| 1 |
.\" Copyright 2020, Michael Kerrisk <[email protected]> |
| 2 |
.\" Copyright, the authors of the Linux man-pages project |
| 3 |
.\" |
| 4 |
.\" SPDX-License-Identifier: Linux-man-pages-copyleft |
| 5 |
.\" |
| 6 |
.TH seccomp_unotify 2 2026-02-22 "Linux man-pages 6.18" |
| 7 |
.SH NAME |
| 8 |
seccomp_unotify \- Seccomp user-space notification mechanism |
| 9 |
.SH LIBRARY |
| 10 |
Standard C library |
| 11 |
.RI ( libc ,\~ \-lc ) |
| 12 |
.SH SYNOPSIS |
| 13 |
.nf |
| 14 |
.B #include <linux/seccomp.h> |
| 15 |
.B #include <linux/filter.h> |
| 16 |
.B #include <linux/audit.h> |
| 17 |
.P |
| 18 |
.BI "int seccomp(unsigned int " operation ", unsigned int " flags \ |
| 19 |
", void *" args ); |
| 20 |
.P |
| 21 |
.B #include <sys/ioctl.h> |
| 22 |
.P |
| 23 |
.BI "int ioctl(int " fd ", SECCOMP_IOCTL_NOTIF_RECV," |
| 24 |
.BI " struct seccomp_notif *" req ); |
| 25 |
.BI "int ioctl(int " fd ", SECCOMP_IOCTL_NOTIF_SEND," |
| 26 |
.BI " struct seccomp_notif_resp *" resp ); |
| 27 |
.BI "int ioctl(int " fd ", SECCOMP_IOCTL_NOTIF_ID_VALID, __u64 *" id ); |
| 28 |
.BI "int ioctl(int " fd ", SECCOMP_IOCTL_NOTIF_ADDFD," |
| 29 |
.BI " struct seccomp_notif_addfd *" addfd ); |
| 30 |
.fi |
| 31 |
.SH DESCRIPTION |
| 32 |
This page describes the user-space notification mechanism provided by the |
| 33 |
Secure Computing (seccomp) facility. |
| 34 |
As well as the use of the |
| 35 |
.B SECCOMP_FILTER_FLAG_NEW_LISTENER |
| 36 |
flag, the |
| 37 |
.B SECCOMP_RET_USER_NOTIF |
| 38 |
action value, and the |
| 39 |
.B SECCOMP_GET_NOTIF_SIZES |
| 40 |
operation described in |
| 41 |
.BR seccomp (2), |
| 42 |
this mechanism involves the use of a number of related |
| 43 |
.BR ioctl (2) |
| 44 |
operations (described below). |
| 45 |
.\" |
| 46 |
.SS Overview |
| 47 |
In conventional usage of a seccomp filter, |
| 48 |
the decision about how to treat a system call is made by the filter itself. |
| 49 |
By contrast, the user-space notification mechanism allows |
| 50 |
the seccomp filter to delegate |
| 51 |
the handling of the system call to another user-space process. |
| 52 |
Note that this mechanism is explicitly |
| 53 |
.B not |
| 54 |
intended as a method implementing security policy; |
| 55 |
see NOTES. |
| 56 |
.P |
| 57 |
In the discussion that follows, |
| 58 |
the thread(s) on which the seccomp filter is installed is (are) |
| 59 |
referred to as the |
| 60 |
.IR target , |
| 61 |
and the process that is notified by the user-space notification |
| 62 |
mechanism is referred to as the |
| 63 |
.IR supervisor . |
| 64 |
.P |
| 65 |
A suitably privileged supervisor can use the user-space notification |
| 66 |
mechanism to perform actions on behalf of the target. |
| 67 |
The advantage of the user-space notification mechanism is that |
| 68 |
the supervisor will |
| 69 |
usually be able to retrieve information about the target and the |
| 70 |
performed system call that the seccomp filter itself cannot. |
| 71 |
(A seccomp filter is limited in the information it can obtain and |
| 72 |
the actions that it can perform because it |
| 73 |
is running on a virtual machine inside the kernel.) |
| 74 |
.P |
| 75 |
An overview of the steps performed by the target and the supervisor |
| 76 |
is as follows: |
| 77 |
.\"------------------------------------- |
| 78 |
.IP (1) 5 |
| 79 |
The target establishes a seccomp filter in the usual manner, |
| 80 |
but with two differences: |
| 81 |
.RS |
| 82 |
.IP \[bu] 3 |
| 83 |
The |
| 84 |
.BR seccomp (2) |
| 85 |
.I flags |
| 86 |
argument includes the flag |
| 87 |
.BR SECCOMP_FILTER_FLAG_NEW_LISTENER . |
| 88 |
Consequently, the return value of the (successful) |
| 89 |
.BR seccomp (2) |
| 90 |
call is a new "listening" |
| 91 |
file descriptor that can be used to receive notifications. |
| 92 |
Only one "listening" seccomp filter can be installed for a thread. |
| 93 |
.\" FIXME |
| 94 |
.\" Is the last sentence above correct? |
| 95 |
.\" |
| 96 |
.\" Kees Cook (25 Oct 2020) notes: |
| 97 |
.\" |
| 98 |
.\" I like this limitation, but I expect that it'll need to change in the |
| 99 |
.\" future. Even with LSMs, we see the need for arbitrary stacking, and the |
| 100 |
.\" idea of there being only 1 supervisor will eventually break down. Right |
| 101 |
.\" now there is only 1 because only container managers are using this |
| 102 |
.\" feature. But if some daemon starts using it to isolate some thread, |
| 103 |
.\" suddenly it might break if a container manager is trying to listen to it |
| 104 |
.\" too, etc. I expect it won't be needed soon, but I do think it'll change. |
| 105 |
.\" |
| 106 |
.IP \[bu] |
| 107 |
In cases where it is appropriate, the seccomp filter returns the action value |
| 108 |
.BR SECCOMP_RET_USER_NOTIF . |
| 109 |
This return value will trigger a notification event. |
| 110 |
.RE |
| 111 |
.\"------------------------------------- |
| 112 |
.IP (2) |
| 113 |
In order that the supervisor can obtain notifications |
| 114 |
using the listening file descriptor, |
| 115 |
(a duplicate of) that file descriptor must be passed from |
| 116 |
the target to the supervisor. |
| 117 |
One way in which this could be done is by passing the file descriptor |
| 118 |
over a UNIX domain socket connection between the target and the supervisor |
| 119 |
(using the |
| 120 |
.B SCM_RIGHTS |
| 121 |
ancillary message type described in |
| 122 |
.BR unix (7)). |
| 123 |
Another way to do this is through the use of |
| 124 |
.BR pidfd_getfd (2). |
| 125 |
.\" Jann Horn: |
| 126 |
.\" Instead of using unix domain sockets to send the fd to the |
| 127 |
.\" parent, I think you could also use clone3() with |
| 128 |
.\" flags==CLONE_FILES|SIGCHLD, dup2() the seccomp fd to an fd |
| 129 |
.\" that was reserved in the parent, call unshare(CLONE_FILES) |
| 130 |
.\" in the child after setting up the seccomp fd, and wake |
| 131 |
.\" up the parent with something like pthread_cond_signal()? |
| 132 |
.\" I'm not sure whether that'd look better or worse in the |
| 133 |
.\" end though, so maybe just ignore this comment. |
| 134 |
.\"------------------------------------- |
| 135 |
.IP (3) |
| 136 |
The supervisor will receive notification events |
| 137 |
on the listening file descriptor. |
| 138 |
These events are returned as structures of type |
| 139 |
.IR seccomp_notif . |
| 140 |
Because this structure and its size may evolve over kernel versions, |
| 141 |
the supervisor must first determine the size of this structure |
| 142 |
using the |
| 143 |
.BR seccomp (2) |
| 144 |
.B SECCOMP_GET_NOTIF_SIZES |
| 145 |
operation, which returns a structure of type |
| 146 |
.IR seccomp_notif_sizes . |
| 147 |
The supervisor allocates a buffer of size |
| 148 |
.I seccomp_notif_sizes.seccomp_notif |
| 149 |
bytes to receive notification events. |
| 150 |
In addition,the supervisor allocates another buffer of size |
| 151 |
.I seccomp_notif_sizes.seccomp_notif_resp |
| 152 |
bytes for the response (a |
| 153 |
.I struct seccomp_notif_resp |
| 154 |
structure) |
| 155 |
that it will provide to the kernel (and thus the target). |
| 156 |
.\"------------------------------------- |
| 157 |
.IP (4) |
| 158 |
The target then performs its workload, |
| 159 |
which includes system calls that will be controlled by the seccomp filter. |
| 160 |
Whenever one of these system calls causes the filter to return the |
| 161 |
.B SECCOMP_RET_USER_NOTIF |
| 162 |
action value, the kernel does |
| 163 |
.I not |
| 164 |
(yet) execute the system call; |
| 165 |
instead, execution of the target is temporarily blocked inside |
| 166 |
the kernel (in a sleep state that is interruptible by signals) |
| 167 |
and a notification event is generated on the listening file descriptor. |
| 168 |
.\"------------------------------------- |
| 169 |
.IP (5) |
| 170 |
The supervisor can now repeatedly monitor the |
| 171 |
listening file descriptor for |
| 172 |
.BR SECCOMP_RET_USER_NOTIF -triggered |
| 173 |
events. |
| 174 |
To do this, the supervisor uses the |
| 175 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 176 |
.BR ioctl (2) |
| 177 |
operation to read information about a notification event; |
| 178 |
this operation blocks until an event is available. |
| 179 |
The operation returns a |
| 180 |
.I seccomp_notif |
| 181 |
structure containing information about the system call |
| 182 |
that is being attempted by the target. |
| 183 |
(As described in NOTES, |
| 184 |
the file descriptor can also be monitored with |
| 185 |
.BR select (2), |
| 186 |
.BR poll (2), |
| 187 |
or |
| 188 |
.BR epoll (7).) |
| 189 |
.\" FIXME |
| 190 |
.\" Christian Brauner: |
| 191 |
.\" |
| 192 |
.\" Do we support O_NONBLOCK with SECCOMP_IOCTL_NOTIF_RECV and if |
| 193 |
.\" not should we? |
| 194 |
.\" |
| 195 |
.\" Michael Kerrisk: |
| 196 |
.\" |
| 197 |
.\" A quick test suggests that O_NONBLOCK has no effect on the blocking |
| 198 |
.\" behavior of SECCOMP_IOCTL_NOTIF_RECV. |
| 199 |
. |
| 200 |
.\"------------------------------------- |
| 201 |
.IP (6) |
| 202 |
The |
| 203 |
.I seccomp_notif |
| 204 |
structure returned by the |
| 205 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 206 |
operation includes the same information (a |
| 207 |
.I seccomp_data |
| 208 |
structure) that was passed to the seccomp filter. |
| 209 |
This information allows the supervisor to discover the system call number and |
| 210 |
the arguments for the target's system call. |
| 211 |
In addition, the notification event contains the ID of the thread |
| 212 |
that triggered the notification and a unique cookie value that |
| 213 |
is used in subsequent |
| 214 |
.B SECCOMP_IOCTL_NOTIF_ID_VALID |
| 215 |
and |
| 216 |
.B SECCOMP_IOCTL_NOTIF_SEND |
| 217 |
operations. |
| 218 |
.IP |
| 219 |
The information in the notification can be used to discover the |
| 220 |
values of pointer arguments for the target's system call. |
| 221 |
(This is something that can't be done from within a seccomp filter.) |
| 222 |
One way in which the supervisor can do this is to open the corresponding |
| 223 |
.IR /proc/ tid /mem |
| 224 |
file (see |
| 225 |
.BR proc (5)) |
| 226 |
and read bytes from the location that corresponds to one of |
| 227 |
the pointer arguments whose value is supplied in the notification event. |
| 228 |
.\" Tycho Andersen mentioned that there are alternatives to /proc/PID/mem, |
| 229 |
.\" such as ptrace() and /proc/PID/map_files |
