man7/vdso.7

   1 .\" Written by Mike Frysinger <vapier@gentoo.org>
   2 .\"
   3 .\" %%%LICENSE_START(PUBLIC_DOMAIN)
   4 .\" This page is in the public domain.
   5 .\" %%%LICENSE_END
   6 .\"
   7 .\" Useful background:
   8 .\"   http://articles.manugarg.com/systemcallinlinux2_6.html
   9 .\"   https://lwn.net/Articles/446528/
  10 .\"   http://www.linuxjournal.com/content/creating-vdso-colonels-other-chicken
  11 .\"   http://www.trilithium.com/johan/2005/08/linux-gate/
  12 .\"
  13 .TH VDSO 7 2015-12-28 "Linux" "Linux Programmer's Manual"
  14 .SH NAME
  15 vdso \- overview of the virtual ELF dynamic shared object
  16 .SH SYNOPSIS
  17 .B #include <sys/auxv.h>
  18
  19 .B void *vdso = (uintptr_t) getauxval(AT_SYSINFO_EHDR);
  20 .SH DESCRIPTION
  21 The "vDSO" (virtual dynamic shared object) is a small shared library that
  22 the kernel automatically maps into the
  23 address space of all user-space applications.
  24 Applications usually do not need to concern themselves with these details
  25 as the vDSO is most commonly called by the C library.
  26 This way you can code in the normal way using standard functions
  27 and the C library will take care
  28 of using any functionality that is available via the vDSO.
  29
  30 Why does the vDSO exist at all?
  31 There are some system calls the kernel provides that
  32 user-space code ends up using frequently,
  33 to the point that such calls can dominate overall performance.
  34 This is due both to the frequency of the call as well as the
  35 context-switch overhead that results
  36 from exiting user space and entering the kernel.
  37
  38 The rest of this documentation is geared toward the curious and/or
  39 C library writers rather than general developers.
  40 If you're trying to call the vDSO in your own application rather than using
  41 the C library, you're most likely doing it wrong.
  42 .SS Example background
  43 Making system calls can be slow.
  44 In x86 32-bit systems, you can trigger a software interrupt
  45 .RI ( "int $0x80" )
  46 to tell the kernel you wish to make a system call.
  47 However, this instruction is expensive: it goes through
  48 the full interrupt-handling paths
  49 in the processor's microcode as well as in the kernel.
  50 Newer processors have faster (but backward incompatible) instructions to
  51 initiate system calls.
  52 Rather than require the C library to figure out if this functionality is
  53 available at run time,
  54 the C library can use functions provided by the kernel in
  55 the vDSO.
  56
  57 Note that the terminology can be confusing.
  58 On x86 systems, the vDSO function
  59 used to determine the preferred method of making a system call is
  60 named "__kernel_vsyscall", but on x86_64,
  61 the term "vsyscall" also refers to an obsolete way to ask the kernel
  62 what time it is or what CPU the caller is on.
  63
  64 One frequently used system call is
  65 .BR gettimeofday (2).
  66 This system call is called both directly by user-space applications
  67 as well as indirectly by
  68 the C library.
  69 Think timestamps or timing loops or polling\(emall of these
  70 frequently need to know what time it is right now.
  71 This information is also not secret\(emany application in any
  72 privilege mode (root or any unprivileged user) will get the same answer.
  73 Thus the kernel arranges for the information required to answer
  74 this question to be placed in memory the process can access.
  75 Now a call to
  76 .BR gettimeofday (2)
  77 changes from a system call to a normal function
  78 call and a few memory accesses.
  79 .SS Finding the vDSO
  80 The base address of the vDSO (if one exists) is passed by the kernel to
  81 each program in the initial auxiliary vector (see
  82 .BR getauxval (3)),
  83 via the
  84 .B AT_SYSINFO_EHDR
  85 tag.
  86
  87 You must not assume the vDSO is mapped at any particular location in the
  88 user's memory map.
  89 The base address will usually be randomized at run time every time a new
  90 process image is created (at
  91 .BR execve (2)
  92 time).
  93 This is done for security reasons,
  94 to prevent "return-to-libc" attacks.
  95
  96 For some architectures, there is also an
  97 .B AT_SYSINFO
  98 tag.
