blob: bb0dd9ae699a4dd0b252f947de761a36a08194d1 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/kernel/seccomp.c
4 *
5 * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com>
6 *
Will Drewrye2cfabdf2012-04-12 16:47:57 -05007 * Copyright (C) 2012 Google, Inc.
8 * Will Drewry <wad@chromium.org>
9 *
10 * This defines a simple but solid secure-computing facility.
11 *
12 * Mode 1 uses a fixed list of allowed system calls.
13 * Mode 2 allows user-defined system call filters in the form
14 * of Berkeley Packet Filters/Linux Socket Filters.
Linus Torvalds1da177e2005-04-16 15:20:36 -070015 */
Kees Cooke68f9d42020-06-15 22:02:56 -070016#define pr_fmt(fmt) "seccomp: " fmt
Linus Torvalds1da177e2005-04-16 15:20:36 -070017
Kees Cook0b5fa222017-06-26 09:24:00 -070018#include <linux/refcount.h>
Eric Paris85e7bac32012-01-03 14:23:05 -050019#include <linux/audit.h>
Roland McGrath5b101742009-02-27 23:25:54 -080020#include <linux/compat.h>
Mike Frysingerb25e6712017-01-19 22:28:57 -060021#include <linux/coredump.h>
Tyler Hicks8e5f1ad2017-08-11 04:33:52 +000022#include <linux/kmemleak.h>
Kees Cook5c307082018-05-01 15:07:31 -070023#include <linux/nospec.h>
24#include <linux/prctl.h>
Will Drewrye2cfabdf2012-04-12 16:47:57 -050025#include <linux/sched.h>
Ingo Molnar68db0cf2017-02-08 18:51:37 +010026#include <linux/sched/task_stack.h>
Will Drewrye2cfabdf2012-04-12 16:47:57 -050027#include <linux/seccomp.h>
Kees Cookc8bee432014-06-27 15:16:33 -070028#include <linux/slab.h>
Kees Cook48dc92b2014-06-25 16:08:24 -070029#include <linux/syscalls.h>
Tyler Hicks8e5f1ad2017-08-11 04:33:52 +000030#include <linux/sysctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Andy Lutomirskia4412fc2014-07-21 18:49:14 -070032#ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
Will Drewrye2cfabdf2012-04-12 16:47:57 -050033#include <asm/syscall.h>
Andy Lutomirskia4412fc2014-07-21 18:49:14 -070034#endif
Will Drewrye2cfabdf2012-04-12 16:47:57 -050035
36#ifdef CONFIG_SECCOMP_FILTER
Tycho Andersen6a21cc52018-12-09 11:24:13 -070037#include <linux/file.h>
Will Drewrye2cfabdf2012-04-12 16:47:57 -050038#include <linux/filter.h>
Kees Cookc2e1f2e2014-06-05 00:23:17 -070039#include <linux/pid.h>
Will Drewryfb0fadf2012-04-12 16:48:02 -050040#include <linux/ptrace.h>
Will Drewrye2cfabdf2012-04-12 16:47:57 -050041#include <linux/security.h>
Will Drewrye2cfabdf2012-04-12 16:47:57 -050042#include <linux/tracehook.h>
43#include <linux/uaccess.h>
Tycho Andersen6a21cc52018-12-09 11:24:13 -070044#include <linux/anon_inodes.h>
Sargun Dhillon9f87dcf2020-06-01 04:25:32 -070045#include <linux/lockdep.h>
Tycho Andersen6a21cc52018-12-09 11:24:13 -070046
Kees Cook47e33c052020-06-15 15:42:46 -070047/*
48 * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
49 * wrong direction flag in the ioctl number. This is the broken one,
50 * which the kernel needs to keep supporting until all userspaces stop
51 * using the wrong command number.
52 */
53#define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR SECCOMP_IOR(2, __u64)
54
Tycho Andersen6a21cc52018-12-09 11:24:13 -070055enum notify_state {
56 SECCOMP_NOTIFY_INIT,
57 SECCOMP_NOTIFY_SENT,
58 SECCOMP_NOTIFY_REPLIED,
59};
60
61struct seccomp_knotif {
62 /* The struct pid of the task whose filter triggered the notification */
63 struct task_struct *task;
64
65 /* The "cookie" for this request; this is unique for this filter. */
66 u64 id;
67
68 /*
69 * The seccomp data. This pointer is valid the entire time this
70 * notification is active, since it comes from __seccomp_filter which
71 * eclipses the entire lifecycle here.
72 */
73 const struct seccomp_data *data;
74
75 /*
76 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
77 * struct seccomp_knotif is created and starts out in INIT. Once the
78 * handler reads the notification off of an FD, it transitions to SENT.
79 * If a signal is received the state transitions back to INIT and
80 * another message is sent. When the userspace handler replies, state
81 * transitions to REPLIED.
82 */
83 enum notify_state state;
84
85 /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
86 int error;
87 long val;
Christian Braunerfb3c53862019-09-20 10:30:05 +020088 u32 flags;
Tycho Andersen6a21cc52018-12-09 11:24:13 -070089
Sargun Dhillon7cf97b12020-06-02 18:10:43 -070090 /*
91 * Signals when this has changed states, such as the listener
92 * dying, a new seccomp addfd message, or changing to REPLIED
93 */
Tycho Andersen6a21cc52018-12-09 11:24:13 -070094 struct completion ready;
95
96 struct list_head list;
Sargun Dhillon7cf97b12020-06-02 18:10:43 -070097
98 /* outstanding addfd requests */
99 struct list_head addfd;
100};
101
102/**
103 * struct seccomp_kaddfd - container for seccomp_addfd ioctl messages
104 *
105 * @file: A reference to the file to install in the other task
106 * @fd: The fd number to install it at. If the fd number is -1, it means the
107 * installing process should allocate the fd as normal.
108 * @flags: The flags for the new file descriptor. At the moment, only O_CLOEXEC
109 * is allowed.
110 * @ret: The return value of the installing process. It is set to the fd num
111 * upon success (>= 0).
112 * @completion: Indicates that the installing process has completed fd
113 * installation, or gone away (either due to successful
114 * reply, or signal)
115 *
116 */
117struct seccomp_kaddfd {
118 struct file *file;
119 int fd;
120 unsigned int flags;
121
122 /* To only be set on reply */
123 int ret;
124 struct completion completion;
125 struct list_head list;
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700126};
127
128/**
129 * struct notification - container for seccomp userspace notifications. Since
130 * most seccomp filters will not have notification listeners attached and this
131 * structure is fairly large, we store the notification-specific stuff in a
132 * separate structure.
133 *
134 * @request: A semaphore that users of this notification can wait on for
135 * changes. Actual reads and writes are still controlled with
136 * filter->notify_lock.
137 * @next_id: The id of the next request.
138 * @notifications: A list of struct seccomp_knotif elements.