| 230 |
(The supervisor must be careful to avoid |
| 231 |
a race condition that can occur when doing this; |
| 232 |
see the description of the |
| 233 |
.B SECCOMP_IOCTL_NOTIF_ID_VALID |
| 234 |
.BR ioctl (2) |
| 235 |
operation below.) |
| 236 |
In addition, |
| 237 |
the supervisor can access other system information that is visible |
| 238 |
in user space but which is not accessible from a seccomp filter. |
| 239 |
.\"------------------------------------- |
| 240 |
.IP (7) |
| 241 |
Having obtained information as per the previous step, |
| 242 |
the supervisor may then choose to perform an action in response |
| 243 |
to the target's system call |
| 244 |
(which, as noted above, is not executed when the seccomp filter returns the |
| 245 |
.B SECCOMP_RET_USER_NOTIF |
| 246 |
action value). |
| 247 |
.IP |
| 248 |
One example use case here relates to containers. |
| 249 |
The target may be located inside a container where |
| 250 |
it does not have sufficient capabilities to mount a filesystem |
| 251 |
in the container's mount namespace. |
| 252 |
However, the supervisor may be a more privileged process that |
| 253 |
does have sufficient capabilities to perform the mount operation. |
| 254 |
.\"------------------------------------- |
| 255 |
.IP (8) |
| 256 |
The supervisor then sends a response to the notification. |
| 257 |
The information in this response is used by the kernel to construct |
| 258 |
a return value for the target's system call and provide |
| 259 |
a value that will be assigned to the |
| 260 |
.I errno |
| 261 |
variable of the target. |
| 262 |
.IP |
| 263 |
The response is sent using the |
| 264 |
.B SECCOMP_IOCTL_NOTIF_SEND |
| 265 |
.BR ioctl (2) |
| 266 |
operation, which is used to transmit a |
| 267 |
.I seccomp_notif_resp |
| 268 |
structure to the kernel. |
| 269 |
This structure includes a cookie value that the supervisor obtained in the |
| 270 |
.I seccomp_notif |
| 271 |
structure returned by the |
| 272 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 273 |
operation. |
| 274 |
This cookie value allows the kernel to associate the response with the |
| 275 |
target. |
| 276 |
This structure must include the cookie value that the supervisor |
| 277 |
obtained in the |
| 278 |
.I seccomp_notif |
| 279 |
structure returned by the |
| 280 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 281 |
operation; |
| 282 |
the cookie allows the kernel to associate the response with the target. |
| 283 |
.\"------------------------------------- |
| 284 |
.IP (9) |
| 285 |
Once the notification has been sent, |
| 286 |
the system call in the target thread unblocks, |
| 287 |
returning the information that was provided by the supervisor |
| 288 |
in the notification response. |
| 289 |
.\"------------------------------------- |
| 290 |
.P |
| 291 |
As a variation on the last two steps, |
| 292 |
the supervisor can send a response that tells the kernel that |
| 293 |
it should execute the target thread's system call; |
| 294 |
see the discussion of |
| 295 |
.BR SECCOMP_USER_NOTIF_FLAG_CONTINUE , |
| 296 |
below. |
| 297 |
.\" |
| 298 |
.SH IOCTL OPERATIONS |
| 299 |
The following |
| 300 |
.BR ioctl (2) |
| 301 |
operations are supported by the seccomp user-space |
| 302 |
notification file descriptor. |
| 303 |
For each of these operations, the first (file descriptor) argument of |
| 304 |
.BR ioctl (2) |
| 305 |
is the listening file descriptor returned by a call to |
| 306 |
.BR seccomp (2) |
| 307 |
with the |
| 308 |
.B SECCOMP_FILTER_FLAG_NEW_LISTENER |
| 309 |
flag. |
| 310 |
.\" |
| 311 |
.SS SECCOMP_IOCTL_NOTIF_RECV |
| 312 |
The |
| 313 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 314 |
operation (available since Linux 5.0) is used to obtain a user-space |
| 315 |
notification event. |
| 316 |
If no such event is currently pending, |
| 317 |
the operation blocks until an event occurs. |
| 318 |
The third |
| 319 |
.BR ioctl (2) |
| 320 |
argument is a pointer to a structure of the following form |
| 321 |
which contains information about the event. |
| 322 |
This structure must be zeroed out before the call. |
| 323 |
.P |
| 324 |
.in +4n |
| 325 |
.EX |
| 326 |
struct seccomp_notif { |
| 327 |
__u64 id; /* Cookie */ |
| 328 |
__u32 pid; /* TID of target thread */ |
| 329 |
__u32 flags; /* Currently unused (0) */ |
| 330 |
struct seccomp_data data; /* See seccomp(2) */ |
| 331 |
}; |
| 332 |
.EE |
| 333 |
.in |
| 334 |
.P |
| 335 |
The fields in this structure are as follows: |
| 336 |
.TP |
| 337 |
.I id |
| 338 |
This is a cookie for the notification. |
| 339 |
Each such cookie is guaranteed to be unique for the corresponding |
| 340 |
seccomp filter. |
| 341 |
.RS |
| 342 |
.IP \[bu] 3 |
| 343 |
The cookie can be used with the |
| 344 |
.B SECCOMP_IOCTL_NOTIF_ID_VALID |
| 345 |
.BR ioctl (2) |
| 346 |
operation described below. |
| 347 |
.IP \[bu] |
| 348 |
When returning a notification response to the kernel, |
| 349 |
the supervisor must include the cookie value in the |
| 350 |
.I seccomp_notif_resp |
| 351 |
structure that is specified as the argument of the |
| 352 |
.B SECCOMP_IOCTL_NOTIF_SEND |
| 353 |
operation. |
| 354 |
.RE |
| 355 |
.TP |
| 356 |
.I pid |
| 357 |
This is the thread ID of the target thread that triggered |
| 358 |
the notification event. |
| 359 |
.TP |
| 360 |
.I flags |
| 361 |
This is a bit mask of flags providing further information on the event. |
| 362 |
In the current implementation, this field is always zero. |
| 363 |
.TP |
| 364 |
.I data |
| 365 |
This is a |
| 366 |
.I seccomp_data |
| 367 |
structure containing information about the system call that |
| 368 |
triggered the notification. |
| 369 |
This is the same structure that is passed to the seccomp filter. |
| 370 |
See |
| 371 |
.BR seccomp (2) |
| 372 |
for details of this structure. |
| 373 |
.P |
| 374 |
On success, |
| 375 |
this operation returns 0; |
| 376 |
on failure, \-1 is returned, and |
| 377 |
.I errno |
| 378 |
is set to indicate the error. |
| 379 |
This operation can fail with the following errors: |
| 380 |
.TP |
| 381 |
.BR EINVAL " (since Linux 5.5)" |
| 382 |
.\" commit 2882d53c9c6f3b8311d225062522f03772cf0179 |
| 383 |
The |
| 384 |
.I seccomp_notif |
| 385 |
structure that was passed to the call contained nonzero fields. |
| 386 |
.TP |
| 387 |
.B ENOENT |
| 388 |
The target thread was killed by a signal as the notification information |
| 389 |
was being generated, |
| 390 |
or the target's (blocked) system call was interrupted by a signal handler. |
| 391 |
.\" FIXME |
| 392 |
.\" From my experiments, |
| 393 |
.\" it appears that if a SECCOMP_IOCTL_NOTIF_RECV is done after |
| 394 |
.\" the target thread terminates, then the ioctl() simply |
| 395 |
.\" blocks (rather than returning an error to indicate that the |
| 396 |
.\" target no longer exists). |
| 397 |
.\" |
| 398 |
.\" I found that surprising, and it required some contortions in |
| 399 |
.\" the example program. It was not possible to code my SIGCHLD |
| 400 |
.\" handler (which reaps the zombie when the worker/target |
| 401 |
.\" terminates) to simply set a flag checked in the main |
| 402 |
.\" handleNotifications() loop, since this created an |
| 403 |
.\" unavoidable race where the child might terminate just after |
| 404 |
.\" I had checked the flag, but before I blocked (forever!) in the |
| 405 |
.\" SECCOMP_IOCTL_NOTIF_RECV operation. Instead, I had to code |
| 406 |
.\" the signal handler to simply call _exit(2) in order to |
| 407 |
.\" terminate the parent process (the supervisor). |
| 408 |
.\" |
| 409 |
.\" Is this expected behavior? It seems to me rather |
| 410 |
.\" desirable that SECCOMP_IOCTL_NOTIF_RECV should give an error |
| 411 |
.\" if the target has terminated. |
| 412 |
.\" |
| 413 |
.\" Jann posted a patch to rectify this, but there was no response |
| 414 |
.\" (Lore link: https://bit.ly/3jvUBxk) to his question about fixing |
| 415 |
.\" this issue. (I've tried building with the patch, but encountered |
| 416 |
.\" an issue with the target process entering D state after a signal.) |
| 417 |
.\" |
| 418 |
.\" For now, this behavior is documented in BUGS. |
| 419 |
.\" |
| 420 |
.\" Kees Cook commented: Let's change [this] ASAP! |
| 421 |
.\" |
| 422 |
.SS SECCOMP_IOCTL_NOTIF_ID_VALID |
| 423 |
The |
| 424 |
.B SECCOMP_IOCTL_NOTIF_ID_VALID |
| 425 |
operation (available since Linux 5.0) is used to check that a notification ID |
| 426 |
returned by an earlier |
| 427 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 428 |
operation is still valid |
| 429 |
(i.e., that the target still exists and its system call |
| 430 |
is still blocked waiting for a response). |
| 431 |
.P |
| 432 |
The third |
| 433 |
.BR ioctl (2) |
| 434 |
argument is a pointer to the cookie |
| 435 |
.RI ( id ) |
| 436 |
returned by the |
| 437 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 438 |
operation. |
| 439 |
.P |
| 440 |
This operation is necessary to avoid race conditions that can occur when the |
| 441 |
.I pid |
| 442 |
returned by the |
| 443 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 444 |
operation terminates, and that process ID is reused by another process. |
| 445 |
An example of this kind of race is the following |
| 446 |
.IP (1) 5 |
| 447 |
A notification is generated on the listening file descriptor. |
| 448 |
The returned |
| 449 |
.I seccomp_notif |
| 450 |
contains the TID of the target thread (in the |
| 451 |
.I pid |
| 452 |
field of the structure). |
| 453 |
.IP (2) |
| 454 |
The target terminates. |
| 455 |
.IP (3) |
| 456 |
Another thread or process is created on the system that by chance reuses the |
| 457 |
TID that was freed when the target terminated. |
| 458 |
.IP (4) |
| 459 |
The supervisor |
| 460 |
.BR open (2)s |
| 461 |
the |
| 462 |
.IR /proc/ tid /mem |
| 463 |
file for the TID obtained in step 1, with the intention of (say) |
| 464 |
inspecting the memory location(s) that containing the argument(s) of |
| 465 |
the system call that triggered the notification in step 1. |
| 466 |
.P |
| 467 |
In the above scenario, the risk is that the supervisor may try |
| 468 |
to access the memory of a process other than the target. |
| 469 |
This race can be avoided by following the call to |
| 470 |
.BR open (2) |
| 471 |
with a |
| 472 |
.B SECCOMP_IOCTL_NOTIF_ID_VALID |
| 473 |
operation to verify that the process that generated the notification |