  99 This is used only for locating the vsyscall entry point and is frequently
 100 omitted or set to 0 (meaning it's not available).
 101 This tag is a throwback to the initial vDSO work (see
 102 .IR History
 103 below) and its use should be avoided.
 104 .SS File format
 105 Since the vDSO is a fully formed ELF image, you can do symbol lookups on it.
 106 This allows new symbols to be added with newer kernel releases,
 107 and allows the C library to detect available functionality at
 108 run time when running under different kernel versions.
 109 Oftentimes the C library will do detection with the first call and then
 110 cache the result for subsequent calls.
 111
 112 All symbols are also versioned (using the GNU version format).
 113 This allows the kernel to update the function signature without breaking
 114 backward compatibility.
 115 This means changing the arguments that the function accepts as well as the
 116 return value.
 117 Thus, when looking up a symbol in the vDSO,
 118 you must always include the version
 119 to match the ABI you expect.
 120
 121 Typically the vDSO follows the naming convention of prefixing
 122 all symbols with "__vdso_" or "__kernel_"
 123 so as to distinguish them from other standard symbols.
 124 For example, the "gettimeofday" function is named "__vdso_gettimeofday".
 125
 126 You use the standard C calling conventions when calling
 127 any of these functions.
 128 No need to worry about weird register or stack behavior.
 129 .SH NOTES
 130 .SS Source
 131 When you compile the kernel,
 132 it will automatically compile and link the vDSO code for you.
 133 You will frequently find it under the architecture-specific directory:
 134
 135     find arch/$ARCH/ -name '*vdso*.so*' -o -name '*gate*.so*'
 136
 137 .SS vDSO names
 138 The name of the vDSO varies across architectures.
 139 It will often show up in things like glibc's
 140 .BR ldd (1)
 141 output.
 142 The exact name should not matter to any code, so do not hardcode it.
 143 .if t \{\
 144 .ft CW
 145 \}
 146 .TS
 147 l l.
 148 user ABI        vDSO name
 149 _
 150 aarch64 linux-vdso.so.1
 151 arm     linux-vdso.so.1
 152 ia64    linux-gate.so.1
 153 ppc/32  linux-vdso32.so.1
 154 ppc/64  linux-vdso64.so.1
 155 s390    linux-vdso32.so.1
 156 s390x   linux-vdso64.so.1
 157 sh      linux-gate.so.1
 158 i386    linux-gate.so.1
 159 x86_64  linux-vdso.so.1
 160 x86/x32 linux-vdso.so.1
 161 .TE
 162 .if t \{\
 163 .in
 164 .ft P
 165 \}
 166 .SS strace(1) and the vDSO
 167 When tracing systems calls with
 168 .BR strace (1),
 169 symbols (system calls) that are exported by the vDSO will
 170 .I not
 171 appear in the trace output.
 172 .SH ARCHITECTURE-SPECIFIC NOTES
 173 The subsections below provide architecture-specific notes
 174 on the vDSO.
 175
 176 Note that the vDSO that is used is based on the ABI of your user-space code
 177 and not the ABI of the kernel.
 178 Thus, for example,
 179 when you run an i386 32-bit ELF binary,
 180 you'll get the same vDSO regardless of whether you run it under
 181 an i386 32-bit kernel or under an x86_64 64-bit kernel.
 182 Therefore, the name of the user-space ABI should be used to determine
 183 which of the sections below is relevant.
 184 .SS ARM functions
 185 .\" See linux/arch/arm/vdso/vdso.lds.S
 186 .\" Commit: 8512287a8165592466cb9cb347ba94892e9c56a5
 187 The table below lists the symbols exported by the vDSO.
 188 .if t \{\
 189 .ft CW
 190 \}
 191 .TS
 192 l l.
 193 symbol  version
 194 _
 195 __vdso_gettimeofday     LINUX_2.6 (exported since Linux 4.1)
 196 __vdso_clock_gettime    LINUX_2.6 (exported since Linux 4.1)
 197 .TE
 198 .if t \{\
 199 .in
 200 .ft P
 201 \}
 202
 203 .\" See linux/arch/arm/kernel/entry-armv.S
 204 .\" See linux/Documentation/arm/kernel_user_helpers.txt
 205 Additionally, the ARM port has a code page full of utility functions.
 206 Since it's just a raw page of code, there is no ELF information for doing
 207 symbol lookups or versioning.
