| 1 |
package cap |
| 2 |
|
| 3 |
import ( |
| 4 |
"bytes" |
| 5 |
"encoding/binary" |
| 6 |
"errors" |
| 7 |
"fmt" |
| 8 |
"io" |
| 9 |
"os" |
| 10 |
"syscall" |
| 11 |
"unsafe" |
| 12 |
) |
| 13 |
|
| 14 |
// uapi/linux/xattr.h defined. |
| 15 |
var ( |
| 16 |
xattrNameCaps, _ = syscall.BytePtrFromString("security.capability") |
| 17 |
) |
| 18 |
|
| 19 |
// uapi/linux/capability.h defined. |
| 20 |
const ( |
| 21 |
vfsCapRevisionMask = uint32(0xff000000) |
| 22 |
vfsCapFlagsMask = ^vfsCapRevisionMask |
| 23 |
vfsCapFlagsEffective = uint32(1) |
| 24 |
|
| 25 |
vfsCapRevision1 = uint32(0x01000000) |
| 26 |
vfsCapRevision2 = uint32(0x02000000) |
| 27 |
vfsCapRevision3 = uint32(0x03000000) |
| 28 |
) |
| 29 |
|
| 30 |
// uapi/asm-generic/fcntl.h defined |
| 31 |
const ( |
| 32 |
O_PATH = 0x200000 |
| 33 |
) |
| 34 |
|
| 35 |
// Data types stored in little-endian order. |
| 36 |
|
| 37 |
type vfsCaps1 struct { |
| 38 |
MagicEtc uint32 |
| 39 |
Data [1]struct { |
| 40 |
Permitted, Inheritable uint32 |
| 41 |
} |
| 42 |
} |
| 43 |
|
| 44 |
type vfsCaps2 struct { |
| 45 |
MagicEtc uint32 |
| 46 |
Data [2]struct { |
| 47 |
Permitted, Inheritable uint32 |
| 48 |
} |
| 49 |
} |
| 50 |
|
| 51 |
type vfsCaps3 struct { |
| 52 |
MagicEtc uint32 |
| 53 |
Data [2]struct { |
| 54 |
Permitted, Inheritable uint32 |
| 55 |
} |
| 56 |
RootID uint32 |
| 57 |
} |
| 58 |
|
| 59 |
// ErrBadSize indicates the loaded file capability has |
| 60 |
// an invalid number of bytes in it. |
| 61 |
var ErrBadSize = errors.New("filecap bad size") |
| 62 |
|
| 63 |
// ErrBadMagic indicates that the kernel preferred magic number for |
| 64 |
// capability Set values is not supported by this package. This |
| 65 |
// generally implies you are using an exceptionally old |
| 66 |
// "../libcap/cap" package. An upgrade is needed, or failing that see |
| 67 |
// [the Fully Capable site] for the way to report or review a bug. |
| 68 |
// |
| 69 |
// [the Fully Capable site]: https://sites.google.com/site/fullycapable/ |
| 70 |
var ErrBadMagic = errors.New("unsupported magic") |
| 71 |
|
| 72 |
// ErrBadPath indicates a failed attempt to set a file capability on |
| 73 |
// an irregular (non-executable) file. |
| 74 |
var ErrBadPath = errors.New("file is not a regular executable") |
| 75 |
|
| 76 |
// ErrOutOfRange indicates an erroneous value for MinExtFlagSize. |
| 77 |
var ErrOutOfRange = errors.New("flag length invalid for export") |
| 78 |
|
| 79 |
// digestFileCap unpacks a file capability and returns it in a *Set |
| 80 |
// form. |
| 81 |
func digestFileCap(d []byte, sz int, err error) (*Set, error) { |
| 82 |
if err != nil { |
| 83 |
return nil, err |
| 84 |
} |
| 85 |
var raw1 vfsCaps1 |
| 86 |
var raw2 vfsCaps2 |
| 87 |
var raw3 vfsCaps3 |
| 88 |
if sz < binary.Size(raw1) || sz > binary.Size(raw3) { |
| 89 |
return nil, ErrBadSize |
| 90 |
} |
| 91 |
b := bytes.NewReader(d[:sz]) |
| 92 |
var magicEtc uint32 |
| 93 |
if err = binary.Read(b, binary.LittleEndian, &magicEtc); err != nil { |
| 94 |
return nil, err |
| 95 |
} |
| 96 |
|
| 97 |
c := NewSet() |
| 98 |
b.Seek(0, io.SeekStart) |
| 99 |
switch magicEtc & vfsCapRevisionMask { |
| 100 |
case vfsCapRevision1: |
| 101 |
if err = binary.Read(b, binary.LittleEndian, &raw1); err != nil { |
| 102 |
return nil, err |
| 103 |
} |
| 104 |
data := raw1.Data[0] |
| 105 |