| 474 |
is still alive. |
| 475 |
(Note that if the target terminates after the latter step, |
| 476 |
a subsequent |
| 477 |
.BR read (2) |
| 478 |
from the file descriptor may return 0, indicating end of file.) |
| 479 |
.\" Jann Horn: |
| 480 |
.\" the PID can be reused, but the /proc/$pid directory is |
| 481 |
.\" internally not associated with the numeric PID, but, |
| 482 |
.\" conceptually speaking, with a specific incarnation of the |
| 483 |
.\" PID, or something like that. (Actually, it is associated |
| 484 |
.\" with the "struct pid", which is not reused, instead of the |
| 485 |
.\" numeric PID. |
| 486 |
.P |
| 487 |
See NOTES for a discussion of other cases where |
| 488 |
.B SECCOMP_IOCTL_NOTIF_ID_VALID |
| 489 |
checks must be performed. |
| 490 |
.P |
| 491 |
On success (i.e., the notification ID is still valid), |
| 492 |
this operation returns 0. |
| 493 |
On failure (i.e., the notification ID is no longer valid), |
| 494 |
\-1 is returned, and |
| 495 |
.I errno |
| 496 |
is set to |
| 497 |
.BR ENOENT . |
| 498 |
.\" |
| 499 |
.SS SECCOMP_IOCTL_NOTIF_SEND |
| 500 |
The |
| 501 |
.B SECCOMP_IOCTL_NOTIF_SEND |
| 502 |
operation (available since Linux 5.0) |
| 503 |
is used to send a notification response back to the kernel. |
| 504 |
The third |
| 505 |
.BR ioctl (2) |
| 506 |
argument of this structure is a pointer to a structure of the following form: |
| 507 |
.P |
| 508 |
.in +4n |
| 509 |
.EX |
| 510 |
struct seccomp_notif_resp { |
| 511 |
__u64 id; /* Cookie value */ |
| 512 |
__s64 val; /* Success return value */ |
| 513 |
__s32 error; /* 0 (success) or negative error number */ |
| 514 |
__u32 flags; /* See below */ |
| 515 |
}; |
| 516 |
.EE |
| 517 |
.in |
| 518 |
.P |
| 519 |
The fields of this structure are as follows: |
| 520 |
.TP |
| 521 |
.I id |
| 522 |
This is the cookie value that was obtained using the |
| 523 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 524 |
operation. |
| 525 |
This cookie value allows the kernel to correctly associate this response |
| 526 |
with the system call that triggered the user-space notification. |
| 527 |
.TP |
| 528 |
.I val |
| 529 |
This is the value that will be used for |
| 530 |
a spoofed success return for the target's system call; |
| 531 |
see below. |
| 532 |
.TP |
| 533 |
.I error |
| 534 |
This is the value that will be used as the error number |
| 535 |
.RI ( errno ) |
| 536 |
for a spoofed error return for the target's system call; |
| 537 |
see below. |
| 538 |
.TP |
| 539 |
.I flags |
| 540 |
This is a bit mask that includes zero or more of the following flags: |
| 541 |
.RS |
| 542 |
.TP |
| 543 |
.BR SECCOMP_USER_NOTIF_FLAG_CONTINUE " (since Linux 5.5)" |
| 544 |
Tell the kernel to execute the target's system call. |
| 545 |
.\" commit fb3c5386b382d4097476ce9647260fc89b34afdb |
| 546 |
.RE |
| 547 |
.P |
| 548 |
Two kinds of response are possible: |
| 549 |
.IP \[bu] 3 |
| 550 |
A response to the kernel telling it to execute the |
| 551 |
target's system call. |
| 552 |
In this case, the |
| 553 |
.I flags |
| 554 |
field includes |
| 555 |
.B SECCOMP_USER_NOTIF_FLAG_CONTINUE |
| 556 |
and the |
| 557 |
.I error |
| 558 |
and |
| 559 |
.I val |
| 560 |
fields must be zero. |
| 561 |
.IP |
| 562 |
This kind of response can be useful in cases where the supervisor needs |
| 563 |
to do deeper analysis of the target's system call than is possible |
| 564 |
from a seccomp filter (e.g., examining the values of pointer arguments), |
| 565 |
and, having decided that the system call does not require emulation |
| 566 |
by the supervisor, the supervisor wants the system call to |
| 567 |
be executed normally in the target. |
| 568 |
.IP |
| 569 |
The |
| 570 |
.B SECCOMP_USER_NOTIF_FLAG_CONTINUE |
| 571 |
flag should be used with caution; |
| 572 |
see NOTES. |
| 573 |
.IP \[bu] |
| 574 |
A spoofed return value for the target's system call. |
| 575 |
In this case, the kernel does not execute the target's system call, |
| 576 |
instead causing the system call to return a spoofed value as specified by |
| 577 |
fields of the |
| 578 |
.I seccomp_notif_resp |
| 579 |
structure. |
| 580 |
The supervisor should set the fields of this structure as follows: |
| 581 |
.RS |
| 582 |
.IP \[bu] 3 |
| 583 |
.I flags |
| 584 |
does not contain |
| 585 |
.BR SECCOMP_USER_NOTIF_FLAG_CONTINUE . |
| 586 |
.IP \[bu] |
| 587 |
.I error |
| 588 |
is set either to 0 for a spoofed "success" return or to a negative |
| 589 |
error number for a spoofed "failure" return. |
| 590 |
In the former case, the kernel causes the target's system call |
| 591 |
to return the value specified in the |
| 592 |
.I val |
| 593 |
field. |
| 594 |
In the latter case, the kernel causes the target's system call |
| 595 |
to return \-1, and |
| 596 |
.I errno |
| 597 |
is assigned the negated |
| 598 |
.I error |
| 599 |
value. |
| 600 |
.IP \[bu] |
| 601 |
.I val |
| 602 |
is set to a value that will be used as the return value for a spoofed |
| 603 |
"success" return for the target's system call. |
| 604 |
The value in this field is ignored if the |
| 605 |
.I error |
| 606 |
field contains a nonzero value. |
| 607 |
.\" FIXME |
| 608 |
.\" Kees Cook suggested: |
| 609 |
.\" |
| 610 |
.\" Strictly speaking, this is architecture specific, but |
| 611 |
.\" all architectures do it this way. Should seccomp enforce |
| 612 |
.\" val == 0 when err != 0 ? |
| 613 |
.\" |
| 614 |
.\" Christian Brauner |
| 615 |
.\" |
| 616 |
.\" Feels like it should, at least for the SEND ioctl where we already |
| 617 |
.\" verify that val and err are both 0 when CONTINUE is specified (as you |
| 618 |
.\" pointed out correctly above). |
| 619 |
.RE |
| 620 |
.P |
| 621 |
On success, |
| 622 |
this operation returns 0; |
| 623 |
on failure, \-1 is returned, and |
| 624 |
.I errno |
| 625 |
is set to indicate the error. |
| 626 |
This operation can fail with the following errors: |
| 627 |
.TP |
| 628 |
.B EINPROGRESS |
| 629 |
A response to this notification has already been sent. |
| 630 |
.TP |
| 631 |
.B EINVAL |
| 632 |
An invalid value was specified in the |
| 633 |
.I flags field. |
| 634 |
.TP |
| 635 |
.B |
| 636 |
.B EINVAL |
| 637 |
The |
| 638 |
.I flags |
| 639 |
field contained |
| 640 |
.BR SECCOMP_USER_NOTIF_FLAG_CONTINUE , |
| 641 |
and the |
| 642 |
.I error |
| 643 |
or |
| 644 |
.I val |
| 645 |
field was not zero. |
| 646 |
.TP |
| 647 |
.B ENOENT |
| 648 |
The blocked system call in the target |
| 649 |
has been interrupted by a signal handler |
| 650 |
or the target has terminated. |
| 651 |
.\" Jann Horn notes: |
| 652 |
.\" you could also get this [ENOENT] if a response has already |
| 653 |
.\" been sent, instead of EINPROGRESS - the only difference is |
| 654 |
.\" whether the target thread has picked up the response yet |
| 655 |
.\" |
| 656 |
.SS SECCOMP_IOCTL_NOTIF_ADDFD |
| 657 |
The |
| 658 |
.B SECCOMP_IOCTL_NOTIF_ADDFD |
| 659 |
operation (available since Linux 5.9) |
| 660 |
allows the supervisor to install a file descriptor |
| 661 |
into the target's file descriptor table. |
| 662 |
Much like the use of |
| 663 |
.B SCM_RIGHTS |
| 664 |
messages described in |
| 665 |
.BR unix (7), |
| 666 |
this operation is semantically equivalent to duplicating |
| 667 |
a file descriptor from the supervisor's file descriptor table |
| 668 |
into the target's file descriptor table. |
| 669 |
.P |
| 670 |
The |
| 671 |
.B SECCOMP_IOCTL_NOTIF_ADDFD |
| 672 |
operation permits the supervisor to emulate a target system call (such as |
| 673 |
.BR socket (2) |
| 674 |
or |
| 675 |
.BR openat (2)) |
| 676 |
that generates a file descriptor. |
| 677 |
The supervisor can perform the system call that generates |
| 678 |
the file descriptor (and associated open file description) |
| 679 |
and then use this operation to allocate |
| 680 |
a file descriptor that refers to the same open file description in the target. |
| 681 |
(For an explanation of open file descriptions, see |
| 682 |
.BR open (2).) |
| 683 |
.P |
| 684 |
Once this operation has been performed, |
| 685 |
the supervisor can close its copy of the file descriptor. |
| 686 |
.P |
| 687 |
In the target, |
| 688 |
the received file descriptor is subject to the same |
| 689 |
Linux Security Module (LSM) checks as are applied to a file descriptor |
| 690 |
that is received in an |
| 691 |
.B SCM_RIGHTS |
| 692 |
ancillary message. |
| 693 |
If the file descriptor refers to a socket, |
| 694 |
it inherits the cgroup version 1 network controller settings |
| 695 |
.RI ( classid |
| 696 |
and |
| 697 |
.IR netprioidx ) |
| 698 |
of the target. |
| 699 |
.P |
| 700 |
The third |
| 701 |
.BR ioctl (2) |
| 702 |
argument is a pointer to a structure of the following form: |
| 703 |
.P |
| 704 |
.in +4n |
| 705 |
.EX |
| 706 |
struct seccomp_notif_addfd { |
| 707 |
__u64 id; /* Cookie value */ |
| 708 |
__u32 flags; /* Flags */ |
| 709 |
__u32 srcfd; /* Local file descriptor number */ |
| 710 |
__u32 newfd; /* 0 or desired file descriptor |
| 711 |
number in target */ |
| 712 |
__u32 newfd_flags; /* Flags to set on target file |
| 713 |
descriptor */ |
| 714 |
}; |
| 715 |
.EE |
| 716 |
.in |
| 717 |
.P |
| 718 |
The fields in this structure are as follows: |
| 719 |
.TP |
| 720 |
.I id |
| 721 |
This field should be set to the notification ID |
| 722 |
(cookie value) that was obtained via |
| 723 |
.BR SECCOMP_IOCTL_NOTIF_RECV . |
| 724 |
.TP |
| 725 |
.I flags |
| 726 |
This field is a bit mask of flags that modify the behavior of the operation. |
| 727 |
Currently, only one flag is supported: |
| 728 |
.RS |
| 729 |
.TP |
| 730 |
.B SECCOMP_ADDFD_FLAG_SETFD |
| 731 |
When allocating the file descriptor in the target, |
| 732 |
use the file descriptor number specified in the |
| 733 |
.I newfd |
| 734 |
field. |
| 735 |
.TP |
| 736 |
.BR SECCOMP_ADDFD_FLAG_SEND " (since Linux 5.14)" |
| 737 |
.\" commit 0ae71c7720e3ae3aabd2e8a072d27f7bd173d25c |
| 738 |
Perform the equivalent of |
| 739 |
.B SECCOMP_IOCTL_NOTIF_ADDFD |
| 740 |
plus |
| 741 |
.B SECCOMP_IOCTL_NOTIF_SEND |
| 742 |
as an atomic operation. |
| 743 |
On successful invocation, the target process's |
| 744 |
.I errno |
| 745 |
will be 0 |
| 746 |
and the return value will be the file descriptor number |
| 747 |
that was allocated in the target. |
| 748 |
If allocating the file descriptor in the target fails, |
| 749 |