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700139 */
140struct notification {
141 struct semaphore request;
142 u64 next_id;
143 struct list_head notifications;
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700144};
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500145
146/**
147 * struct seccomp_filter - container for seccomp BPF programs
148 *
Christian Braunerb707dde2020-05-31 13:50:28 +0200149 * @refs: Reference count to manage the object lifetime.
150 * A filter's reference count is incremented for each directly
151 * attached task, once for the dependent filter, and if
152 * requested for the user notifier. When @refs reaches zero,
153 * the filter can be freed.
Christian Brauner99cdb8b2020-05-31 13:50:30 +0200154 * @users: A filter's @users count is incremented for each directly
155 * attached task (filter installation, fork(), thread_sync),
156 * and once for the dependent filter (tracked in filter->prev).
157 * When it reaches zero it indicates that no direct or indirect
158 * users of that filter exist. No new tasks can get associated with
159 * this filter after reaching 0. The @users count is always smaller
160 * or equal to @refs. Hence, reaching 0 for @users does not mean
161 * the filter can be freed.
Tyler Hickse66a3992017-08-11 04:33:56 +0000162 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500163 * @prev: points to a previously installed, or inherited, filter
Mickaël Salaün285fdfc2016-09-20 19:39:47 +0200164 * @prog: the BPF program to evaluate
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700165 * @notif: the struct that holds all notification related information
166 * @notify_lock: A lock for all notification-related accesses.
Christian Brauner76194c42020-06-01 11:50:07 -0700167 * @wqh: A wait queue for poll if a notifier is in use.
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500168 *
169 * seccomp_filter objects are organized in a tree linked via the @prev
170 * pointer. For any task, it appears to be a singly-linked list starting
171 * with current->seccomp.filter, the most recently attached or inherited filter.
172 * However, multiple filters may share a @prev node, by way of fork(), which
173 * results in a unidirectional tree existing in memory. This is similar to
174 * how namespaces work.
175 *
176 * seccomp_filter objects should never be modified after being attached
Christian Braunerb707dde2020-05-31 13:50:28 +0200177 * to a task_struct (other than @refs).
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500178 */
179struct seccomp_filter {
Christian Braunerb707dde2020-05-31 13:50:28 +0200180 refcount_t refs;
Christian Brauner99cdb8b2020-05-31 13:50:30 +0200181 refcount_t users;
Tyler Hickse66a3992017-08-11 04:33:56 +0000182 bool log;
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500183 struct seccomp_filter *prev;
Alexei Starovoitov7ae457c2014-07-30 20:34:16 -0700184 struct bpf_prog *prog;
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700185 struct notification *notif;
186 struct mutex notify_lock;
Christian Brauner76194c42020-06-01 11:50:07 -0700187 wait_queue_head_t wqh;
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500188};
189
190/* Limit any path through the tree to 256KB worth of instructions. */
191#define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
192
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100193/*
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500194 * Endianness is explicitly ignored and left for BPF program authors to manage
195 * as per the specific architecture.
196 */
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100197static void populate_seccomp_data(struct seccomp_data *sd)
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500198{
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100199 struct task_struct *task = current;
200 struct pt_regs *regs = task_pt_regs(task);
Daniel Borkmann2eac7642014-04-14 21:02:59 +0200201 unsigned long args[6];
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500202
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100203 sd->nr = syscall_get_nr(task, regs);
Dmitry V. Levin16add412019-03-18 02:30:18 +0300204 sd->arch = syscall_get_arch(task);
Steven Rostedt (Red Hat)b35f5492016-11-07 16:26:37 -0500205 syscall_get_arguments(task, regs, args);
Daniel Borkmann2eac7642014-04-14 21:02:59 +0200206 sd->args[0] = args[0];
207 sd->args[1] = args[1];
208 sd->args[2] = args[2];
209 sd->args[3] = args[3];
210 sd->args[4] = args[4];
211 sd->args[5] = args[5];
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100212 sd->instruction_pointer = KSTK_EIP(task);
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500213}
214
215/**
216 * seccomp_check_filter - verify seccomp filter code
217 * @filter: filter to verify
218 * @flen: length of filter
219 *
Alexei Starovoitov4df95ff2014-07-30 20:34:14 -0700220 * Takes a previously checked filter (by bpf_check_classic) and
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500221 * redirects all filter code that loads struct sk_buff data
222 * and related data through seccomp_bpf_load. It also
223 * enforces length and alignment checking of those loads.
224 *
225 * Returns 0 if the rule set is legal or -EINVAL if not.
226 */
227static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
228{
229 int pc;
230 for (pc = 0; pc < flen; pc++) {
231 struct sock_filter *ftest = &filter[pc];
232 u16 code = ftest->code;
233 u32 k = ftest->k;
234
235 switch (code) {
Daniel Borkmann34805932014-05-29 10:22:50 +0200236 case BPF_LD | BPF_W | BPF_ABS:
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100237 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500238 /* 32-bit aligned and not out of bounds. */
239 if (k >= sizeof(struct seccomp_data) || k & 3)
240 return -EINVAL;
241 continue;
Daniel Borkmann34805932014-05-29 10:22:50 +0200242 case BPF_LD | BPF_W | BPF_LEN:
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100243 ftest->code = BPF_LD | BPF_IMM;
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500244 ftest->k = sizeof(struct seccomp_data);
245 continue;
Daniel Borkmann34805932014-05-29 10:22:50 +0200246 case BPF_LDX | BPF_W | BPF_LEN:
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100247 ftest->code = BPF_LDX | BPF_IMM;
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500248 ftest->k = sizeof(struct seccomp_data);
249 continue;
250 /* Explicitly include allowed calls. */
Daniel Borkmann34805932014-05-29 10:22:50 +0200251 case BPF_RET | BPF_K:
252 case BPF_RET | BPF_A:
253 case BPF_ALU | BPF_ADD | BPF_K:
254 case BPF_ALU | BPF_ADD | BPF_X:
255 case BPF_ALU | BPF_SUB | BPF_K:
256 case BPF_ALU | BPF_SUB | BPF_X:
257 case BPF_ALU | BPF_MUL | BPF_K:
258 case BPF_ALU | BPF_MUL | BPF_X:
259 case BPF_ALU | BPF_DIV | BPF_K:
260 case BPF_ALU | BPF_DIV | BPF_X:
261 case BPF_ALU | BPF_AND | BPF_K:
262 case BPF_ALU | BPF_AND | BPF_X:
263 case BPF_ALU | BPF_OR | BPF_K:
264 case BPF_ALU | BPF_OR | BPF_X:
265 case BPF_ALU | BPF_XOR | BPF_K:
266 case BPF_ALU | BPF_XOR | BPF_X:
267 case BPF_ALU | BPF_LSH | BPF_K:
268 case BPF_ALU | BPF_LSH | BPF_X:
269 case BPF_ALU | BPF_RSH | BPF_K:
270 case BPF_ALU | BPF_RSH | BPF_X:
271 case BPF_ALU | BPF_NEG:
272 case BPF_LD | BPF_IMM:
273 case BPF_LDX | BPF_IMM:
274 case BPF_MISC | BPF_TAX:
275 case BPF_MISC | BPF_TXA:
276 case BPF_LD | BPF_MEM:
277 case BPF_LDX | BPF_MEM:
278 case BPF_ST:
279 case BPF_STX:
280 case BPF_JMP | BPF_JA:
281 case BPF_JMP | BPF_JEQ | BPF_K:
282 case BPF_JMP | BPF_JEQ | BPF_X:
283 case BPF_JMP | BPF_JGE | BPF_K:
284 case BPF_JMP | BPF_JGE | BPF_X:
285 case BPF_JMP | BPF_JGT | BPF_K:
286 case BPF_JMP | BPF_JGT | BPF_X:
287 case BPF_JMP | BPF_JSET | BPF_K:
288 case BPF_JMP | BPF_JSET | BPF_X:
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500289 continue;
290 default:
291 return -EINVAL;
292 }
293 }
294 return 0;
295}
296
297/**
Mickaël Salaün285fdfc2016-09-20 19:39:47 +0200298 * seccomp_run_filters - evaluates all seccomp filters against @sd
299 * @sd: optional seccomp data to be passed to filters
Kees Cookdeb4de82017-08-02 15:00:40 -0700300 * @match: stores struct seccomp_filter that resulted in the return value,
301 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
302 * be unchanged.