 208 It does provide support for different versions though.
 209
 210 For information on this code page,
 211 it's best to refer to the kernel documentation
 212 as it's extremely detailed and covers everything you need to know:
 213 .IR Documentation/arm/kernel_user_helpers.txt .
 214 .SS aarch64 functions
 215 .\" See linux/arch/arm64/kernel/vdso/vdso.lds.S
 216 The table below lists the symbols exported by the vDSO.
 217 .if t \{\
 218 .ft CW
 219 \}
 220 .TS
 221 l l.
 222 symbol  version
 223 _
 224 __kernel_rt_sigreturn   LINUX_2.6.39
 225 __kernel_gettimeofday   LINUX_2.6.39
 226 __kernel_clock_gettime  LINUX_2.6.39
 227 __kernel_clock_getres   LINUX_2.6.39
 228 .TE
 229 .if t \{\
 230 .in
 231 .ft P
 232 \}
 233 .SS bfin (Blackfin) functions
 234 .\" See linux/arch/blackfin/kernel/fixed_code.S
 235 .\" See http://docs.blackfin.uclinux.org/doku.php?id=linux-kernel:fixed-code
 236 As this CPU lacks a memory management unit (MMU),
 237 it doesn't set up a vDSO in the normal sense.
 238 Instead, it maps at boot time a few raw functions into
 239 a fixed location in memory.
 240 User-space applications then call directly into that region.
 241 There is no provision for backward compatibility
 242 beyond sniffing raw opcodes,
 243 but as this is an embedded CPU, it can get away with things\(emsome of the
 244 object formats it runs aren't even ELF based (they're bFLT/FLAT).
 245
 246 For information on this code page,
 247 it's best to refer to the public documentation:
 248 .br
 249 http://docs.blackfin.uclinux.org/doku.php?id=linux-kernel:fixed-code
 250 .SS ia64 (Itanium) functions
 251 .\" See linux/arch/ia64/kernel/gate.lds.S
 252 .\" Also linux/arch/ia64/kernel/fsys.S and linux/Documentation/ia64/fsys.txt
 253 The table below lists the symbols exported by the vDSO.
 254 .if t \{\
 255 .ft CW
 256 \}
 257 .TS
 258 l l.
 259 symbol  version
 260 _
 261 __kernel_sigtramp       LINUX_2.5
 262 __kernel_syscall_via_break      LINUX_2.5
 263 __kernel_syscall_via_epc        LINUX_2.5
 264 .TE
 265 .if t \{\
 266 .in
 267 .ft P
 268 \}
 269
 270 The Itanium port is somewhat tricky.
 271 In addition to the vDSO above, it also has "light-weight system calls"
 272 (also known as "fast syscalls" or "fsys").
 273 You can invoke these via the
 274 .I __kernel_syscall_via_epc
 275 vDSO helper.
 276 The system calls listed here have the same semantics as if you called them
 277 directly via
 278 .BR syscall (2),
 279 so refer to the relevant
 280 documentation for each.
 281 The table below lists the functions available via this mechanism.
 282 .if t \{\
 283 .ft CW
 284 \}
 285 .TS
 286 l.
 287 function
 288 _
 289 clock_gettime
 290 getcpu
 291 getpid
 292 getppid
 293 gettimeofday
 294 set_tid_address
 295 .TE
 296 .if t \{\
 297 .in
 298 .ft P
 299 \}
 300 .SS parisc (hppa) functions
 301 .\" See linux/arch/parisc/kernel/syscall.S
 302 .\" See linux/Documentation/parisc/registers
 303 The parisc port has a code page full of utility functions
 304 called a gateway page.
 305 Rather than use the normal ELF auxiliary vector approach,
 306 it passes the address of
 307 the page to the process via the SR2 register.
 308 The permissions on the page are such that merely executing those addresses
 309 automatically executes with kernel privileges and not in user space.
 310 This is done to match the way HP-UX works.
 311
 312 Since it's just a raw page of code, there is no ELF information for doing
 313 symbol lookups or versioning.