c.flat[0][Permitted] = data.Permitted |
| 106 |
c.flat[0][Inheritable] = data.Inheritable |
| 107 |
if raw1.MagicEtc&vfsCapFlagsMask == vfsCapFlagsEffective { |
| 108 |
c.flat[0][Effective] = data.Inheritable | data.Permitted |
| 109 |
} |
| 110 |
case vfsCapRevision2: |
| 111 |
if err = binary.Read(b, binary.LittleEndian, &raw2); err != nil { |
| 112 |
return nil, err |
| 113 |
} |
| 114 |
for i, data := range raw2.Data { |
| 115 |
c.flat[i][Permitted] = data.Permitted |
| 116 |
c.flat[i][Inheritable] = data.Inheritable |
| 117 |
if raw2.MagicEtc&vfsCapFlagsMask == vfsCapFlagsEffective { |
| 118 |
c.flat[i][Effective] = data.Inheritable | data.Permitted |
| 119 |
} |
| 120 |
} |
| 121 |
case vfsCapRevision3: |
| 122 |
if err = binary.Read(b, binary.LittleEndian, &raw3); err != nil { |
| 123 |
return nil, err |
| 124 |
} |
| 125 |
for i, data := range raw3.Data { |
| 126 |
c.flat[i][Permitted] = data.Permitted |
| 127 |
c.flat[i][Inheritable] = data.Inheritable |
| 128 |
if raw3.MagicEtc&vfsCapFlagsMask == vfsCapFlagsEffective { |
| 129 |
c.flat[i][Effective] = data.Inheritable | data.Permitted |
| 130 |
} |
| 131 |
} |
| 132 |
c.nsRoot = int(raw3.RootID) |
| 133 |
default: |
| 134 |
return nil, ErrBadMagic |
| 135 |
} |
| 136 |
return c, nil |
| 137 |
} |
| 138 |
|
| 139 |
//go:uintptrescapes |
| 140 |
|
| 141 |
// GetFd returns the file capabilities of an open (*os.File).Fd(). |
| 142 |
func GetFd(file *os.File) (*Set, error) { |
| 143 |
var raw3 vfsCaps3 |
| 144 |
d := make([]byte, binary.Size(raw3)) |
| 145 |
sz, _, oErr := multisc.r6(syscall.SYS_FGETXATTR, fd(file), uintptr(unsafe.Pointer(xattrNameCaps)), uintptr(unsafe.Pointer(&d[0])), uintptr(len(d)), 0, 0) |
| 146 |
var err error |
| 147 |
if oErr != 0 { |
| 148 |
err = oErr |
| 149 |
} |
| 150 |
return digestFileCap(d, int(sz), err) |
| 151 |
} |
| 152 |
|
| 153 |
//go:uintptrescapes |
| 154 |
|
| 155 |
// GetFile returns the file capabilities of a named file. |
| 156 |
func GetFile(path string) (*Set, error) { |
| 157 |
p, err := syscall.BytePtrFromString(path) |
| 158 |
if err != nil { |
| 159 |
return nil, err |
| 160 |
} |
| 161 |
var raw3 vfsCaps3 |
| 162 |
d := make([]byte, binary.Size(raw3)) |
| 163 |
sz, _, oErr := multisc.r6(syscall.SYS_GETXATTR, uintptr(unsafe.Pointer(p)), uintptr(unsafe.Pointer(xattrNameCaps)), uintptr(unsafe.Pointer(&d[0])), uintptr(len(d)), 0, 0) |
| 164 |
if oErr != 0 { |
| 165 |
err = oErr |
| 166 |
} |
| 167 |
return digestFileCap(d, int(sz), err) |
| 168 |
} |
| 169 |
|
| 170 |
// GetNSOwner returns the namespace owner UID of the capability Set. |
| 171 |
func (c *Set) GetNSOwner() (int, error) { |
| 172 |
if magic < kv3 { |
| 173 |
return 0, ErrBadMagic |
| 174 |
} |
| 175 |
c.mu.RLock() |
| 176 |
defer c.mu.RUnlock() |
| 177 |
return c.nsRoot, nil |
| 178 |
} |
| 179 |
|
| 180 |
// SetNSOwner adds an explicit namespace owner UID to the capability |
| 181 |
// Set. This is only honored when generating file capabilities, and is |
| 182 |
// generally for use by a setup process when installing binaries that |
| 183 |
// use file capabilities to become capable inside a namespace to be |
| 184 |
// administered by that UID. If capability aware code within that |
| 185 |
// namespace writes file capabilities without explicitly setting such |
| 186 |
// a UID, the kernel will fix-up the capabilities to be specific to |