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500303 *
304 * Returns valid seccomp BPF response codes.
305 */
Kees Cook0466bdb2017-08-11 13:12:11 -0700306#define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
Kees Cookdeb4de82017-08-02 15:00:40 -0700307static u32 seccomp_run_filters(const struct seccomp_data *sd,
308 struct seccomp_filter **match)
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500309{
Will Drewryacf3b2c2012-04-12 16:47:59 -0500310 u32 ret = SECCOMP_RET_ALLOW;
Pranith Kumar8225d382014-11-21 10:06:01 -0500311 /* Make sure cross-thread synced filter points somewhere sane. */
312 struct seccomp_filter *f =
Will Deacon506458e2017-10-24 11:22:48 +0100313 READ_ONCE(current->seccomp.filter);
Will Drewryacf3b2c2012-04-12 16:47:59 -0500314
315 /* Ensure unexpected behavior doesn't result in failing open. */
Igor Stoppa0d42d732018-09-05 23:34:43 +0300316 if (WARN_ON(f == NULL))
Kees Cook4d3b0b02017-08-11 13:01:39 -0700317 return SECCOMP_RET_KILL_PROCESS;
Will Drewryacf3b2c2012-04-12 16:47:59 -0500318
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500319 /*
320 * All filters in the list are evaluated and the lowest BPF return
Will Drewryacf3b2c2012-04-12 16:47:59 -0500321 * value always takes priority (ignoring the DATA).
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500322 */
Kees Cook3ba25302014-06-27 15:01:35 -0700323 for (; f; f = f->prev) {
David Miller3d9f773c2020-02-24 15:01:43 +0100324 u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
Alexei Starovoitov8f577ca2014-05-13 19:50:47 -0700325
Kees Cook0466bdb2017-08-11 13:12:11 -0700326 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
Will Drewryacf3b2c2012-04-12 16:47:59 -0500327 ret = cur_ret;
Kees Cookdeb4de82017-08-02 15:00:40 -0700328 *match = f;
329 }
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500330 }
331 return ret;
332}
Kees Cook1f41b4502014-06-25 15:38:02 -0700333#endif /* CONFIG_SECCOMP_FILTER */
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500334
Kees Cook1f41b4502014-06-25 15:38:02 -0700335static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
336{
Guenter Roeck69f6a342014-08-10 20:50:30 -0700337 assert_spin_locked(&current->sighand->siglock);
Kees Cookdbd952122014-06-27 15:18:48 -0700338
Kees Cook1f41b4502014-06-25 15:38:02 -0700339 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
340 return false;
341
342 return true;
343}
344
Thomas Gleixner8bf37d82018-05-04 15:12:06 +0200345void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
Kees Cook5c307082018-05-01 15:07:31 -0700346
Kees Cook3ba25302014-06-27 15:01:35 -0700347static inline void seccomp_assign_mode(struct task_struct *task,
Kees Cook00a02d02018-05-03 14:56:12 -0700348 unsigned long seccomp_mode,
349 unsigned long flags)
Kees Cook1f41b4502014-06-25 15:38:02 -0700350{
Guenter Roeck69f6a342014-08-10 20:50:30 -0700351 assert_spin_locked(&task->sighand->siglock);
Kees Cookdbd952122014-06-27 15:18:48 -0700352
Kees Cook3ba25302014-06-27 15:01:35 -0700353 task->seccomp.mode = seccomp_mode;
354 /*
355 * Make sure TIF_SECCOMP cannot be set before the mode (and
356 * filter) is set.
357 */
358 smp_mb__before_atomic();
Kees Cook00a02d02018-05-03 14:56:12 -0700359 /* Assume default seccomp processes want spec flaw mitigation. */
360 if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
Thomas Gleixner8bf37d82018-05-04 15:12:06 +0200361 arch_seccomp_spec_mitigate(task);
Kees Cook3ba25302014-06-27 15:01:35 -0700362 set_tsk_thread_flag(task, TIF_SECCOMP);
Kees Cook1f41b4502014-06-25 15:38:02 -0700363}
364
365#ifdef CONFIG_SECCOMP_FILTER
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700366/* Returns 1 if the parent is an ancestor of the child. */
367static int is_ancestor(struct seccomp_filter *parent,
368 struct seccomp_filter *child)
369{
370 /* NULL is the root ancestor. */
371 if (parent == NULL)
372 return 1;
373 for (; child; child = child->prev)
374 if (child == parent)
375 return 1;
376 return 0;
377}
378
379/**
380 * seccomp_can_sync_threads: checks if all threads can be synchronized
381 *
382 * Expects sighand and cred_guard_mutex locks to be held.
383 *
384 * Returns 0 on success, -ve on error, or the pid of a thread which was
Tycho Andersen6beff002019-03-06 13:14:12 -0700385 * either not in the correct seccomp mode or did not have an ancestral
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700386 * seccomp filter.