the target's system call continues to be blocked |
| 750 |
until a successful response is sent. |
| 751 |
.RE |
| 752 |
.TP |
| 753 |
.I srcfd |
| 754 |
This field should be set to the number of the file descriptor |
| 755 |
in the supervisor that is to be duplicated. |
| 756 |
.TP |
| 757 |
.I newfd |
| 758 |
This field determines which file descriptor number is allocated in the target. |
| 759 |
If the |
| 760 |
.B SECCOMP_ADDFD_FLAG_SETFD |
| 761 |
flag is set, |
| 762 |
then this field specifies which file descriptor number should be allocated. |
| 763 |
If this file descriptor number is already open in the target, |
| 764 |
it is atomically closed and reused. |
| 765 |
If the descriptor duplication fails due to an LSM check, or if |
| 766 |
.I srcfd |
| 767 |
is not a valid file descriptor, |
| 768 |
the file descriptor |
| 769 |
.I newfd |
| 770 |
will not be closed in the target process. |
| 771 |
.IP |
| 772 |
If the |
| 773 |
.B SECCOMP_ADDFD_FLAG_SETFD |
| 774 |
flag it not set, then this field must be 0, |
| 775 |
and the kernel allocates the lowest unused file descriptor number |
| 776 |
in the target. |
| 777 |
.TP |
| 778 |
.I newfd_flags |
| 779 |
This field is a bit mask specifying flags that should be set on |
| 780 |
the file descriptor that is received in the target process. |
| 781 |
Currently, only the following flag is implemented: |
| 782 |
.RS |
| 783 |
.TP |
| 784 |
.B O_CLOEXEC |
| 785 |
Set the close-on-exec flag on the received file descriptor. |
| 786 |
.RE |
| 787 |
.P |
| 788 |
On success, this |
| 789 |
.BR ioctl (2) |
| 790 |
call returns the number of the file descriptor that was allocated |
| 791 |
in the target. |
| 792 |
Assuming that the emulated system call is one that returns |
| 793 |
a file descriptor as its function result (e.g., |
| 794 |
.BR socket (2)), |
| 795 |
this value can be used as the return value |
| 796 |
.RI ( resp.val ) |
| 797 |
that is supplied in the response that is subsequently sent with the |
| 798 |
.B SECCOMP_IOCTL_NOTIF_SEND |
| 799 |
operation. |
| 800 |
.P |
| 801 |
On error, \-1 is returned and |
| 802 |
.I errno |
| 803 |
is set to indicate the error. |
| 804 |
.P |
| 805 |
This operation can fail with the following errors: |
| 806 |
.TP |
| 807 |
.B EBADF |
| 808 |
Allocating the file descriptor in the target would cause the target's |
| 809 |
.B RLIMIT_NOFILE |
| 810 |
limit to be exceeded (see |
| 811 |
.BR getrlimit (2)). |
| 812 |
.TP |
| 813 |
.B EBUSY |
| 814 |
If the flag |
| 815 |
.B SECCOMP_IOCTL_NOTIF_SEND |
| 816 |
is used, this means the operation can't proceed until other |
| 817 |
.B SECCOMP_IOCTL_NOTIF_ADDFD |
| 818 |
requests are processed. |
| 819 |
.TP |
| 820 |
.B EINPROGRESS |
| 821 |
The user-space notification specified in the |
| 822 |
.I id |
| 823 |
field exists but has not yet been fetched (by a |
| 824 |
.BR SECCOMP_IOCTL_NOTIF_RECV ) |
| 825 |
or has already been responded to (by a |
| 826 |
.BR SECCOMP_IOCTL_NOTIF_SEND ). |
| 827 |
.TP |
| 828 |
.B EINVAL |
| 829 |
An invalid flag was specified in the |
| 830 |
.I flags |
| 831 |
or |
| 832 |
.I newfd_flags |
| 833 |
field, or the |
| 834 |
.I newfd |
| 835 |
field is nonzero and the |
| 836 |
.B SECCOMP_ADDFD_FLAG_SETFD |
| 837 |
flag was not specified in the |
| 838 |
.I flags |
| 839 |
field. |
| 840 |
.TP |
| 841 |
.B EMFILE |
| 842 |
The file descriptor number specified in |
| 843 |
.I newfd |
| 844 |
exceeds the limit specified in |
| 845 |
.IR /proc/sys/fs/nr_open . |
| 846 |
.TP |
| 847 |
.B ENOENT |
| 848 |
The blocked system call in the target |
| 849 |
has been interrupted by a signal handler |
| 850 |
or the target has terminated. |
| 851 |
.P |
| 852 |
Here is some sample code (with error handling omitted) that uses the |
| 853 |
.B SECCOMP_ADDFD_FLAG_SETFD |
| 854 |
operation (here, to emulate a call to |
| 855 |
.BR openat (2)): |
| 856 |
.P |
| 857 |
.EX |
| 858 |
.in +4n |
| 859 |
int fd, removeFd; |
| 860 |
\& |
| 861 |
fd = openat(req\->data.args[0], path, req\->data.args[2], |
| 862 |
req\->data.args[3]); |
| 863 |
\& |
| 864 |
struct seccomp_notif_addfd addfd; |
| 865 |
addfd.id = req\->id; /* Cookie from SECCOMP_IOCTL_NOTIF_RECV */ |
| 866 |
addfd.srcfd = fd; |
| 867 |
addfd.newfd = 0; |
| 868 |
addfd.flags = 0; |
| 869 |
addfd.newfd_flags = O_CLOEXEC; |
| 870 |
\& |
| 871 |
targetFd = ioctl(notifyFd, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); |
| 872 |
\& |
| 873 |
close(fd); /* No longer needed in supervisor */ |
| 874 |
\& |
| 875 |
struct seccomp_notif_resp *resp; |
| 876 |
/* Code to allocate 'resp' omitted */ |
| 877 |
resp\->id = req\->id; |
| 878 |
resp\->error = 0; /* "Success" */ |
| 879 |
resp\->val = targetFd; |
| 880 |
resp\->flags = 0; |
| 881 |
ioctl(notifyFd, SECCOMP_IOCTL_NOTIF_SEND, resp); |
| 882 |
.in |
| 883 |
.EE |
| 884 |
.SH NOTES |
| 885 |
One example use case for the user-space notification |
| 886 |
mechanism is to allow a container manager |
| 887 |
(a process which is typically running with more privilege than |
| 888 |
the processes inside the container) |
| 889 |
to mount block devices or create device nodes for the container. |
| 890 |
The mount use case provides an example of where the |
| 891 |
.B SECCOMP_USER_NOTIF_FLAG_CONTINUE |
| 892 |
.BR ioctl (2) |
| 893 |
operation is useful. |
| 894 |
Upon receiving a notification for the |
| 895 |
.BR mount (2) |
| 896 |
system call, the container manager (the "supervisor") can distinguish |
| 897 |
a request to mount a block filesystem |
| 898 |
(which would not be possible for a "target" process inside the container) |
| 899 |
and mount that file system. |
| 900 |
If, on the other hand, the container manager detects that the operation |
| 901 |
could be performed by the process inside the container |
| 902 |
(e.g., a mount of a |
| 903 |
.BR tmpfs (5) |
| 904 |
filesystem), it can notify the kernel that the target process's |
| 905 |
.BR mount (2) |
| 906 |
system call can continue. |
| 907 |
.\" |
| 908 |
.SS select()/poll()/epoll semantics |
| 909 |
The file descriptor returned when |
| 910 |
.BR seccomp (2) |
| 911 |
is employed with the |
| 912 |
.B SECCOMP_FILTER_FLAG_NEW_LISTENER |
| 913 |
flag can be monitored using |
| 914 |
.BR poll (2), |
| 915 |
.BR epoll (7), |
| 916 |
and |
| 917 |
.BR select (2). |
| 918 |
These interfaces indicate that the file descriptor is ready as follows: |
| 919 |
.IP \[bu] 3 |
| 920 |
When a notification is pending, |
| 921 |
these interfaces indicate that the file descriptor is readable. |
| 922 |
Following such an indication, a subsequent |
| 923 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 924 |
.BR ioctl (2) |
| 925 |
will not block, returning either information about a notification |
| 926 |
or else failing with the error |
| 927 |
.B EINTR |
| 928 |
if the target has been killed by a signal or its system call |
| 929 |
has been interrupted by a signal handler. |
| 930 |
.IP \[bu] |
| 931 |
After the notification has been received (i.e., by the |
| 932 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 933 |
.BR ioctl (2) |
| 934 |
operation), these interfaces indicate that the file descriptor is writable, |
| 935 |
meaning that a notification response can be sent using the |
| 936 |
.B SECCOMP_IOCTL_NOTIF_SEND |
| 937 |
.BR ioctl (2) |
| 938 |
operation. |
| 939 |
.IP \[bu] |
| 940 |
After the last thread using the filter has terminated and been reaped using |
| 941 |
.BR waitpid (2) |
| 942 |
(or similar), |
| 943 |
the file descriptor indicates an end-of-file condition (readable in |
| 944 |
.BR select (2); |
| 945 |
.BR POLLHUP / EPOLLHUP |
| 946 |
in |
| 947 |
.BR poll (2)/ |
| 948 |
.BR epoll_wait (2)). |
| 949 |
.SS Design goals; use of SECCOMP_USER_NOTIF_FLAG_CONTINUE |
| 950 |
The intent of the user-space notification feature is |
| 951 |
to allow system calls to be performed on behalf of the target. |
| 952 |
The target's system call should either be handled by the supervisor or |
| 953 |
allowed to continue normally in the kernel (where standard security |
| 954 |
policies will be applied). |
| 955 |
.P |
| 956 |
.BR "Note well" : |
| 957 |
this mechanism must not be used to make security policy decisions |
| 958 |
about the system call, |
| 959 |
which would be inherently race-prone for reasons described next. |
| 960 |
.P |
| 961 |
The |
| 962 |
.B SECCOMP_USER_NOTIF_FLAG_CONTINUE |
| 963 |
flag must be used with caution. |
| 964 |
If set by the supervisor, the target's system call will continue. |
| 965 |
However, there is a time-of-check, time-of-use race here, |
| 966 |
since an attacker could exploit the interval of time where the target is |
| 967 |
blocked waiting on the "continue" response to do things such as |
| 968 |
rewriting the system call arguments. |
| 969 |
.P |
| 970 |
Note furthermore that a user-space notifier can be bypassed if |
| 971 |
the existing filters allow the use of |
| 972 |
.BR seccomp (2) |
| 973 |
or |
| 974 |
.BR prctl (2) |
| 975 |
to install a filter that returns an action value with a higher precedence than |
| 976 |
.B SECCOMP_RET_USER_NOTIF |
| 977 |
(see |
| 978 |
.BR seccomp (2)). |
| 979 |
.P |
| 980 |
It should thus be absolutely clear that the |
| 981 |
seccomp user-space notification mechanism |
| 982 |
.B can not |
| 983 |
be used to implement a security policy! |
| 984 |
It should only ever be used in scenarios where a more privileged process |
| 985 |
supervises the system calls of a lesser privileged target to |
| 986 |
get around kernel-enforced security restrictions when |
| 987 |
the supervisor deems this safe. |
| 988 |
In other words, |
| 989 |
in order to continue a system call, the supervisor should be sure that |
| 990 |
another security mechanism or the kernel itself will sufficiently block |
| 991 |
the system call if its arguments are rewritten to something unsafe. |
| 992 |
.\" |
| 993 |
.SS Caveats regarding the use of |
| 994 |
.IR /proc/ tid /mem |
| 995 |
The discussion above noted the need to use the |
| 996 |
.B SECCOMP_IOCTL_NOTIF_ID_VALID |
| 997 |
.BR ioctl (2) |
| 998 |
when opening the |
| 999 |
.IR /proc/ tid /mem |
| 1000 |
file of the target |
| 1001 |
to avoid the possibility of accessing the memory of the wrong process |
| 1002 |
in the event that the target terminates and its ID |
| 1003 |
is recycled by another (unrelated) thread. |
| 1004 |
However, the use of this |
| 1005 |
.BR ioctl (2) |
| 1006 |
operation is also necessary in other situations, |