 314 Simply call into the appropriate offset via the branch instruction,
 315 for example:
 316
 317     ble <offset>(%sr2, %r0)
 318 .if t \{\
 319 .ft CW
 320 \}
 321 .TS
 322 l l.
 323 offset  function
 324 _
 325 00b0    lws_entry
 326 00e0    set_thread_pointer
 327 0100    linux_gateway_entry (syscall)
 328 0268    syscall_nosys
 329 0274    tracesys
 330 0324    tracesys_next
 331 0368    tracesys_exit
 332 03a0    tracesys_sigexit
 333 03b8    lws_start
 334 03dc    lws_exit_nosys
 335 03e0    lws_exit
 336 03e4    lws_compare_and_swap64
 337 03e8    lws_compare_and_swap
 338 0404    cas_wouldblock
 339 0410    cas_action
 340 .TE
 341 .if t \{\
 342 .in
 343 .ft P
 344 \}
 345 .SS ppc/32 functions
 346 .\" See linux/arch/powerpc/kernel/vdso32/vdso32.lds.S
 347 The table below lists the symbols exported by the vDSO.
 348 The functions marked with a
 349 .I *
 350 are available only when the kernel is
 351 a PowerPC64 (64-bit) kernel.
 352 .if t \{\
 353 .ft CW
 354 \}
 355 .TS
 356 l l.
 357 symbol  version
 358 _
 359 __kernel_clock_getres   LINUX_2.6.15
 360 __kernel_clock_gettime  LINUX_2.6.15
 361 __kernel_datapage_offset        LINUX_2.6.15
 362 __kernel_get_syscall_map        LINUX_2.6.15
 363 __kernel_get_tbfreq     LINUX_2.6.15
 364 __kernel_getcpu \fI*\fR LINUX_2.6.15
 365 __kernel_gettimeofday   LINUX_2.6.15
 366 __kernel_sigtramp_rt32  LINUX_2.6.15
 367 __kernel_sigtramp32     LINUX_2.6.15
 368 __kernel_sync_dicache   LINUX_2.6.15
 369 __kernel_sync_dicache_p5        LINUX_2.6.15
 370 .TE
 371 .if t \{\
 372 .in
 373 .ft P
 374 \}
 375 .SS ppc/64 functions
 376 .\" See linux/arch/powerpc/kernel/vdso64/vdso64.lds.S
 377 The table below lists the symbols exported by the vDSO.
 378 .if t \{\
 379 .ft CW
 380 \}
 381 .TS
 382 l l.
 383 symbol  version
 384 _
 385 __kernel_clock_getres   LINUX_2.6.15
 386 __kernel_clock_gettime  LINUX_2.6.15
 387 __kernel_datapage_offset        LINUX_2.6.15
 388 __kernel_get_syscall_map        LINUX_2.6.15
 389 __kernel_get_tbfreq     LINUX_2.6.15
 390 __kernel_getcpu LINUX_2.6.15
 391 __kernel_gettimeofday   LINUX_2.6.15
 392 __kernel_sigtramp_rt64  LINUX_2.6.15
 393 __kernel_sync_dicache   LINUX_2.6.15
 394 __kernel_sync_dicache_p5        LINUX_2.6.15
 395 .TE
 396 .if t \{\
 397 .in
 398 .ft P
 399 \}
 400 .SS s390 functions
 401 .\" See linux/arch/s390/kernel/vdso32/vdso32.lds.S
 402 The table below lists the symbols exported by the vDSO.
 403 .if t \{\
 404 .ft CW
 405 \}
 406 .TS
 407 l l.
 408 symbol  version
 409 _
 410 __kernel_clock_getres   LINUX_2.6.29
 411 __kernel_clock_gettime  LINUX_2.6.29
 412 __kernel_gettimeofday   LINUX_2.6.29
 413 .TE
 414 .if t \{\
 415 .in
 416 .ft P
 417 \}
 418 .SS s390x functions
 419 .\" See linux/arch/s390/kernel/vdso64/vdso64.lds.S
 420 The table below lists the symbols exported by the vDSO.
 421 .if t \{\
 422 .ft CW
 423 \}
 424 .TS
 425 l l.