387 */
388static inline pid_t seccomp_can_sync_threads(void)
389{
390 struct task_struct *thread, *caller;
391
392 BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
Guenter Roeck69f6a342014-08-10 20:50:30 -0700393 assert_spin_locked(&current->sighand->siglock);
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700394
395 /* Validate all threads being eligible for synchronization. */
396 caller = current;
397 for_each_thread(caller, thread) {
398 pid_t failed;
399
400 /* Skip current, since it is initiating the sync. */
401 if (thread == caller)
402 continue;
403
404 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
405 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
406 is_ancestor(thread->seccomp.filter,
407 caller->seccomp.filter)))
408 continue;
409
410 /* Return the first thread that cannot be synchronized. */
411 failed = task_pid_vnr(thread);
412 /* If the pid cannot be resolved, then return -ESRCH */
Igor Stoppa0d42d732018-09-05 23:34:43 +0300413 if (WARN_ON(failed == 0))
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700414 failed = -ESRCH;
415 return failed;
416 }
417
418 return 0;
419}
420
Christian Brauner3a15fb62020-05-31 13:50:29 +0200421static inline void seccomp_filter_free(struct seccomp_filter *filter)
422{
423 if (filter) {
424 bpf_prog_destroy(filter->prog);
425 kfree(filter);
426 }
427}
428
Christian Brauner99cdb8b2020-05-31 13:50:30 +0200429static void __seccomp_filter_orphan(struct seccomp_filter *orig)
430{
431 while (orig && refcount_dec_and_test(&orig->users)) {
432 if (waitqueue_active(&orig->wqh))
433 wake_up_poll(&orig->wqh, EPOLLHUP);
434 orig = orig->prev;
435 }
436}
437
Christian Brauner3a15fb62020-05-31 13:50:29 +0200438static void __put_seccomp_filter(struct seccomp_filter *orig)
439{
440 /* Clean up single-reference branches iteratively. */
441 while (orig && refcount_dec_and_test(&orig->refs)) {
442 struct seccomp_filter *freeme = orig;
443 orig = orig->prev;
444 seccomp_filter_free(freeme);
445 }
446}
447
Christian Brauner99cdb8b2020-05-31 13:50:30 +0200448static void __seccomp_filter_release(struct seccomp_filter *orig)
449{
450 /* Notify about any unused filters in the task's former filter tree. */
451 __seccomp_filter_orphan(orig);
452 /* Finally drop all references to the task's former tree. */
453 __put_seccomp_filter(orig);
454}
455
Christian Brauner3a15fb62020-05-31 13:50:29 +0200456/**
Christian Brauner99cdb8b2020-05-31 13:50:30 +0200457 * seccomp_filter_release - Detach the task from its filter tree,
458 * drop its reference count, and notify
459 * about unused filters
Christian Brauner3a15fb62020-05-31 13:50:29 +0200460 *
461 * This function should only be called when the task is exiting as
462 * it detaches it from its filter tree. As such, READ_ONCE() and
463 * barriers are not needed here, as would normally be needed.
464 */
465void seccomp_filter_release(struct task_struct *tsk)
466{
467 struct seccomp_filter *orig = tsk->seccomp.filter;
468
469 /* Detach task from its filter tree. */
470 tsk->seccomp.filter = NULL;
Christian Brauner99cdb8b2020-05-31 13:50:30 +0200471 __seccomp_filter_release(orig);
Christian Brauner3a15fb62020-05-31 13:50:29 +0200472}
473
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700474/**
475 * seccomp_sync_threads: sets all threads to use current's filter
476 *
477 * Expects sighand and cred_guard_mutex locks to be held, and for
478 * seccomp_can_sync_threads() to have returned success already
479 * without dropping the locks.
480 *
481 */
Kees Cook00a02d02018-05-03 14:56:12 -0700482static inline void seccomp_sync_threads(unsigned long flags)
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700483{
484 struct task_struct *thread, *caller;
485
486 BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
Guenter Roeck69f6a342014-08-10 20:50:30 -0700487 assert_spin_locked(&current->sighand->siglock);
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700488
489 /* Synchronize all threads. */
490 caller = current;
491 for_each_thread(caller, thread) {
492 /* Skip current, since it needs no changes. */
493 if (thread == caller)
494 continue;
495
496 /* Get a task reference for the new leaf node. */
497 get_seccomp_filter(caller);
Christian Brauner99cdb8b2020-05-31 13:50:30 +0200498
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700499 /*
500 * Drop the task reference to the shared ancestor since
501 * current's path will hold a reference. (This also
502 * allows a put before the assignment.)
503 */
Christian Brauner99cdb8b2020-05-31 13:50:30 +0200504 __seccomp_filter_release(thread->seccomp.filter);
505
506 /* Make our new filter tree visible. */
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700507 smp_store_release(&thread->seccomp.filter,
508 caller->seccomp.filter);
Kees Cookc818c032020-05-13 14:11:26 -0700509 atomic_set(&thread->seccomp.filter_count,
510 atomic_read(&thread->seccomp.filter_count));
Jann Horn103502a2015-12-26 06:00:48 +0100511
512 /*
513 * Don't let an unprivileged task work around
514 * the no_new_privs restriction by creating
515 * a thread that sets it up, enters seccomp,
516 * then dies.
517 */
518 if (task_no_new_privs(caller))
519 task_set_no_new_privs(thread);
520
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700521 /*
522 * Opt the other thread into seccomp if needed.
523 * As threads are considered to be trust-realm
524 * equivalent (see ptrace_may_access), it is safe to
525 * allow one thread to transition the other.
526 */
Jann Horn103502a2015-12-26 06:00:48 +0100527 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
Kees Cook00a02d02018-05-03 14:56:12 -0700528 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
529 flags);
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700530 }
531}
532
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500533/**
Kees Cookc8bee432014-06-27 15:16:33 -0700534 * seccomp_prepare_filter: Prepares a seccomp filter for use.
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500535 * @fprog: BPF program to install
536 *
Kees Cookc8bee432014-06-27 15:16:33 -0700537 * Returns filter on success or an ERR_PTR on failure.
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500538 */
Kees Cookc8bee432014-06-27 15:16:33 -0700539static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500540{
Daniel Borkmannac67eb22015-05-06 16:12:30 +0200541 struct seccomp_filter *sfilter;
542 int ret;
Masahiro Yamada97f26452016-08-03 13:45:50 -0700543 const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500544
545 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
Kees Cookc8bee432014-06-27 15:16:33 -0700546 return ERR_PTR(-EINVAL);
Nicolas Schichand9e12f42e2015-05-06 16:12:28 +0200547
Kees Cookc8bee432014-06-27 15:16:33 -0700548 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500549
550 /*
Fabian Frederick119ce5c2014-06-06 14:37:53 -0700551 * Installing a seccomp filter requires that the task has
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500552 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
553 * This avoids scenarios where unprivileged tasks can affect the
554 * behavior of privileged children.