| 1007 |
as explained in the following paragraphs. |
| 1008 |
.P |
| 1009 |
Consider the following scenario, where the supervisor |
| 1010 |
tries to read the pathname argument of a target's blocked |
| 1011 |
.BR mount (2) |
| 1012 |
system call: |
| 1013 |
.IP (1) 5 |
| 1014 |
From one of its functions |
| 1015 |
.RI ( func() ), |
| 1016 |
the target calls |
| 1017 |
.BR mount (2), |
| 1018 |
which triggers a user-space notification and causes the target to block. |
| 1019 |
.IP (2) |
| 1020 |
The supervisor receives the notification, opens |
| 1021 |
.IR /proc/ tid /mem , |
| 1022 |
and (successfully) performs the |
| 1023 |
.B SECCOMP_IOCTL_NOTIF_ID_VALID |
| 1024 |
check. |
| 1025 |
.IP (3) |
| 1026 |
The target receives a signal, which causes the |
| 1027 |
.BR mount (2) |
| 1028 |
to abort. |
| 1029 |
.IP (4) |
| 1030 |
The signal handler executes in the target, and returns. |
| 1031 |
.IP (5) |
| 1032 |
Upon return from the handler, the execution of |
| 1033 |
.I func() |
| 1034 |
resumes, and it returns (and perhaps other functions are called, |
| 1035 |
overwriting the memory that had been used for the stack frame of |
| 1036 |
.IR func() ). |
| 1037 |
.IP (6) |
| 1038 |
Using the address provided in the notification information, |
| 1039 |
the supervisor reads from the target's memory location that used to |
| 1040 |
contain the pathname. |
| 1041 |
.IP (7) |
| 1042 |
The supervisor now calls |
| 1043 |
.BR mount (2) |
| 1044 |
with some arbitrary bytes obtained in the previous step. |
| 1045 |
.P |
| 1046 |
The conclusion from the above scenario is this: |
| 1047 |
since the target's blocked system call may be interrupted by a signal handler, |
| 1048 |
the supervisor must be written to expect that the |
| 1049 |
target may abandon its system call at |
| 1050 |
.B any |
| 1051 |
time; |
| 1052 |
in such an event, any information that the supervisor obtained from |
| 1053 |
the target's memory must be considered invalid. |
| 1054 |
.P |
| 1055 |
To prevent such scenarios, |
| 1056 |
every read from the target's memory must be separated from use of |
| 1057 |
the bytes so obtained by a |
| 1058 |
.B SECCOMP_IOCTL_NOTIF_ID_VALID |
| 1059 |
check. |
| 1060 |
In the above example, the check would be placed between the two final steps. |
| 1061 |
An example of such a check is shown in EXAMPLES. |
| 1062 |
.P |
| 1063 |
Following on from the above, it should be clear that |
| 1064 |
a write by the supervisor into the target's memory can |
| 1065 |
.B never |
| 1066 |
be considered safe. |
| 1067 |
.\" |
| 1068 |
.SS Caveats regarding blocking system calls |
| 1069 |
Suppose that the target performs a blocking system call (e.g., |
| 1070 |
.BR accept (2)) |
| 1071 |
that the supervisor should handle. |
| 1072 |
The supervisor might then in turn execute the same blocking system call. |
| 1073 |
.P |
| 1074 |
In this scenario, |
| 1075 |
it is important to note that if the target's system call is now |
| 1076 |
interrupted by a signal, the supervisor is |
| 1077 |
.I not |
| 1078 |
informed of this. |
| 1079 |
If the supervisor does not take suitable steps to |
| 1080 |
actively discover that the target's system call has been canceled, |
| 1081 |
various difficulties can occur. |
| 1082 |
Taking the example of |
| 1083 |
.BR accept (2), |
| 1084 |
the supervisor might remain blocked in its |
| 1085 |
.BR accept (2) |
| 1086 |
holding a port number that the target |
| 1087 |
(which, after the interruption by the signal handler, |
| 1088 |
perhaps closed its listening socket) |
| 1089 |
might expect to be able to reuse in a |
| 1090 |
.BR bind (2) |
| 1091 |
call. |
| 1092 |
.P |
| 1093 |
Therefore, when the supervisor wishes to emulate a blocking system call, |
| 1094 |
it must do so in such a way that it gets informed if the target's |
| 1095 |
system call is interrupted by a signal handler. |
| 1096 |
For example, if the supervisor itself executes the same |
| 1097 |
blocking system call, then it could employ a separate thread |
| 1098 |
that uses the |
| 1099 |
.B SECCOMP_IOCTL_NOTIF_ID_VALID |
| 1100 |
operation to check if the target is still blocked in its system call. |
| 1101 |
Alternatively, in the |
| 1102 |
.BR accept (2) |
| 1103 |
example, the supervisor might use |
| 1104 |
.BR poll (2) |
| 1105 |
to monitor both the notification file descriptor |
| 1106 |
(so as to discover when the target's |
| 1107 |
.BR accept (2) |
| 1108 |
call has been interrupted) and the listening file descriptor |
| 1109 |
(so as to know when a connection is available). |
| 1110 |
.P |
| 1111 |
If the target's system call is interrupted, |
| 1112 |
the supervisor must take care to release resources (e.g., file descriptors) |
| 1113 |
that it acquired on behalf of the target. |
| 1114 |
.\" |
| 1115 |
.SS Interaction with SA_RESTART signal handlers |
| 1116 |
Consider the following scenario: |
| 1117 |
.IP (1) 5 |
| 1118 |
The target process has used |
| 1119 |
.BR sigaction (2) |
| 1120 |
to install a signal handler with the |
| 1121 |
.B SA_RESTART |
| 1122 |
flag. |
| 1123 |
.IP (2) |
| 1124 |
The target has made a system call that triggered a seccomp |
| 1125 |
user-space notification and the target is currently blocked |
| 1126 |
until the supervisor sends a notification response. |
| 1127 |
.IP (3) |
| 1128 |
A signal is delivered to the target and the signal handler is executed. |
| 1129 |
.IP (4) |
| 1130 |
When (if) the supervisor attempts to send a notification response, the |
| 1131 |
.B SECCOMP_IOCTL_NOTIF_SEND |
| 1132 |
.BR ioctl (2)) |
| 1133 |
operation will fail with the |
| 1134 |
.B ENOENT |
| 1135 |
error. |
| 1136 |
.P |
| 1137 |
In this scenario, the kernel will restart the target's system call. |
| 1138 |
Consequently, the supervisor will receive another user-space notification. |
| 1139 |
Thus, depending on how many times the blocked system call |
| 1140 |
is interrupted by a signal handler, |
| 1141 |
the supervisor may receive multiple notifications for |
| 1142 |
the same instance of a system call in the target. |
| 1143 |
.P |
| 1144 |
One oddity is that system call restarting as described in this scenario |
| 1145 |
will occur even for the blocking system calls listed in |
| 1146 |
.BR signal (7) |
| 1147 |
that would |
| 1148 |
.B never |
| 1149 |
normally be restarted by the |
| 1150 |
.B SA_RESTART |
| 1151 |
flag. |
| 1152 |
.\" FIXME |
| 1153 |
.\" About the above, Kees Cook commented: |
| 1154 |
.\" |
| 1155 |
.\" Does this need fixing? I imagine the correct behavior for this case |
| 1156 |
.\" would be a response to _SEND of EINPROGRESS and the target would see |
| 1157 |
.\" EINTR normally? |
| 1158 |
.\" |
| 1159 |
.\" I mean, it's not like seccomp doesn't already expose weirdness with |
| 1160 |
.\" syscall restarts. Not even arm64 compat agrees[3] with arm32 in this |
| 1161 |
.\" regard. :( |
| 1162 |
. |
| 1163 |
.\" FIXME |
| 1164 |
.\" Michael Kerrisk: |
| 1165 |
.\" I wonder about the effect of this oddity for system calls that |
| 1166 |
.\" are normally nonrestartable because they have timeouts. My |
| 1167 |
.\" understanding is that the kernel doesn't restart those system |
| 1168 |
.\" calls because it's impossible for the kernel to restart the call |
| 1169 |
.\" with the right timeout value. I wonder what happens when those |
| 1170 |
.\" system calls are restarted in the scenario we're discussing.) |
| 1171 |
.P |
| 1172 |
Furthermore, if the supervisor response is a file descriptor |
| 1173 |
added with |
| 1174 |
.BR SECCOMP_IOCTL_NOTIF_ADDFD , |
| 1175 |
then the flag |
| 1176 |
.B SECCOMP_ADDFD_FLAG_SEND |
| 1177 |
can be used to atomically add the file descriptor and return that value, |
| 1178 |
making sure no file descriptors are inadvertently leaked into the target. |
| 1179 |
.SH BUGS |
| 1180 |
If a |
| 1181 |
.B SECCOMP_IOCTL_NOTIF_RECV |
| 1182 |
.BR ioctl (2) |
| 1183 |
operation |
| 1184 |
.\" or a poll/epoll/select |
| 1185 |
is performed after the target terminates, then the |
| 1186 |
.BR ioctl (2) |
| 1187 |
call simply blocks (rather than returning an error to indicate that the |
| 1188 |
target no longer exists). |
| 1189 |
.\" FIXME |
| 1190 |
.\" Comment from Kees Cook: |
| 1191 |
.\" |
| 1192 |
.\" I want this fixed. It caused me no end of pain when building the |
| 1193 |
.\" selftests, and ended up spawning my implementing a global test timeout |
| 1194 |
.\" in kselftest. :P Before the usage counter refactor, there was no sane |
| 1195 |
.\" way to deal with this, but now I think we're close. |
| 1196 |
.\" |
| 1197 |
.SH EXAMPLES |
| 1198 |
The (somewhat contrived) program shown below demonstrates the use of |
| 1199 |
the interfaces described in this page. |
| 1200 |
The program creates a child process that serves as the "target" process. |
| 1201 |
The child process installs a seccomp filter that returns the |
| 1202 |
.B SECCOMP_RET_USER_NOTIF |
| 1203 |
action value if a call is made to |
| 1204 |
.BR mkdir (2). |
| 1205 |
The child process then calls |
| 1206 |
.BR mkdir (2) |
| 1207 |
once for each of the supplied command-line arguments, |
| 1208 |
and reports the result returned by the call. |
| 1209 |
After processing all arguments, the child process terminates. |
| 1210 |
.P |
| 1211 |
The parent process acts as the supervisor, listening for the notifications |
| 1212 |
that are generated when the target process calls |
| 1213 |
.BR mkdir (2). |
| 1214 |
When such a notification occurs, |
| 1215 |
the supervisor examines the memory of the target process (using |
| 1216 |
.IR /proc/ pid /mem ) |
| 1217 |
to discover the pathname argument that was supplied to the |
| 1218 |
.BR mkdir (2) |
| 1219 |
call, and performs one of the following actions: |
| 1220 |
.IP \[bu] 3 |
| 1221 |
If the pathname begins with the prefix "/tmp/", |
| 1222 |
then the supervisor attempts to create the specified directory, |
| 1223 |
and then spoofs a return for the target process based on the return |
| 1224 |
value of the supervisor's |
| 1225 |
.BR mkdir (2) |
| 1226 |
call. |
| 1227 |
In the event that that call succeeds, |
| 1228 |
the spoofed success return value is the length of the pathname. |
| 1229 |
.IP \[bu] |
| 1230 |
If the pathname begins with "./" (i.e., it is a relative pathname), |
| 1231 |
the supervisor sends a |
| 1232 |
.B SECCOMP_USER_NOTIF_FLAG_CONTINUE |
| 1233 |
response to the kernel to say that the kernel should execute |