| 187 |
// that owner. In this way, the kernel prevents filesystem |
| 188 |
// capabilities from leaking out of that restricted namespace. |
| 189 |
func (c *Set) SetNSOwner(uid int) { |
| 190 |
c.mu.Lock() |
| 191 |
defer c.mu.Unlock() |
| 192 |
c.nsRoot = uid |
| 193 |
} |
| 194 |
|
| 195 |
// packFileCap transforms a system capability into a VFS form. Because |
| 196 |
// of the way Linux stores capabilities in the file extended |
| 197 |
// attributes, the process is a little lossy with respect to effective |
| 198 |
// bits. |
| 199 |
func (c *Set) packFileCap() ([]byte, error) { |
| 200 |
c.mu.RLock() |
| 201 |
defer c.mu.RUnlock() |
| 202 |
|
| 203 |
var magic uint32 |
| 204 |
switch words { |
| 205 |
case 1: |
| 206 |
if c.nsRoot != 0 { |
| 207 |
return nil, ErrBadSet // nsRoot not supported for single DWORD caps. |
| 208 |
} |
| 209 |
magic = vfsCapRevision1 |
| 210 |
case 2: |
| 211 |
if c.nsRoot == 0 { |
| 212 |
magic = vfsCapRevision2 |
| 213 |
break |
| 214 |
} |
| 215 |
magic = vfsCapRevision3 |
| 216 |
} |
| 217 |
if magic == 0 { |
| 218 |
return nil, ErrBadSize |
| 219 |
} |
| 220 |
eff := uint32(0) |
| 221 |
for _, f := range c.flat { |
| 222 |
eff |= (f[Permitted] | f[Inheritable]) & f[Effective] |
| 223 |
} |
| 224 |
if eff != 0 { |
| 225 |
magic |= vfsCapFlagsEffective |
| 226 |
} |
| 227 |
b := new(bytes.Buffer) |
| 228 |
binary.Write(b, binary.LittleEndian, magic) |
| 229 |
for _, f := range c.flat { |
| 230 |
binary.Write(b, binary.LittleEndian, f[Permitted]) |
| 231 |
binary.Write(b, binary.LittleEndian, f[Inheritable]) |
| 232 |
} |
| 233 |
if c.nsRoot != 0 { |
| 234 |
binary.Write(b, binary.LittleEndian, c.nsRoot) |
| 235 |
} |
| 236 |
return b.Bytes(), nil |
| 237 |
} |
| 238 |
|
| 239 |
// specialFile confirms that the open file descriptor is not a regular |
| 240 |
// file. |
| 241 |
func specialFile(filed uintptr) bool { |
| 242 |
var stat syscall.Stat_t |
| 243 |
if err := syscall.Fstat(int(filed), &stat); err != nil { |
| 244 |
return false |
| 245 |
} |
| 246 |
return (stat.Mode & syscall.S_IFMT) != syscall.S_IFREG |
| 247 |
} |
| 248 |
|
| 249 |
//go:uintptrescapes |
| 250 |
|
| 251 |
// setFiled sets the file capability of an open file descriptor. |
| 252 |
func (c *Set) setFiled(filed uintptr) error { |
| 253 |
if c == nil { |
| 254 |
if _, _, err := multisc.r6(syscall.SYS_FREMOVEXATTR, filed, uintptr(unsafe.Pointer(xattrNameCaps)), 0, 0, 0, 0); err != 0 { |
| 255 |
return err |
| 256 |
} |
| 257 |
return nil |
| 258 |
} |
| 259 |
c.mu.RLock() |
| 260 |
defer c.mu.RUnlock() |
| 261 |
d, err := c.packFileCap() |
| 262 |
if err != nil { |
| 263 |
return err |
| 264 |
} |
| 265 |
if specialFile(filed) { |
| 266 |
return ErrBadPath |
| 267 |
} |
| 268 |
if _, _, err := multisc.r6(syscall.SYS_FSETXATTR, filed, uintptr(unsafe.Pointer(xattrNameCaps)), uintptr(unsafe.Pointer(&d[0])), uintptr(len(d)), 0, 0); err != 0 { |
| 269 |
return err |
| 270 |
} |
| 271 |
return nil |
| 272 |
} |
| 273 |
|
| 274 |
// SetFd attempts to set the file capabilities of an open |
| 275 |
// (*os.File).Fd(). This function can also be used to delete a file's |
| 276 |
// capabilities, by calling with c = nil. |
| 277 |
// |
| 278 |
// Note, Linux does not store the full Effective Flag in the metadata |
| 279 |
// for the file. Only a single Effective bit is stored in this |