555 */
Kees Cook1d4457f2014-05-21 15:23:46 -0700556 if (!task_no_new_privs(current) &&
Micah Mortonc1a85a02019-01-07 16:10:53 -0800557 security_capable(current_cred(), current_user_ns(),
558 CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) != 0)
Kees Cookc8bee432014-06-27 15:16:33 -0700559 return ERR_PTR(-EACCES);
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500560
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100561 /* Allocate a new seccomp_filter */
Daniel Borkmannac67eb22015-05-06 16:12:30 +0200562 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
563 if (!sfilter)
Nicolas Schichand9e12f42e2015-05-06 16:12:28 +0200564 return ERR_PTR(-ENOMEM);
Daniel Borkmannac67eb22015-05-06 16:12:30 +0200565
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700566 mutex_init(&sfilter->notify_lock);
Daniel Borkmannac67eb22015-05-06 16:12:30 +0200567 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
Tycho Andersenf8e529ed2015-10-27 09:23:59 +0900568 seccomp_check_filter, save_orig);
Daniel Borkmannac67eb22015-05-06 16:12:30 +0200569 if (ret < 0) {
570 kfree(sfilter);
571 return ERR_PTR(ret);
Nicolas Schichand9e12f42e2015-05-06 16:12:28 +0200572 }
Alexei Starovoitovbd4cf0e2014-03-28 18:58:25 +0100573
Christian Braunerb707dde2020-05-31 13:50:28 +0200574 refcount_set(&sfilter->refs, 1);
Christian Brauner99cdb8b2020-05-31 13:50:30 +0200575 refcount_set(&sfilter->users, 1);
Christian Brauner76194c42020-06-01 11:50:07 -0700576 init_waitqueue_head(&sfilter->wqh);
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500577
Daniel Borkmannac67eb22015-05-06 16:12:30 +0200578 return sfilter;
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500579}
580
581/**
Kees Cookc8bee432014-06-27 15:16:33 -0700582 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500583 * @user_filter: pointer to the user data containing a sock_fprog.
584 *
585 * Returns 0 on success and non-zero otherwise.
586 */
Kees Cookc8bee432014-06-27 15:16:33 -0700587static struct seccomp_filter *
588seccomp_prepare_user_filter(const char __user *user_filter)
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500589{
590 struct sock_fprog fprog;
Kees Cookc8bee432014-06-27 15:16:33 -0700591 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500592
593#ifdef CONFIG_COMPAT
Andy Lutomirski5c380652016-03-22 14:24:52 -0700594 if (in_compat_syscall()) {
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500595 struct compat_sock_fprog fprog32;
596 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
597 goto out;
598 fprog.len = fprog32.len;
599 fprog.filter = compat_ptr(fprog32.filter);
600 } else /* falls through to the if below. */
601#endif
602 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
603 goto out;
Kees Cookc8bee432014-06-27 15:16:33 -0700604 filter = seccomp_prepare_filter(&fprog);
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500605out:
Kees Cookc8bee432014-06-27 15:16:33 -0700606 return filter;
607}
608
609/**
610 * seccomp_attach_filter: validate and attach filter
611 * @flags: flags to change filter behavior
612 * @filter: seccomp filter to add to the current process
613 *
Kees Cookdbd952122014-06-27 15:18:48 -0700614 * Caller must be holding current->sighand->siglock lock.
615 *
Tycho Andersen7a0df7f2019-03-06 13:14:13 -0700616 * Returns 0 on success, -ve on error, or
617 * - in TSYNC mode: the pid of a thread which was either not in the correct
618 * seccomp mode or did not have an ancestral seccomp filter
619 * - in NEW_LISTENER mode: the fd of the new listener
Kees Cookc8bee432014-06-27 15:16:33 -0700620 */
621static long seccomp_attach_filter(unsigned int flags,
622 struct seccomp_filter *filter)
623{
624 unsigned long total_insns;
625 struct seccomp_filter *walker;
626
Guenter Roeck69f6a342014-08-10 20:50:30 -0700627 assert_spin_locked(&current->sighand->siglock);
Kees Cookdbd952122014-06-27 15:18:48 -0700628
Kees Cookc8bee432014-06-27 15:16:33 -0700629 /* Validate resulting filter length. */
630 total_insns = filter->prog->len;
631 for (walker = current->seccomp.filter; walker; walker = walker->prev)
632 total_insns += walker->prog->len + 4; /* 4 instr penalty */
633 if (total_insns > MAX_INSNS_PER_PATH)
634 return -ENOMEM;
635
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700636 /* If thread sync has been requested, check that it is possible. */
637 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
638 int ret;
639
640 ret = seccomp_can_sync_threads();
Tycho Andersen51891492020-03-04 11:05:17 -0700641 if (ret) {
642 if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
643 return -ESRCH;
644 else
645 return ret;
646 }
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700647 }
648
Tyler Hickse66a3992017-08-11 04:33:56 +0000649 /* Set log flag, if present. */
650 if (flags & SECCOMP_FILTER_FLAG_LOG)
651 filter->log = true;
652
Kees Cookc8bee432014-06-27 15:16:33 -0700653 /*
654 * If there is an existing filter, make it the prev and don't drop its
655 * task reference.
656 */
657 filter->prev = current->seccomp.filter;
658 current->seccomp.filter = filter;
Kees Cookc818c032020-05-13 14:11:26 -0700659 atomic_inc(&current->seccomp.filter_count);
Kees Cookc8bee432014-06-27 15:16:33 -0700660
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700661 /* Now that the new filter is in place, synchronize to all threads. */
662 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
Kees Cook00a02d02018-05-03 14:56:12 -0700663 seccomp_sync_threads(flags);
Kees Cookc2e1f2e2014-06-05 00:23:17 -0700664
Kees Cookc8bee432014-06-27 15:16:33 -0700665 return 0;
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500666}
667
Colin Ian King084f5602017-09-29 14:26:48 +0100668static void __get_seccomp_filter(struct seccomp_filter *filter)
Oleg Nesterov66a733e2017-09-27 09:25:30 -0600669{
Christian Braunerb707dde2020-05-31 13:50:28 +0200670 refcount_inc(&filter->refs);
Oleg Nesterov66a733e2017-09-27 09:25:30 -0600671}
672
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500673/* get_seccomp_filter - increments the reference count of the filter on @tsk */
674void get_seccomp_filter(struct task_struct *tsk)
675{
676 struct seccomp_filter *orig = tsk->seccomp.filter;
677 if (!orig)
678 return;
Oleg Nesterov66a733e2017-09-27 09:25:30 -0600679 __get_seccomp_filter(orig);
Christian Brauner99cdb8b2020-05-31 13:50:30 +0200680 refcount_inc(&orig->users);
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500681}
682
Eric W. Biedermanae7795b2018-09-25 11:27:20 +0200683static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason)
Mike Frysingerb25e6712017-01-19 22:28:57 -0600684{
Eric W. Biederman3b10db2b2017-08-18 19:56:27 -0500685 clear_siginfo(info);
Mike Frysingerb25e6712017-01-19 22:28:57 -0600686 info->si_signo = SIGSYS;
687 info->si_code = SYS_SECCOMP;
688 info->si_call_addr = (void __user *)KSTK_EIP(current);