| 1234 |
the target process's |
| 1235 |
.BR mkdir (2) |
| 1236 |
call. |
| 1237 |
.IP \[bu] |
| 1238 |
If the pathname begins with some other prefix, |
| 1239 |
the supervisor spoofs an error return for the target process, |
| 1240 |
so that the target process's |
| 1241 |
.BR mkdir (2) |
| 1242 |
call appears to fail with the error |
| 1243 |
.B EOPNOTSUPP |
| 1244 |
("Operation not supported"). |
| 1245 |
Additionally, if the specified pathname is exactly "/bye", |
| 1246 |
then the supervisor terminates. |
| 1247 |
.P |
| 1248 |
This program can be used to demonstrate various aspects of the |
| 1249 |
behavior of the seccomp user-space notification mechanism. |
| 1250 |
To help aid such demonstrations, |
| 1251 |
the program logs various messages to show the operation |
| 1252 |
of the target process (lines prefixed "T:") and the supervisor |
| 1253 |
(indented lines prefixed "S:"). |
| 1254 |
.P |
| 1255 |
In the following example, the target attempts to create the directory |
| 1256 |
.IR /tmp/x . |
| 1257 |
Upon receiving the notification, the supervisor creates the directory on the |
| 1258 |
target's behalf, |
| 1259 |
and spoofs a success return to be received by the target process's |
| 1260 |
.BR mkdir (2) |
| 1261 |
call. |
| 1262 |
.P |
| 1263 |
.in +4n |
| 1264 |
.EX |
| 1265 |
.RB $ " ./seccomp_unotify /tmp/x" ; |
| 1266 |
T: PID = 23168 |
| 1267 |
\& |
| 1268 |
T: about to mkdir("/tmp/x") |
| 1269 |
S: got notification (ID 0x17445c4a0f4e0e3c) for PID 23168 |
| 1270 |
S: executing: mkdir("/tmp/x", 0700) |
| 1271 |
S: success! spoofed return = 6 |
| 1272 |
S: sending response (flags = 0; val = 6; error = 0) |
| 1273 |
T: SUCCESS: mkdir(2) returned 6 |
| 1274 |
\& |
| 1275 |
T: terminating |
| 1276 |
S: target has terminated; bye |
| 1277 |
.EE |
| 1278 |
.in |
| 1279 |
.P |
| 1280 |
In the above output, note that the spoofed return value seen by the target |
| 1281 |
process is 6 (the length of the pathname |
| 1282 |
.IR /tmp/x ), |
| 1283 |
whereas a normal |
| 1284 |
.BR mkdir (2) |
| 1285 |
call returns 0 on success. |
| 1286 |
.P |
| 1287 |
In the next example, the target attempts to create a directory using the |
| 1288 |
relative pathname |
| 1289 |
.IR ./sub . |
| 1290 |
Since this pathname starts with "./", |
| 1291 |
the supervisor sends a |
| 1292 |
.B SECCOMP_USER_NOTIF_FLAG_CONTINUE |
| 1293 |
response to the kernel, |
| 1294 |
and the kernel then (successfully) executes the target process's |
| 1295 |
.BR mkdir (2) |
| 1296 |
call. |
| 1297 |
.P |
| 1298 |
.in +4n |
| 1299 |
.EX |
| 1300 |
.RB $ " ./seccomp_unotify ./sub" ; |
| 1301 |
T: PID = 23204 |
| 1302 |
\& |
| 1303 |
T: about to mkdir("./sub") |
| 1304 |
S: got notification (ID 0xddb16abe25b4c12) for PID 23204 |
| 1305 |
S: target can execute system call |
| 1306 |
S: sending response (flags = 0x1; val = 0; error = 0) |
| 1307 |
T: SUCCESS: mkdir(2) returned 0 |
| 1308 |
\& |
| 1309 |
T: terminating |
| 1310 |
S: target has terminated; bye |
| 1311 |
.EE |
| 1312 |
.in |
| 1313 |
.P |
| 1314 |
If the target process attempts to create a directory with |
| 1315 |
a pathname that doesn't start with "." and doesn't begin with the prefix |
| 1316 |
"/tmp/", then the supervisor spoofs an error return |
| 1317 |
.RB ( EOPNOTSUPP , |
| 1318 |
"Operation not supported") |
| 1319 |
for the target's |
| 1320 |
.BR mkdir (2) |
| 1321 |
call (which is not executed): |
| 1322 |
.P |
| 1323 |
.in +4n |
| 1324 |
.EX |
| 1325 |
.RB $ " ./seccomp_unotify /xxx" ; |
| 1326 |
T: PID = 23178 |
| 1327 |
\& |
| 1328 |
T: about to mkdir("/xxx") |
| 1329 |
S: got notification (ID 0xe7dc095d1c524e80) for PID 23178 |
| 1330 |
S: spoofing error response (Operation not supported) |
| 1331 |
S: sending response (flags = 0; val = 0; error = \-95) |
| 1332 |
T: ERROR: mkdir(2): Operation not supported |
| 1333 |
\& |
| 1334 |
T: terminating |
| 1335 |
S: target has terminated; bye |
| 1336 |
.EE |
| 1337 |
.in |
| 1338 |
.P |
| 1339 |
In the next example, |
| 1340 |
the target process attempts to create a directory with the pathname |
| 1341 |
.BR /tmp/nosuchdir/b . |
| 1342 |
Upon receiving the notification, |
| 1343 |
the supervisor attempts to create that directory, but the |
| 1344 |
.BR mkdir (2) |
| 1345 |
call fails because the directory |
| 1346 |
.B /tmp/nosuchdir |
| 1347 |
does not exist. |
| 1348 |
Consequently, the supervisor spoofs an error return that passes the error |
| 1349 |
that it received back to the target process's |
| 1350 |
.BR mkdir (2) |
| 1351 |
call. |
| 1352 |
.P |
| 1353 |
.in +4n |
| 1354 |
.EX |
| 1355 |
.RB $ " ./seccomp_unotify /tmp/nosuchdir/b" ; |
| 1356 |
T: PID = 23199 |
| 1357 |
\& |
| 1358 |
T: about to mkdir("/tmp/nosuchdir/b") |
| 1359 |
S: got notification (ID 0x8744454293506046) for PID 23199 |
| 1360 |
S: executing: mkdir("/tmp/nosuchdir/b", 0700) |
| 1361 |
S: failure! (errno = 2; No such file or directory) |
| 1362 |
S: sending response (flags = 0; val = 0; error = \-2) |
| 1363 |
T: ERROR: mkdir(2): No such file or directory |
| 1364 |
\& |
| 1365 |
T: terminating |
| 1366 |
S: target has terminated; bye |
| 1367 |
.EE |
| 1368 |
.in |
| 1369 |
.P |
| 1370 |
If the supervisor receives a notification and sees that the |
| 1371 |
argument of the target's |
| 1372 |
.BR mkdir (2) |
| 1373 |
is the string "/bye", then (as well as spoofing an |
| 1374 |
.B EOPNOTSUPP |
| 1375 |
error), the supervisor terminates. |
| 1376 |
If the target process subsequently executes another |
| 1377 |
.BR mkdir (2) |
| 1378 |
that triggers its seccomp filter to return the |
| 1379 |
.B SECCOMP_RET_USER_NOTIF |
| 1380 |
action value, then the kernel causes the target process's system call to |
| 1381 |
fail with the error |
| 1382 |
.B ENOSYS |
| 1383 |
("Function not implemented"). |
| 1384 |
This is demonstrated by the following example: |
| 1385 |
.P |
| 1386 |
.in +4n |
| 1387 |
.EX |
| 1388 |
.RB $ " ./seccomp_unotify /bye /tmp/y" ; |
| 1389 |
T: PID = 23185 |
| 1390 |
\& |
| 1391 |
T: about to mkdir("/bye") |
| 1392 |
S: got notification (ID 0xa81236b1d2f7b0f4) for PID 23185 |
| 1393 |
S: spoofing error response (Operation not supported) |
| 1394 |
S: sending response (flags = 0; val = 0; error = \-95) |
| 1395 |
S: terminating ********** |
| 1396 |
T: ERROR: mkdir(2): Operation not supported |
| 1397 |
\& |
| 1398 |
T: about to mkdir("/tmp/y") |
| 1399 |
T: ERROR: mkdir(2): Function not implemented |
| 1400 |
\& |
| 1401 |
T: terminating |
| 1402 |
.EE |
| 1403 |
.in |
| 1404 |
.\" |
| 1405 |
.SS Program source |
| 1406 |
.\" SRC BEGIN (seccomp_unotify.c) |
| 1407 |
.EX |
| 1408 |
#define _GNU_SOURCE |
| 1409 |
#include <err.h> |
| 1410 |
#include <errno.h> |
| 1411 |
#include <fcntl.h> |
| 1412 |
#include <limits.h> |
| 1413 |
#include <linux/audit.h> |
| 1414 |
#include <linux/filter.h> |
| 1415 |
#include <linux/seccomp.h> |
| 1416 |
#include <signal.h> |
| 1417 |
#include <stdbool.h> |
| 1418 |
#include <stdcountof.h> |
| 1419 |
#include <stddef.h> |
| 1420 |
#include <stdint.h> |
| 1421 |
#include <stdio.h> |
| 1422 |
#include <stdlib.h> |
| 1423 |
#include <string.h> |
| 1424 |
#include <sys/ioctl.h> |
| 1425 |
#include <sys/prctl.h> |
| 1426 |
#include <sys/socket.h> |
| 1427 |
#include <sys/stat.h> |
| 1428 |
#include <sys/syscall.h> |
| 1429 |
#include <sys/types.h> |
| 1430 |
#include <sys/un.h> |
| 1431 |
#include <unistd.h> |
| 1432 |
\& |
| 1433 |
#define streq(...) (strcmp(__VA_ARGS__) == 0) |
| 1434 |
\& |
| 1435 |
/* Send the file descriptor \[aq]fd\[aq] over the connected UNIX domain socket |
| 1436 |
\[aq]sockfd\[aq]. Returns 0 on success, or \-1 on error. */ |
| 1437 |
\& |
| 1438 |
static int |
| 1439 |
sendfd(int sockfd, int fd) |
| 1440 |
{ |
| 1441 |
int data; |
| 1442 |
struct iovec iov; |
| 1443 |
struct msghdr msgh; |
| 1444 |
struct cmsghdr *cmsgp; |
| 1445 |
\& |
| 1446 |
/* Allocate a char array of suitable size to hold the ancillary data. |
| 1447 |
However, since this buffer is in reality a \[aq]struct cmsghdr\[aq], use a |
| 1448 |
union to ensure that it is suitably aligned. */ |
| 1449 |
union { |
| 1450 |
char buf[CMSG_SPACE(sizeof(int))]; |
| 1451 |
/* Space large enough to hold an \[aq]int\[aq] */ |
| 1452 |
struct cmsghdr align; |
| 1453 |
} controlMsg; |
| 1454 |
\& |
| 1455 |
/* The \[aq]msg_name\[aq] field can be used to specify the address of the |
| 1456 |
destination socket when sending a datagram. However, we do not |
| 1457 |
need to use this field because \[aq]sockfd\[aq] is a connected socket. */ |
| 1458 |
\& |
| 1459 |
msgh.msg_name = NULL; |
| 1460 |
msgh.msg_namelen = 0; |
| 1461 |
\& |
| 1462 |
/* On Linux, we must transmit at least one byte of real data in |
| 1463 |
order to send ancillary data. We transmit an arbitrary integer |
| 1464 |
whose value is ignored by recvfd(). */ |
| 1465 |
\& |
| 1466 |
msgh.msg_iov = &iov; |
| 1467 |
msgh.msg_iovlen = 1; |
| 1468 |
iov.iov_base = &data; |
| 1469 |
iov.iov_len = sizeof(int); |
| 1470 |
data = 12345; |
| 1471 |
\& |
| 1472 |
/* Set \[aq]msghdr\[aq] fields that describe ancillary data */ |
| 1473 |
\& |
| 1474 |
msgh.msg_control = controlMsg.buf; |
| 1475 |
msgh.msg_controllen = sizeof(controlMsg.buf); |
| 1476 |
\& |
| 1477 |
/* Set up ancillary data describing file descriptor to send */ |
| 1478 |
\& |
| 1479 |
cmsgp = CMSG_FIRSTHDR(&msgh); |
| 1480 |
cmsgp\->cmsg_level = SOL_SOCKET; |
| 1481 |
cmsgp\->cmsg_type = SCM_RIGHTS; |
| 1482 |
cmsgp\->cmsg_len = CMSG_LEN(sizeof(int)); |
| 1483 |
memcpy(CMSG_DATA(cmsgp), &fd, sizeof(int)); |
| 1484 |
\& |
| 1485 |
/* Send real plus ancillary data */ |
| 1486 |
\& |
| 1487 |
if (sendmsg(sockfd, &msgh, 0) == \-1) |
| 1488 |
return \-1; |
| 1489 |
\& |
| 1490 |
return 0; |
| 1491 |
} |
| 1492 |
\& |
| 1493 |
/* Receive a file descriptor on a connected UNIX domain socket. Returns |
| 1494 |
the received file descriptor on success, or \-1 on error. */ |
| 1495 |
\& |
| 1496 |
static int |
| 1497 |
recvfd(int sockfd) |
| 1498 |
{ |
| 1499 |
int data, fd; |
| 1500 |
ssize_t nr; |
| 1501 |
struct iovec iov; |
| 1502 |
struct msghdr msgh; |
| 1503 |
\& |
| 1504 |
/* Allocate a char buffer for the ancillary data. See the comments |
| 1505 |
in sendfd() */ |
| 1506 |
union { |
| 1507 |
char buf[CMSG_SPACE(sizeof(int))]; |
| 1508 |
struct cmsghdr align; |
| 1509 |
} controlMsg; |
| 1510 |
struct cmsghdr *cmsgp; |
| 1511 |
\& |
| 1512 |
/* The \[aq]msg_name\[aq] field can be used to obtain the address of the |