| 280 |
// metadata. This single bit is non-zero if the Effective Flag has any |
| 281 |
// overlapping bits with the Permitted or Inheritable Flags of c. This |
| 282 |
// may appear suboptimal, but the reasoning behind it is sound. |
| 283 |
// Namely, the purpose of the Effective bit it to support capabability |
| 284 |
// unaware binaries that will only work if they magically launch with |
| 285 |
// the needed Values already raised (this bit is sometimes referred to |
| 286 |
// simply as the 'legacy' bit). |
| 287 |
// |
| 288 |
// Historical note: without *full* support for runtime capability |
| 289 |
// manipulation, as it is provided in this "../libcap/cap" package, |
| 290 |
// this was previously the only way for Go programs to make use of |
| 291 |
// file capabilities. |
| 292 |
// |
| 293 |
// The preferred way that a binary will actually manipulate its |
| 294 |
// file-acquired capabilities is to carefully and deliberately use |
| 295 |
// this package (or libcap, assisted by libpsx, for threaded C/C++ |
| 296 |
// family code). |
| 297 |
func (c *Set) SetFd(file *os.File) error { |
| 298 |
return c.setFiled(fd(file)) |
| 299 |
} |
| 300 |
|
| 301 |
//go:uintptrescapes |
| 302 |
|
| 303 |
// SetFile attempts to set the file capabilities of the specified |
| 304 |
// filename. This function can also be used to delete a file's |
| 305 |
// capabilities, by calling with c = nil. |
| 306 |
// |
| 307 |
// Note, see the comment for SetFd() for some non-obvious behavior of |
| 308 |
// Linux for the Effective Flag on the modified file. |
| 309 |
func (c *Set) SetFile(path string) error { |
| 310 |
filed, err := syscall.Open(path, syscall.O_RDONLY|syscall.O_NOFOLLOW, 0) |
| 311 |
if err == nil { |
| 312 |
defer syscall.Close(filed) |
| 313 |
return c.setFiled(uintptr(filed)) |
| 314 |
} |
| 315 |
|
| 316 |
// Corner case: attempting to set a capability on a file not |
| 317 |
// readable by the caller. |
| 318 |
|
| 319 |
filed, err = syscall.Open(path, O_PATH|syscall.O_NOFOLLOW, 0) |
| 320 |
if err != nil { |
| 321 |
return err |
| 322 |
} |
| 323 |
defer syscall.Close(filed) |
| 324 |
if specialFile(uintptr(filed)) { |
| 325 |
return ErrBadPath |
| 326 |
} |
| 327 |
stable := fmt.Sprint(procRoot, "/self/fd/", filed) |
| 328 |
p, err := syscall.BytePtrFromString(stable) |
| 329 |
if err != nil { |
| 330 |
return err |
| 331 |
} |
| 332 |
if c == nil { |
| 333 |
if _, _, err := multisc.r6(syscall.SYS_REMOVEXATTR, uintptr(unsafe.Pointer(p)), uintptr(unsafe.Pointer(xattrNameCaps)), 0, 0, 0, 0); err != 0 { |
| 334 |
return err |
| 335 |
} |
| 336 |
return nil |
| 337 |
} |
| 338 |
c.mu.RLock() |
| 339 |
defer c.mu.RUnlock() |
| 340 |
d, err := c.packFileCap() |
| 341 |
if err != nil { |
| 342 |
return err |
| 343 |
} |
| 344 |
if _, _, err := multisc.r6(syscall.SYS_SETXATTR, uintptr(unsafe.Pointer(p)), uintptr(unsafe.Pointer(xattrNameCaps)), uintptr(unsafe.Pointer(&d[0])), uintptr(len(d)), 0, 0); err != 0 { |
| 345 |
return err |
| 346 |
} |
| 347 |
return nil |
| 348 |
} |
| 349 |
|
| 350 |
// ExtMagic is the 32-bit (little endian) magic for an external |
| 351 |
// capability set. It can be used to transmit capabilities in binary |
| 352 |
// format in a Linux portable way. The format is: |
| 353 |
// <ExtMagic><byte:length><length-bytes*3-of-cap-data>. |
| 354 |