689 info->si_errno = reason;
Dmitry V. Levin16add412019-03-18 02:30:18 +0300690 info->si_arch = syscall_get_arch(current);
Mike Frysingerb25e6712017-01-19 22:28:57 -0600691 info->si_syscall = syscall;
692}
693
Will Drewrybb6ea432012-04-12 16:48:01 -0500694/**
695 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
696 * @syscall: syscall number to send to userland
697 * @reason: filter-supplied reason code to send to userland (via si_errno)
698 *
699 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
700 */
701static void seccomp_send_sigsys(int syscall, int reason)
702{
Eric W. Biedermanae7795b2018-09-25 11:27:20 +0200703 struct kernel_siginfo info;
Mike Frysingerb25e6712017-01-19 22:28:57 -0600704 seccomp_init_siginfo(&info, syscall, reason);
Eric W. Biedermana89e9b82019-05-15 10:11:09 -0500705 force_sig_info(&info);
Will Drewrybb6ea432012-04-12 16:48:01 -0500706}
Will Drewrye2cfabdf2012-04-12 16:47:57 -0500707#endif /* CONFIG_SECCOMP_FILTER */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000709/* For use with seccomp_actions_logged */
Kees Cook4d3b0b02017-08-11 13:01:39 -0700710#define SECCOMP_LOG_KILL_PROCESS (1 << 0)
711#define SECCOMP_LOG_KILL_THREAD (1 << 1)
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000712#define SECCOMP_LOG_TRAP (1 << 2)
713#define SECCOMP_LOG_ERRNO (1 << 3)
714#define SECCOMP_LOG_TRACE (1 << 4)
Tyler Hicks59f5cf42017-08-11 04:33:57 +0000715#define SECCOMP_LOG_LOG (1 << 5)
716#define SECCOMP_LOG_ALLOW (1 << 6)
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700717#define SECCOMP_LOG_USER_NOTIF (1 << 7)
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000718
Kees Cook4d3b0b02017-08-11 13:01:39 -0700719static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
720 SECCOMP_LOG_KILL_THREAD |
Kees Cookfd768752017-08-11 12:53:18 -0700721 SECCOMP_LOG_TRAP |
722 SECCOMP_LOG_ERRNO |
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700723 SECCOMP_LOG_USER_NOTIF |
Kees Cookfd768752017-08-11 12:53:18 -0700724 SECCOMP_LOG_TRACE |
Tyler Hicks59f5cf42017-08-11 04:33:57 +0000725 SECCOMP_LOG_LOG;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000726
Tyler Hickse66a3992017-08-11 04:33:56 +0000727static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
728 bool requested)
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000729{
730 bool log = false;
731
732 switch (action) {
733 case SECCOMP_RET_ALLOW:
Tyler Hickse66a3992017-08-11 04:33:56 +0000734 break;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000735 case SECCOMP_RET_TRAP:
Tyler Hickse66a3992017-08-11 04:33:56 +0000736 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
737 break;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000738 case SECCOMP_RET_ERRNO:
Tyler Hickse66a3992017-08-11 04:33:56 +0000739 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
740 break;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000741 case SECCOMP_RET_TRACE:
Tyler Hickse66a3992017-08-11 04:33:56 +0000742 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000743 break;
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700744 case SECCOMP_RET_USER_NOTIF:
745 log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
746 break;
Tyler Hicks59f5cf42017-08-11 04:33:57 +0000747 case SECCOMP_RET_LOG:
748 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
749 break;
Kees Cookfd768752017-08-11 12:53:18 -0700750 case SECCOMP_RET_KILL_THREAD:
Kees Cookfd768752017-08-11 12:53:18 -0700751 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
Kees Cook4d3b0b02017-08-11 13:01:39 -0700752 break;
753 case SECCOMP_RET_KILL_PROCESS:
754 default:
755 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000756 }
757
758 /*
Tyler Hicks326bee02018-05-04 01:08:15 +0000759 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
760 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
761 * any action from being logged by removing the action name from the
762 * seccomp_actions_logged sysctl.
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000763 */
Tyler Hicks326bee02018-05-04 01:08:15 +0000764 if (!log)
765 return;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000766
Tyler Hicks326bee02018-05-04 01:08:15 +0000767 audit_seccomp(syscall, signr, action);
Tyler Hicks0ddec0f2017-08-11 04:33:54 +0000768}
769
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770/*
771 * Secure computing mode 1 allows only read/write/exit/sigreturn.
772 * To be fully secure this must be combined with rlimit
773 * to limit the stack allocations too.
774 */
Matt Redfearncb4253a2016-03-29 09:35:34 +0100775static const int mode1_syscalls[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
Kees Cookfe4bfff82020-06-19 12:20:15 -0700777 -1, /* negative terminated */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778};
779
Andy Lutomirskia4412fc2014-07-21 18:49:14 -0700780static void __secure_computing_strict(int this_syscall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781{
Kees Cookfe4bfff82020-06-19 12:20:15 -0700782 const int *allowed_syscalls = mode1_syscalls;
Andy Lutomirskia4412fc2014-07-21 18:49:14 -0700783#ifdef CONFIG_COMPAT
Andy Lutomirski5c380652016-03-22 14:24:52 -0700784 if (in_compat_syscall())
Kees Cookfe4bfff82020-06-19 12:20:15 -0700785 allowed_syscalls = get_compat_mode1_syscalls();
Andy Lutomirskia4412fc2014-07-21 18:49:14 -0700786#endif
787 do {
Kees Cookfe4bfff82020-06-19 12:20:15 -0700788 if (*allowed_syscalls == this_syscall)
Andy Lutomirskia4412fc2014-07-21 18:49:14 -0700789 return;
Kees Cookfe4bfff82020-06-19 12:20:15 -0700790 } while (*++allowed_syscalls != -1);
Andy Lutomirskia4412fc2014-07-21 18:49:14 -0700791
792#ifdef SECCOMP_DEBUG
793 dump_stack();
794#endif
Kees Cookfd768752017-08-11 12:53:18 -0700795 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
Andy Lutomirskia4412fc2014-07-21 18:49:14 -0700796 do_exit(SIGKILL);
797}
798
799#ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
800void secure_computing_strict(int this_syscall)
801{
802 int mode = current->seccomp.mode;
803
Masahiro Yamada97f26452016-08-03 13:45:50 -0700804 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
Tycho Andersen13c4a902015-06-13 09:02:48 -0600805 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
806 return;
807
Kees Cook221272f2015-06-15 15:29:16 -0700808 if (mode == SECCOMP_MODE_DISABLED)
Andy Lutomirskia4412fc2014-07-21 18:49:14 -0700809 return;
810 else if (mode == SECCOMP_MODE_STRICT)
811 __secure_computing_strict(this_syscall);
812 else
813 BUG();
814}
815#else
Andy Lutomirski13aa72f2014-07-21 18:49:15 -0700816
817#ifdef CONFIG_SECCOMP_FILTER
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700818static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
819{
820 /*
821 * Note: overflow is ok here, the id just needs to be unique per
822 * filter.