| 1513 |
sending socket. However, we do not need this information. */ |
| 1514 |
\& |
| 1515 |
msgh.msg_name = NULL; |
| 1516 |
msgh.msg_namelen = 0; |
| 1517 |
\& |
| 1518 |
/* Specify buffer for receiving real data */ |
| 1519 |
\& |
| 1520 |
msgh.msg_iov = &iov; |
| 1521 |
msgh.msg_iovlen = 1; |
| 1522 |
iov.iov_base = &data; /* Real data is an \[aq]int\[aq] */ |
| 1523 |
iov.iov_len = sizeof(int); |
| 1524 |
\& |
| 1525 |
/* Set \[aq]msghdr\[aq] fields that describe ancillary data */ |
| 1526 |
\& |
| 1527 |
msgh.msg_control = controlMsg.buf; |
| 1528 |
msgh.msg_controllen = sizeof(controlMsg.buf); |
| 1529 |
\& |
| 1530 |
/* Receive real plus ancillary data; real data is ignored */ |
| 1531 |
\& |
| 1532 |
nr = recvmsg(sockfd, &msgh, 0); |
| 1533 |
if (nr == \-1) |
| 1534 |
return \-1; |
| 1535 |
\& |
| 1536 |
cmsgp = CMSG_FIRSTHDR(&msgh); |
| 1537 |
\& |
| 1538 |
/* Check the validity of the \[aq]cmsghdr\[aq] */ |
| 1539 |
\& |
| 1540 |
if (cmsgp == NULL |
| 1541 |
|| cmsgp\->cmsg_len != CMSG_LEN(sizeof(int)) |
| 1542 |
|| cmsgp\->cmsg_level != SOL_SOCKET |
| 1543 |
|| cmsgp\->cmsg_type != SCM_RIGHTS) |
| 1544 |
{ |
| 1545 |
errno = EINVAL; |
| 1546 |
return \-1; |
| 1547 |
} |
| 1548 |
\& |
| 1549 |
/* Return the received file descriptor to our caller */ |
| 1550 |
\& |
| 1551 |
memcpy(&fd, CMSG_DATA(cmsgp), sizeof(int)); |
| 1552 |
return fd; |
| 1553 |
} |
| 1554 |
\& |
| 1555 |
static void |
| 1556 |
sigchldHandler(int sig) |
| 1557 |
{ |
| 1558 |
char msg[] = "\[rs]tS: target has terminated; bye\[rs]n"; |
| 1559 |
\& |
| 1560 |
write(STDOUT_FILENO, msg, sizeof(msg) \- 1); |
| 1561 |
_exit(EXIT_SUCCESS); |
| 1562 |
} |
| 1563 |
\& |
| 1564 |
static int |
| 1565 |
seccomp(unsigned int operation, unsigned int flags, void *args) |
| 1566 |
{ |
| 1567 |
return syscall(SYS_seccomp, operation, flags, args); |
| 1568 |
} |
| 1569 |
\& |
| 1570 |
/* The following is the x86\-64\-specific BPF boilerplate code for checking |
| 1571 |
that the BPF program is running on the right architecture + ABI. At |
| 1572 |
completion of these instructions, the accumulator contains the system |
| 1573 |
call number. */ |
| 1574 |
\& |
| 1575 |
/* For the x32 ABI, all system call numbers have bit 30 set */ |
| 1576 |
\& |
| 1577 |
#define X32_SYSCALL_BIT 0x40000000 |
| 1578 |
\& |
| 1579 |
#define X86_64_CHECK_ARCH_AND_LOAD_SYSCALL_NR \[rs] |
| 1580 |
BPF_STMT(BPF_LD | BPF_W | BPF_ABS, \[rs] |
| 1581 |
(offsetof(struct seccomp_data, arch))), \[rs] |
| 1582 |
BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, AUDIT_ARCH_X86_64, 0, 2), \[rs] |
| 1583 |
BPF_STMT(BPF_LD | BPF_W | BPF_ABS, \[rs] |
| 1584 |
(offsetof(struct seccomp_data, nr))), \[rs] |
| 1585 |
BPF_JUMP(BPF_JMP | BPF_JGE | BPF_K, X32_SYSCALL_BIT, 0, 1), \[rs] |
| 1586 |
BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL_PROCESS) |
| 1587 |
\& |
| 1588 |
/* installNotifyFilter() installs a seccomp filter that generates |
| 1589 |
user\-space notifications (SECCOMP_RET_USER_NOTIF) when the process |
| 1590 |
calls mkdir(2); the filter allows all other system calls. |
| 1591 |
\& |
| 1592 |
The function return value is a file descriptor from which the |
| 1593 |
user\-space notifications can be fetched. */ |
| 1594 |
\& |
| 1595 |
static int |
| 1596 |
installNotifyFilter(void) |
| 1597 |
{ |
| 1598 |
int notifyFd; |
| 1599 |
\& |
| 1600 |
struct sock_filter filter[] = { |
| 1601 |
X86_64_CHECK_ARCH_AND_LOAD_SYSCALL_NR, |
| 1602 |
\& |
| 1603 |
/* mkdir() triggers notification to user\-space supervisor */ |
| 1604 |
\& |
| 1605 |
BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, SYS_mkdir, 0, 1), |
| 1606 |
BPF_STMT(BPF_RET + BPF_K, SECCOMP_RET_USER_NOTIF), |
| 1607 |
\& |
| 1608 |
/* Every other system call is allowed */ |
| 1609 |
\& |
| 1610 |
BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW), |
| 1611 |
}; |
| 1612 |
\& |
| 1613 |
struct sock_fprog prog = { |
| 1614 |
.len = countof(filter), |
| 1615 |
.filter = filter, |
| 1616 |
}; |
| 1617 |
\& |
| 1618 |
/* Install the filter with the SECCOMP_FILTER_FLAG_NEW_LISTENER flag; |
| 1619 |
as a result, seccomp() returns a notification file descriptor. */ |
| 1620 |
\& |
| 1621 |
notifyFd = seccomp(SECCOMP_SET_MODE_FILTER, |
| 1622 |
SECCOMP_FILTER_FLAG_NEW_LISTENER, &prog); |
| 1623 |
if (notifyFd == \-1) |
| 1624 |
err(EXIT_FAILURE, "seccomp\-install\-notify\-filter"); |
| 1625 |
\& |
| 1626 |
return notifyFd; |
| 1627 |
} |
| 1628 |
\& |
| 1629 |
/* Close a pair of sockets created by socketpair() */ |
| 1630 |
\& |
| 1631 |
static void |
| 1632 |
closeSocketPair(int sockPair[2]) |
| 1633 |
{ |
| 1634 |
if (close(sockPair[0]) == \-1) |
| 1635 |
err(EXIT_FAILURE, "closeSocketPair\-close\-0"); |
| 1636 |
if (close(sockPair[1]) == \-1) |
| 1637 |
err(EXIT_FAILURE, "closeSocketPair\-close\-1"); |
| 1638 |
} |
| 1639 |
\& |
| 1640 |
/* Implementation of the target process. Create a child process that: |
| 1641 |
\& |
| 1642 |
(1) installs a seccomp filter with the |
| 1643 |
SECCOMP_FILTER_FLAG_NEW_LISTENER flag; |
| 1644 |
(2) writes the seccomp notification file descriptor returned from |
| 1645 |
the previous step onto the UNIX domain socket, \[aq]sockPair[0]\[aq]; |
| 1646 |
(3) calls mkdir(2) for each element of \[aq]argv\[aq]. |
| 1647 |
\& |
| 1648 |
The function return value in the parent is the PID of the child |
| 1649 |
process; the child does not return from this function. */ |
| 1650 |
\& |
| 1651 |
static pid_t |
| 1652 |
targetProcess(int sockPair[2], char *argv[]) |
| 1653 |
{ |
| 1654 |
int notifyFd, s; |
| 1655 |
pid_t targetPid; |
| 1656 |
\& |
| 1657 |
targetPid = fork(); |
| 1658 |
\& |
| 1659 |
if (targetPid == \-1) |
| 1660 |
err(EXIT_FAILURE, "fork"); |
| 1661 |
\& |
| 1662 |
if (targetPid > 0) /* In parent, return PID of child */ |
| 1663 |
return targetPid; |
| 1664 |
\& |
| 1665 |
/* Child falls through to here */ |
| 1666 |
\& |
| 1667 |
printf("T: PID = %ld\[rs]n", (long) getpid()); |
| 1668 |
\& |
| 1669 |
/* Install seccomp filter(s) */ |
| 1670 |
\& |
| 1671 |
if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) |
| 1672 |
err(EXIT_FAILURE, "prctl"); |
| 1673 |
\& |
| 1674 |
notifyFd = installNotifyFilter(); |
| 1675 |
\& |
| 1676 |
/* Pass the notification file descriptor to the tracing process over |
| 1677 |
a UNIX domain socket */ |
| 1678 |
\& |
| 1679 |
if (sendfd(sockPair[0], notifyFd) == \-1) |
| 1680 |
err(EXIT_FAILURE, "sendfd"); |
| 1681 |
\& |
| 1682 |
/* Notification and socket FDs are no longer needed in target */ |
| 1683 |
\& |
| 1684 |
if (close(notifyFd) == \-1) |
| 1685 |
err(EXIT_FAILURE, "close\-target\-notify\-fd"); |
| 1686 |
\& |
| 1687 |
closeSocketPair(sockPair); |
| 1688 |
\& |
| 1689 |
/* Perform a mkdir() call for each of the command\-line arguments */ |
| 1690 |
\& |
| 1691 |
for (char **ap = argv; *ap != NULL; ap++) { |
| 1692 |
printf("\[rs]nT: about to mkdir(\[rs]"%s\[rs]")\[rs]n", *ap); |
| 1693 |
\& |
| 1694 |
s = mkdir(*ap, 0700); |
| 1695 |
if (s == \-1) |
| 1696 |
perror("T: ERROR: mkdir(2)"); |
| 1697 |
else |
| 1698 |
printf("T: SUCCESS: mkdir(2) returned %d\[rs]n", s); |
| 1699 |
} |
| 1700 |
\& |
| 1701 |
printf("\[rs]nT: terminating\[rs]n"); |
| 1702 |
exit(EXIT_SUCCESS); |
| 1703 |
} |
| 1704 |
\& |
| 1705 |
/* Check that the notification ID provided by a SECCOMP_IOCTL_NOTIF_RECV |
| 1706 |
operation is still valid. It will no longer be valid if the target |
| 1707 |
process has terminated or is no longer blocked in the system call that |
| 1708 |
generated the notification (because it was interrupted by a signal). |
| 1709 |
\& |
| 1710 |
This operation can be used when doing such things as accessing |
| 1711 |
/proc/PID files in the target process in order to avoid TOCTOU race |
| 1712 |
conditions where the PID that is returned by SECCOMP_IOCTL_NOTIF_RECV |
| 1713 |
terminates and is reused by another process. */ |
| 1714 |
\& |
| 1715 |
static bool |
| 1716 |
cookieIsValid(int notifyFd, uint64_t id) |
| 1717 |
{ |
| 1718 |
return ioctl(notifyFd, SECCOMP_IOCTL_NOTIF_ID_VALID, &id) == 0; |
| 1719 |
} |
| 1720 |
\& |
| 1721 |
/* Access the memory of the target process in order to fetch the |
| 1722 |
pathname referred to by the system call argument \[aq]argNum\[aq] in |
| 1723 |
\[aq]req\->data.args[]\[aq]. The pathname is returned in \[aq]path\[aq], |
| 1724 |
a buffer of \[aq]size\[aq] bytes allocated by the caller. |
| 1725 |
\& |
| 1726 |
Returns true if the pathname is successfully fetched, and false |
| 1727 |
otherwise. For possible causes of failure, see the comments below. */ |
| 1728 |
\& |
| 1729 |
static bool |
| 1730 |
getTargetPathname(struct seccomp_notif *req, int notifyFd, |
| 1731 |
int argNum, char *path, size_t size) |
| 1732 |
{ |
| 1733 |
int procMemFd; |
| 1734 |
char procMemPath[PATH_MAX]; |
| 1735 |
ssize_t nread; |
| 1736 |
\& |
| 1737 |
snprintf(procMemPath, sizeof(procMemPath), "/proc/%d/mem", req\->pid); |
| 1738 |
\& |
| 1739 |
procMemFd = open(procMemPath, O_RDONLY | O_CLOEXEC); |
| 1740 |
if (procMemFd == \-1) |
| 1741 |
return false; |
| 1742 |
\& |
| 1743 |
/* Check that the process whose info we are accessing is still alive |
| 1744 |
and blocked in the system call that caused the notification. |
| 1745 |
If the SECCOMP_IOCTL_NOTIF_ID_VALID operation (performed in |
| 1746 |
cookieIsValid()) succeeded, we know that the /proc/PID/mem file |
| 1747 |
descriptor that we opened corresponded to the process for which we |
| 1748 |
received a notification. If that process subsequently terminates, |
| 1749 |
then read() on that file descriptor will return 0 (EOF). */ |
| 1750 |
\& |
| 1751 |
if (!cookieIsValid(notifyFd, req\->id)) { |
| 1752 |
close(procMemFd); |
| 1753 |
return false; |
| 1754 |
} |
| 1755 |
\& |
| 1756 |
/* Read bytes at the location containing the pathname argument */ |
| 1757 |
\& |
| 1758 |
nread = pread(procMemFd, path, size, req\->data.args[argNum]); |
| 1759 |
\& |
| 1760 |
close(procMemFd); |
| 1761 |
\& |
| 1762 |
if (nread <= 0) |
| 1763 |
return false; |
| 1764 |
\& |
| 1765 |
/* Once again check that the notification ID is still valid. The |
| 1766 |
case we are particularly concerned about here is that just |
| 1767 |
before we fetched the pathname, the target\[aq]s blocked system |
| 1768 |
call was interrupted by a signal handler, and after the handler |
| 1769 |
returned, the target carried on execution (past the interrupted |
| 1770 |