 426 symbol  version
 427 _
 428 __kernel_clock_getres   LINUX_2.6.29
 429 __kernel_clock_gettime  LINUX_2.6.29
 430 __kernel_gettimeofday   LINUX_2.6.29
 431 .TE
 432 .if t \{\
 433 .in
 434 .ft P
 435 \}
 436 .SS sh (SuperH) functions
 437 .\" See linux/arch/sh/kernel/vsyscall/vsyscall.lds.S
 438 The table below lists the symbols exported by the vDSO.
 439 .if t \{\
 440 .ft CW
 441 \}
 442 .TS
 443 l l.
 444 symbol  version
 445 _
 446 __kernel_rt_sigreturn   LINUX_2.6
 447 __kernel_sigreturn      LINUX_2.6
 448 __kernel_vsyscall       LINUX_2.6
 449 .TE
 450 .if t \{\
 451 .in
 452 .ft P
 453 \}
 454 .SS i386 functions
 455 .\" See linux/arch/x86/vdso/vdso32/vdso32.lds.S
 456 The table below lists the symbols exported by the vDSO.
 457 .if t \{\
 458 .ft CW
 459 \}
 460 .TS
 461 l l.
 462 symbol  version
 463 _
 464 __kernel_sigreturn      LINUX_2.5
 465 __kernel_rt_sigreturn   LINUX_2.5
 466 __kernel_vsyscall       LINUX_2.5
 467 .\" Added in 7a59ed415f5b57469e22e41fc4188d5399e0b194 and updated
 468 .\" in 37c975545ec63320789962bf307f000f08fabd48.
 469 __vdso_clock_gettime    LINUX_2.6 (exported since Linux 3.15)
 470 __vdso_gettimeofday     LINUX_2.6 (exported since Linux 3.15)
 471 __vdso_time     LINUX_2.6 (exported since Linux 3.15)
 472 .TE
 473 .if t \{\
 474 .in
 475 .ft P
 476 \}
 477 .SS x86_64 functions
 478 .\" See linux/arch/x86/vdso/vdso.lds.S
 479 The table below lists the symbols exported by the vDSO.
 480 All of these symbols are also available without the "__vdso_" prefix, but
 481 you should ignore those and stick to the names below.
 482 .if t \{\
 483 .ft CW
 484 \}
 485 .TS
 486 l l.
 487 symbol  version
 488 _
 489 __vdso_clock_gettime    LINUX_2.6
 490 __vdso_getcpu   LINUX_2.6
 491 __vdso_gettimeofday     LINUX_2.6
 492 __vdso_time     LINUX_2.6
 493 .TE
 494 .if t \{\
 495 .in
 496 .ft P
 497 \}
 498 .SS x86/x32 functions
 499 .\" See linux/arch/x86/vdso/vdso32.lds.S
 500 The table below lists the symbols exported by the vDSO.
 501 .if t \{\
 502 .ft CW
 503 \}
 504 .TS
 505 l l.
 506 symbol  version
 507 _
 508 __vdso_clock_gettime    LINUX_2.6
 509 __vdso_getcpu   LINUX_2.6
 510 __vdso_gettimeofday     LINUX_2.6
 511 __vdso_time     LINUX_2.6
 512 .TE
 513 .if t \{\
 514 .in
 515 .ft P
 516 \}
 517 .SS History
 518 The vDSO was originally just a single function\(emthe vsyscall.
 519 In older kernels, you might see that name
 520 in a process's memory map rather than "vdso".
 521 Over time, people realized that this mechanism
 522 was a great way to pass more functionality
 523 to user space, so it was reconceived as a vDSO in the current format.
 524 .SH SEE ALSO
 525 .BR syscalls (2),
 526 .BR getauxval (3),
 527 .BR proc (5)
 528
 529 The documents, examples, and source code in the Linux source code tree:
 530 .in +4n
 531 .nf
 532
 533 Documentation/ABI/stable/vdso
 534 Documentation/ia64/fsys.txt
 535 Documentation/vDSO/* (includes examples of using the vDSO)
 536
 537 find arch/ -iname '*vdso*' -o -iname '*gate*'
 538 .fi
 539 .in