823 */
824 lockdep_assert_held(&filter->notify_lock);
825 return filter->notif->next_id++;
826}
827
Sargun Dhillon7cf97b12020-06-02 18:10:43 -0700828static void seccomp_handle_addfd(struct seccomp_kaddfd *addfd)
829{
830 /*
831 * Remove the notification, and reset the list pointers, indicating
832 * that it has been handled.
833 */
834 list_del_init(&addfd->list);
835 addfd->ret = receive_fd_replace(addfd->fd, addfd->file, addfd->flags);
836 complete(&addfd->completion);
837}
838
Christian Braunerfb3c53862019-09-20 10:30:05 +0200839static int seccomp_do_user_notification(int this_syscall,
840 struct seccomp_filter *match,
841 const struct seccomp_data *sd)
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700842{
843 int err;
Christian Braunerfb3c53862019-09-20 10:30:05 +0200844 u32 flags = 0;
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700845 long ret = 0;
846 struct seccomp_knotif n = {};
Sargun Dhillon7cf97b12020-06-02 18:10:43 -0700847 struct seccomp_kaddfd *addfd, *tmp;
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700848
849 mutex_lock(&match->notify_lock);
850 err = -ENOSYS;
851 if (!match->notif)
852 goto out;
853
854 n.task = current;
855 n.state = SECCOMP_NOTIFY_INIT;
856 n.data = sd;
857 n.id = seccomp_next_notify_id(match);
858 init_completion(&n.ready);
859 list_add(&n.list, &match->notif->notifications);
Sargun Dhillon7cf97b12020-06-02 18:10:43 -0700860 INIT_LIST_HEAD(&n.addfd);
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700861
862 up(&match->notif->request);
Christian Brauner76194c42020-06-01 11:50:07 -0700863 wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM);
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700864 mutex_unlock(&match->notify_lock);
865
866 /*
867 * This is where we wait for a reply from userspace.
868 */
Sargun Dhillon7cf97b12020-06-02 18:10:43 -0700869wait:
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700870 err = wait_for_completion_interruptible(&n.ready);
871 mutex_lock(&match->notify_lock);
872 if (err == 0) {
Sargun Dhillon7cf97b12020-06-02 18:10:43 -0700873 /* Check if we were woken up by a addfd message */
874 addfd = list_first_entry_or_null(&n.addfd,
875 struct seccomp_kaddfd, list);
876 if (addfd && n.state != SECCOMP_NOTIFY_REPLIED) {
877 seccomp_handle_addfd(addfd);
878 mutex_unlock(&match->notify_lock);
879 goto wait;
880 }
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700881 ret = n.val;
882 err = n.error;
Christian Braunerfb3c53862019-09-20 10:30:05 +0200883 flags = n.flags;
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700884 }
885
Sargun Dhillon7cf97b12020-06-02 18:10:43 -0700886 /* If there were any pending addfd calls, clear them out */
887 list_for_each_entry_safe(addfd, tmp, &n.addfd, list) {
888 /* The process went away before we got a chance to handle it */
889 addfd->ret = -ESRCH;
890 list_del_init(&addfd->list);
891 complete(&addfd->completion);
892 }
893
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700894 /*
895 * Note that it's possible the listener died in between the time when
Sargun Dhillon7cf97b12020-06-02 18:10:43 -0700896 * we were notified of a response (or a signal) and when we were able to
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700897 * re-acquire the lock, so only delete from the list if the
898 * notification actually exists.
899 *
900 * Also note that this test is only valid because there's no way to
901 * *reattach* to a notifier right now. If one is added, we'll need to
902 * keep track of the notif itself and make sure they match here.
903 */
904 if (match->notif)
905 list_del(&n.list);
906out:
907 mutex_unlock(&match->notify_lock);
Christian Braunerfb3c53862019-09-20 10:30:05 +0200908
909 /* Userspace requests to continue the syscall. */
910 if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
911 return 0;
912
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700913 syscall_set_return_value(current, task_pt_regs(current),
914 err, ret);
Christian Braunerfb3c53862019-09-20 10:30:05 +0200915 return -1;
Tycho Andersen6a21cc52018-12-09 11:24:13 -0700916}
917
Kees Cookce6526e2016-06-01 19:29:15 -0700918static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
919 const bool recheck_after_trace)
Andy Lutomirski13aa72f2014-07-21 18:49:15 -0700920{
921 u32 filter_ret, action;
Kees Cookdeb4de82017-08-02 15:00:40 -0700922 struct seccomp_filter *match = NULL;
Andy Lutomirski13aa72f2014-07-21 18:49:15 -0700923 int data;
Tycho Andersendb511392018-12-09 11:24:11 -0700924 struct seccomp_data sd_local;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
Kees Cook3ba25302014-06-27 15:01:35 -0700926 /*
927 * Make sure that any changes to mode from another thread have
928 * been seen after TIF_SECCOMP was seen.
929 */
930 rmb();
931
Tycho Andersendb511392018-12-09 11:24:11 -0700932 if (!sd) {
933 populate_seccomp_data(&sd_local);
934 sd = &sd_local;
935 }
936
Kees Cookdeb4de82017-08-02 15:00:40 -0700937 filter_ret = seccomp_run_filters(sd, &match);
Andy Lutomirski13aa72f2014-07-21 18:49:15 -0700938 data = filter_ret & SECCOMP_RET_DATA;
Kees Cook0466bdb2017-08-11 13:12:11 -0700939 action = filter_ret & SECCOMP_RET_ACTION_FULL;
Andy Lutomirski87b526d2012-10-01 11:40:45 -0700940
Andy Lutomirski13aa72f2014-07-21 18:49:15 -0700941 switch (action) {
942 case SECCOMP_RET_ERRNO:
Kees Cook580c57f2015-02-17 13:48:00 -0800943 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
944 if (data > MAX_ERRNO)
945 data = MAX_ERRNO;
Andy Lutomirskid39bd002014-07-21 18:49:16 -0700946 syscall_set_return_value(current, task_pt_regs(current),
Andy Lutomirski13aa72f2014-07-21 18:49:15 -0700947 -data, 0);
948 goto skip;
949
950 case SECCOMP_RET_TRAP:
951 /* Show the handler the original registers. */
Andy Lutomirskid39bd002014-07-21 18:49:16 -0700952 syscall_rollback(current, task_pt_regs(current));
Andy Lutomirski13aa72f2014-07-21 18:49:15 -0700953 /* Let the filter pass back 16 bits of data. */
954 seccomp_send_sigsys(this_syscall, data);
955 goto skip;
956
957 case SECCOMP_RET_TRACE:
Kees Cookce6526e2016-06-01 19:29:15 -0700958 /* We've been put in this state by the ptracer already. */
959 if (recheck_after_trace)
960 return 0;
961
Kees Cook8112c4f2016-06-01 16:02:17 -0700962 /* ENOSYS these calls if there is no tracer attached. */
963 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
964 syscall_set_return_value(current,
965 task_pt_regs(current),
966 -ENOSYS, 0);
967 goto skip;
968 }
969
970 /* Allow the BPF to provide the event message */
971 ptrace_event(PTRACE_EVENT_SECCOMP, data);
972 /*
973 * The delivery of a fatal signal during event
Kees Cook485a2522016-08-10 16:28:09 -0700974 * notification may silently skip tracer notification,
975 * which could leave us with a potentially unmodified
976 * syscall that the tracer would have liked to have
977 * changed. Since the process is about to die, we just
978 * force the syscall to be skipped and let the signal
979 * kill the process and correctly handle any tracer exit
980 * notifications.