system call). In that case, we have no guarantees about what we |
| 1771 |
are reading, since the target\[aq]s memory may have been arbitrarily |
| 1772 |
changed by subsequent operations. */ |
| 1773 |
\& |
| 1774 |
if (!cookieIsValid(notifyFd, req\->id)) { |
| 1775 |
perror("\[rs]tS: notification ID check failed!!!"); |
| 1776 |
return false; |
| 1777 |
} |
| 1778 |
\& |
| 1779 |
/* Even if the target\[aq]s system call was not interrupted by a signal, |
| 1780 |
we have no guarantees about what was in the memory of the target |
| 1781 |
process. (The memory may have been modified by another thread, or |
| 1782 |
even by an external attacking process.) We therefore treat the |
| 1783 |
buffer returned by pread() as untrusted input. The buffer should |
| 1784 |
contain a terminating null byte; if not, then we will trigger an |
| 1785 |
error for the target process. */ |
| 1786 |
\& |
| 1787 |
if (strnlen(path, nread) < nread) |
| 1788 |
return true; |
| 1789 |
\& |
| 1790 |
return false; |
| 1791 |
} |
| 1792 |
\& |
| 1793 |
/* Allocate buffers for the seccomp user\-space notification request and |
| 1794 |
response structures. It is the caller\[aq]s responsibility to free the |
| 1795 |
buffers returned via \[aq]req\[aq] and \[aq]resp\[aq]. */ |
| 1796 |
\& |
| 1797 |
static void |
| 1798 |
allocSeccompNotifBuffers(struct seccomp_notif **req, |
| 1799 |
struct seccomp_notif_resp **resp, |
| 1800 |
struct seccomp_notif_sizes *sizes) |
| 1801 |
{ |
| 1802 |
size_t resp_size; |
| 1803 |
\& |
| 1804 |
/* Discover the sizes of the structures that are used to receive |
| 1805 |
notifications and send notification responses, and allocate |
| 1806 |
buffers of those sizes. */ |
| 1807 |
\& |
| 1808 |
if (seccomp(SECCOMP_GET_NOTIF_SIZES, 0, sizes) == \-1) |
| 1809 |
err(EXIT_FAILURE, "seccomp\-SECCOMP_GET_NOTIF_SIZES"); |
| 1810 |
\& |
| 1811 |
*req = malloc(sizes\->seccomp_notif); |
| 1812 |
if (*req == NULL) |
| 1813 |
err(EXIT_FAILURE, "malloc\-seccomp_notif"); |
| 1814 |
\& |
| 1815 |
/* When allocating the response buffer, we must allow for the fact |
| 1816 |
that the user\-space binary may have been built with user\-space |
| 1817 |
headers where \[aq]struct seccomp_notif_resp\[aq] is bigger than the |
| 1818 |
response buffer expected by the (older) kernel. Therefore, we |
| 1819 |
allocate a buffer that is the maximum of the two sizes. This |
| 1820 |
ensures that if the supervisor places bytes into the response |
| 1821 |
structure that are past the response size that the kernel expects, |
| 1822 |
then the supervisor is not touching an invalid memory location. */ |
| 1823 |
\& |
| 1824 |
resp_size = sizes\->seccomp_notif_resp; |
| 1825 |
if (sizeof(struct seccomp_notif_resp) > resp_size) |
| 1826 |
resp_size = sizeof(struct seccomp_notif_resp); |
| 1827 |
\& |
| 1828 |
*resp = malloc(resp_size); |
| 1829 |
if (*resp == NULL) |
| 1830 |
err(EXIT_FAILURE, "malloc\-seccomp_notif_resp"); |
| 1831 |
\& |
| 1832 |
} |
| 1833 |
\& |
| 1834 |
/* Handle notifications that arrive via the SECCOMP_RET_USER_NOTIF file |
| 1835 |
descriptor, \[aq]notifyFd\[aq]. */ |
| 1836 |
\& |
| 1837 |
static void |
| 1838 |
handleNotifications(int notifyFd) |
| 1839 |
{ |
| 1840 |
bool pathOK; |
| 1841 |
char path[PATH_MAX]; |
| 1842 |
struct seccomp_notif *req; |
| 1843 |
struct seccomp_notif_resp *resp; |
| 1844 |
struct seccomp_notif_sizes sizes; |
| 1845 |
\& |
| 1846 |
allocSeccompNotifBuffers(&req, &resp, &sizes); |
| 1847 |
\& |
| 1848 |
/* Loop handling notifications */ |
| 1849 |
\& |
| 1850 |
for (;;) { |
| 1851 |
\& |
| 1852 |
/* Wait for next notification, returning info in \[aq]*req\[aq] */ |
| 1853 |
\& |
| 1854 |
memset(req, 0, sizes.seccomp_notif); |
| 1855 |
if (ioctl(notifyFd, SECCOMP_IOCTL_NOTIF_RECV, req) == \-1) { |
| 1856 |
if (errno == EINTR) |
| 1857 |
continue; |
| 1858 |
err(EXIT_FAILURE, "\[rs]tS: ioctl\-SECCOMP_IOCTL_NOTIF_RECV"); |
| 1859 |
} |
| 1860 |
\& |
| 1861 |
printf("\[rs]tS: got notification (ID %#llx) for PID %d\[rs]n", |
| 1862 |
req\->id, req\->pid); |
| 1863 |
\& |
| 1864 |
/* The only system call that can generate a notification event |
| 1865 |
is mkdir(2). Nevertheless, we check that the notified system |
| 1866 |
call is indeed mkdir() as kind of future\-proofing of this |
| 1867 |
code in case the seccomp filter is later modified to |
| 1868 |
generate notifications for other system calls. */ |
| 1869 |
\& |
| 1870 |
if (req\->data.nr != SYS_mkdir) { |
| 1871 |
printf("\[rs]tS: notification contained unexpected " |
| 1872 |
"system call number; bye!!!\[rs]n"); |
| 1873 |
exit(EXIT_FAILURE); |
| 1874 |
} |
| 1875 |
\& |
| 1876 |
pathOK = getTargetPathname(req, notifyFd, 0, path, sizeof(path)); |
| 1877 |
\& |
| 1878 |
/* Prepopulate some fields of the response */ |
| 1879 |
\& |
| 1880 |
resp\->id = req\->id; /* Response includes notification ID */ |
| 1881 |
resp\->flags = 0; |
| 1882 |
resp\->val = 0; |
| 1883 |
\& |
| 1884 |
/* If getTargetPathname() failed, trigger an EINVAL error |
| 1885 |
response (sending this response may yield an error if the |
| 1886 |
failure occurred because the notification ID was no longer |
| 1887 |
valid); if the directory is in /tmp, then create it on behalf |
| 1888 |
of the supervisor; if the pathname starts with \[aq].\[aq], tell the |
| 1889 |
kernel to let the target process execute the mkdir(); |
| 1890 |
otherwise, give an error for a directory pathname in any other |
| 1891 |
location. */ |
| 1892 |
\& |
| 1893 |
if (!pathOK) { |
| 1894 |
resp\->error = \-EINVAL; |
| 1895 |
printf("\[rs]tS: spoofing error for invalid pathname (%s)\[rs]n", |
| 1896 |
strerror(\-resp\->error)); |
| 1897 |
} else if (strncmp(path, "/tmp/", strlen("/tmp/")) == 0) { |
| 1898 |
printf("\[rs]tS: executing: mkdir(\[rs]"%s\[rs]", %#llo)\[rs]n", |
| 1899 |
path, req\->data.args[1]); |
| 1900 |
\& |
| 1901 |
if (mkdir(path, req\->data.args[1]) == 0) { |
| 1902 |
resp\->error = 0; /* "Success" */ |
| 1903 |
resp\->val = strlen(path); /* Used as return value of |
| 1904 |
mkdir() in target */ |
| 1905 |
printf("\[rs]tS: success! spoofed return = %lld\[rs]n", |
| 1906 |
resp\->val); |
| 1907 |
} else { |
| 1908 |
\& |
| 1909 |
/* If mkdir() failed in the supervisor, pass the error |
| 1910 |
back to the target */ |
| 1911 |
\& |
| 1912 |
resp\->error = \-errno; |
| 1913 |
printf("\[rs]tS: failure! (errno = %d; %s)\[rs]n", errno, |
| 1914 |
strerror(errno)); |
| 1915 |
} |
| 1916 |
} else if (strncmp(path, "./", strlen("./")) == 0) { |
| 1917 |
resp\->error = resp\->val = 0; |
| 1918 |
resp\->flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE; |
| 1919 |
printf("\[rs]tS: target can execute system call\[rs]n"); |
| 1920 |
} else { |
| 1921 |
resp\->error = \-EOPNOTSUPP; |
| 1922 |
printf("\[rs]tS: spoofing error response (%s)\[rs]n", |
| 1923 |
strerror(\-resp\->error)); |
| 1924 |
} |
| 1925 |
\& |
| 1926 |
/* Send a response to the notification */ |
| 1927 |
\& |
| 1928 |
printf("\[rs]tS: sending response " |
| 1929 |
"(flags = %#x; val = %lld; error = %d)\[rs]n", |
| 1930 |
resp\->flags, resp\->val, resp\->error); |
| 1931 |
\& |
| 1932 |
if (ioctl(notifyFd, SECCOMP_IOCTL_NOTIF_SEND, resp) == \-1) { |
| 1933 |
if (errno == ENOENT) |
| 1934 |
printf("\[rs]tS: response failed with ENOENT; " |
| 1935 |
"perhaps target process\[aq]s syscall was " |
| 1936 |
"interrupted by a signal?\[rs]n"); |
| 1937 |
else |
| 1938 |
perror("ioctl\-SECCOMP_IOCTL_NOTIF_SEND"); |
| 1939 |
} |
| 1940 |
\& |
| 1941 |
/* If the pathname is just "/bye", then the supervisor breaks out |
| 1942 |
of the loop and terminates. This allows us to see what happens |
| 1943 |
if the target process makes further calls to mkdir(2). */ |
| 1944 |
\& |
| 1945 |
if (streq(path, "/bye")) |
| 1946 |
break; |
| 1947 |
} |
| 1948 |
\& |
| 1949 |
free(req); |
| 1950 |
free(resp); |
| 1951 |
printf("\[rs]tS: terminating **********\[rs]n"); |
| 1952 |
exit(EXIT_FAILURE); |
| 1953 |
} |
| 1954 |
\& |
| 1955 |
/* Implementation of the supervisor process: |
| 1956 |
\& |
| 1957 |
(1) obtains the notification file descriptor from \[aq]sockPair[1]\[aq] |
| 1958 |
(2) handles notifications that arrive on that file descriptor. */ |
| 1959 |
\& |
| 1960 |
static void |
| 1961 |
supervisor(int sockPair[2]) |
| 1962 |
{ |
| 1963 |
int notifyFd; |
| 1964 |
\& |
| 1965 |
notifyFd = recvfd(sockPair[1]); |
| 1966 |
\& |
| 1967 |
if (notifyFd == \-1) |
| 1968 |
err(EXIT_FAILURE, "recvfd"); |
| 1969 |
\& |
| 1970 |
closeSocketPair(sockPair); /* We no longer need the socket pair */ |
| 1971 |
\& |
| 1972 |
handleNotifications(notifyFd); |
| 1973 |
} |
| 1974 |
\& |
| 1975 |
int |
| 1976 |
main(int argc, char *argv[]) |
| 1977 |
{ |
| 1978 |
int sockPair[2]; |
| 1979 |
struct sigaction sa; |
| 1980 |
\& |
| 1981 |
setbuf(stdout, NULL); |
| 1982 |
\& |
| 1983 |
if (argc < 2) { |
| 1984 |
fprintf(stderr, "At least one pathname argument is required\[rs]n"); |
| 1985 |
exit(EXIT_FAILURE); |
| 1986 |
} |
| 1987 |
\& |
| 1988 |
/* Create a UNIX domain socket that is used to pass the seccomp |
| 1989 |
notification file descriptor from the target process to the |
| 1990 |
supervisor process. */ |
| 1991 |
\& |
| 1992 |
if (socketpair(AF_UNIX, SOCK_STREAM, 0, sockPair) == \-1) |
| 1993 |
err(EXIT_FAILURE, "socketpair"); |
| 1994 |
\& |
| 1995 |
/* Create a child process\-\-the "target"\-\-that installs seccomp |
| 1996 |
filtering. The target process writes the seccomp notification |
| 1997 |
file descriptor onto \[aq]sockPair[0]\[aq] and then calls mkdir(2) for |
| 1998 |
each directory in the command\-line arguments. */ |
| 1999 |
\& |
| 2000 |
(void) targetProcess(sockPair, &argv[optind]); |
| 2001 |
\& |
| 2002 |
/* Catch SIGCHLD when the target terminates, so that the |
| 2003 |
supervisor can also terminate. */ |
| 2004 |
\& |
| 2005 |
sa.sa_handler = sigchldHandler; |
| 2006 |
sa.sa_flags = 0; |
| 2007 |
sigemptyset(&sa.sa_mask); |
| 2008 |
if (sigaction(SIGCHLD, &sa, NULL) == \-1) |
| 2009 |
err(EXIT_FAILURE, "sigaction"); |
| 2010 |
\& |
| 2011 |
supervisor(sockPair); |
| 2012 |
\& |
| 2013 |
exit(EXIT_SUCCESS); |
| 2014 |
} |
| 2015 |
.EE |
| 2016 |
.\" SRC END |
| 2017 |
.SH SEE ALSO |
| 2018 |
.BR ioctl (2), |
| 2019 |
.BR pidfd_getfd (2), |
| 2020 |
.BR pidfd_open (2), |
| 2021 |
.BR seccomp (2) |
| 2022 |
.P |
| 2023 |
A further example program can be found in the kernel source file |
| 2024 |
.IR samples/seccomp/user-trap.c . |