const ExtMagic = uint32(0x5101c290) |
| 355 |
|
| 356 |
// Import imports a Set from a byte array where it has been stored in |
| 357 |
// a portable (lossless) way. That is values exported by |
| 358 |
// libcap.cap_copy_ext() and Export(). |
| 359 |
func Import(d []byte) (*Set, error) { |
| 360 |
b := bytes.NewBuffer(d) |
| 361 |
var m uint32 |
| 362 |
if err := binary.Read(b, binary.LittleEndian, &m); err != nil { |
| 363 |
return nil, ErrBadSize |
| 364 |
} else if m != ExtMagic { |
| 365 |
return nil, ErrBadMagic |
| 366 |
} |
| 367 |
var n byte |
| 368 |
if err := binary.Read(b, binary.LittleEndian, &n); err != nil { |
| 369 |
return nil, ErrBadSize |
| 370 |
} |
| 371 |
c := NewSet() |
| 372 |
if int(n) > 4*words { |
| 373 |
return nil, ErrBadSize |
| 374 |
} |
| 375 |
f := make([]byte, 3) |
| 376 |
for i := 0; i < words; i++ { |
| 377 |
for j := uint(0); n > 0 && j < 4; j++ { |
| 378 |
n-- |
| 379 |
if x, err := b.Read(f); err != nil || x != 3 { |
| 380 |
return nil, ErrBadSize |
| 381 |
} |
| 382 |
sh := 8 * j |
| 383 |
c.flat[i][Effective] |= uint32(f[0]) << sh |
| 384 |
c.flat[i][Permitted] |= uint32(f[1]) << sh |
| 385 |
c.flat[i][Inheritable] |= uint32(f[2]) << sh |
| 386 |
} |
| 387 |
} |
| 388 |
return c, nil |
| 389 |
} |
| 390 |
|
| 391 |
// MinExtFlagSize defaults to 8 in order to be equivalent to libcap |
| 392 |
// defaults. Setting it to zero can generate smaller external |
| 393 |
// representations. Such smaller representations can be imported by |
| 394 |
// libcap and the Go package just fine, we just default to the default |
| 395 |
// libcap representation for legacy reasons. |
| 396 |
var MinExtFlagSize = uint(8) |
| 397 |
|
| 398 |
// Export exports a Set into a lossless byte array format where it is |
| 399 |
// stored in a portable way. Note, any namespace owner in the Set |
| 400 |
// content is not exported by this function. |
| 401 |
// |
| 402 |
// Note, Export() generates exported byte streams that are importable |
| 403 |
// by libcap.cap_copy_int() as well as Import(). |
| 404 |
func (c *Set) Export() ([]byte, error) { |
| 405 |
if err := c.good(); err != nil { |
| 406 |
return nil, err |
| 407 |
} |
| 408 |
if MinExtFlagSize > 255 { |
| 409 |
return nil, ErrOutOfRange |
| 410 |
} |
| 411 |
b := new(bytes.Buffer) |
| 412 |
binary.Write(b, binary.LittleEndian, ExtMagic) |
| 413 |
c.mu.RLock() |
| 414 |
defer c.mu.RUnlock() |
| 415 |
var n = uint(0) |
| 416 |
for i, f := range c.flat { |
| 417 |
if nn := 4 * uint(i); nn+4 > n { |
| 418 |
if u := f[Effective] | f[Permitted] | f[Inheritable]; u != 0 { |
| 419 |
n = nn |
| 420 |
for ; u != 0; u >>= 8 { |
| 421 |
n++ |
| 422 |
} |
| 423 |
} |
| 424 |
} |
| 425 |
} |
| 426 |
if n < MinExtFlagSize { |
| 427 |
n = MinExtFlagSize |
| 428 |
} |
| 429 |
b.Write([]byte{byte(n)}) |
| 430 |
for _, f := range c.flat { |
| 431 |
if n == 0 { |
| 432 |
break |
| 433 |
} |
| 434 |
eff, per, inh := f[Effective], f[Permitted], f[Inheritable] |
| 435 |
for i := 0; n > 0 && i < 4; i++ { |
| 436 |
n-- |
| 437 |
b.Write([]byte{ |
| 438 |
byte(eff & 0xff), |
| 439 |
byte(per & 0xff), |
| 440 |
byte(inh & 0xff), |
| 441 |
}) |
| 442 |
eff >>= 8 |
| 443 |
per >>= 8 |
| 444 |
inh >>= 8 |
| 445 |
} |
| 446 |
} |
| 447 |
for n > 0 { |
| 448 |
n-- |
| 449 |
b.Write([]byte{0, 0, 0}) |
| 450 |
} |
| 451 |
return b.Bytes(), nil |
| 452 |
} |