Kees Cook8112c4f2016-06-01 16:02:17 -0700981 */
982 if (fatal_signal_pending(current))
Kees Cook485a2522016-08-10 16:28:09 -0700983 goto skip;
Kees Cook8112c4f2016-06-01 16:02:17 -0700984 /* Check if the tracer forced the syscall to be skipped. */
985 this_syscall = syscall_get_nr(current, task_pt_regs(current));
986 if (this_syscall < 0)
987 goto skip;
988
Kees Cookce6526e2016-06-01 19:29:15 -0700989 /*
990 * Recheck the syscall, since it may have changed. This
991 * intentionally uses a NULL struct seccomp_data to force
992 * a reload of all registers. This does not goto skip since
993 * a skip would have already been reported.
994 */
995 if (__seccomp_filter(this_syscall, NULL, true))
996 return -1;
997
Kees Cook8112c4f2016-06-01 16:02:17 -0700998 return 0;
Andy Lutomirski13aa72f2014-07-21 18:49:15 -0700999
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001000 case SECCOMP_RET_USER_NOTIF:
Christian Braunerfb3c53862019-09-20 10:30:05 +02001001 if (seccomp_do_user_notification(this_syscall, match, sd))
1002 goto skip;
1003
1004 return 0;
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001005
Tyler Hicks59f5cf42017-08-11 04:33:57 +00001006 case SECCOMP_RET_LOG:
1007 seccomp_log(this_syscall, 0, action, true);
1008 return 0;
1009
Andy Lutomirski13aa72f2014-07-21 18:49:15 -07001010 case SECCOMP_RET_ALLOW:
Kees Cookdeb4de82017-08-02 15:00:40 -07001011 /*
1012 * Note that the "match" filter will always be NULL for
1013 * this action since SECCOMP_RET_ALLOW is the starting
1014 * state in seccomp_run_filters().
1015 */
Kees Cook8112c4f2016-06-01 16:02:17 -07001016 return 0;
Andy Lutomirski13aa72f2014-07-21 18:49:15 -07001017
Kees Cookfd768752017-08-11 12:53:18 -07001018 case SECCOMP_RET_KILL_THREAD:
Kees Cook4d3b0b02017-08-11 13:01:39 -07001019 case SECCOMP_RET_KILL_PROCESS:
Kees Cook131b6352017-02-23 09:24:24 -08001020 default:
Tyler Hickse66a3992017-08-11 04:33:56 +00001021 seccomp_log(this_syscall, SIGSYS, action, true);
Kees Cookd7276e32017-02-07 15:18:51 -08001022 /* Dump core only if this is the last remaining thread. */
Kees Cook4d3b0b02017-08-11 13:01:39 -07001023 if (action == SECCOMP_RET_KILL_PROCESS ||
1024 get_nr_threads(current) == 1) {
Eric W. Biedermanae7795b2018-09-25 11:27:20 +02001025 kernel_siginfo_t info;
Kees Cook131b6352017-02-23 09:24:24 -08001026
Kees Cookd7276e32017-02-07 15:18:51 -08001027 /* Show the original registers in the dump. */
1028 syscall_rollback(current, task_pt_regs(current));
1029 /* Trigger a manual coredump since do_exit skips it. */
1030 seccomp_init_siginfo(&info, this_syscall, data);
1031 do_coredump(&info);
1032 }
Kees Cook4d3b0b02017-08-11 13:01:39 -07001033 if (action == SECCOMP_RET_KILL_PROCESS)
1034 do_group_exit(SIGSYS);
1035 else
1036 do_exit(SIGSYS);
Will Drewry8156b452012-04-17 14:48:58 -05001037 }
Andy Lutomirski13aa72f2014-07-21 18:49:15 -07001038
1039 unreachable();
1040
1041skip:
Tyler Hickse66a3992017-08-11 04:33:56 +00001042 seccomp_log(this_syscall, 0, action, match ? match->log : false);
Kees Cook8112c4f2016-06-01 16:02:17 -07001043 return -1;
1044}
1045#else
Kees Cookce6526e2016-06-01 19:29:15 -07001046static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1047 const bool recheck_after_trace)
Kees Cook8112c4f2016-06-01 16:02:17 -07001048{
1049 BUG();
Andy Lutomirski13aa72f2014-07-21 18:49:15 -07001050}
1051#endif
1052
Kees Cook8112c4f2016-06-01 16:02:17 -07001053int __secure_computing(const struct seccomp_data *sd)
Andy Lutomirski13aa72f2014-07-21 18:49:15 -07001054{
1055 int mode = current->seccomp.mode;
Kees Cook8112c4f2016-06-01 16:02:17 -07001056 int this_syscall;
Andy Lutomirski13aa72f2014-07-21 18:49:15 -07001057
Masahiro Yamada97f26452016-08-03 13:45:50 -07001058 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
Tycho Andersen13c4a902015-06-13 09:02:48 -06001059 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
Kees Cook8112c4f2016-06-01 16:02:17 -07001060 return 0;
1061
1062 this_syscall = sd ? sd->nr :
1063 syscall_get_nr(current, task_pt_regs(current));
Tycho Andersen13c4a902015-06-13 09:02:48 -06001064
Andy Lutomirski13aa72f2014-07-21 18:49:15 -07001065 switch (mode) {
1066 case SECCOMP_MODE_STRICT:
1067 __secure_computing_strict(this_syscall); /* may call do_exit */
Kees Cook8112c4f2016-06-01 16:02:17 -07001068 return 0;
Andy Lutomirski13aa72f2014-07-21 18:49:15 -07001069 case SECCOMP_MODE_FILTER:
Kees Cookce6526e2016-06-01 19:29:15 -07001070 return __seccomp_filter(this_syscall, sd, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 default:
1072 BUG();
1073 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074}
Andy Lutomirskia4412fc2014-07-21 18:49:14 -07001075#endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
Andrea Arcangeli1d9d02f2007-07-15 23:41:32 -07001076
1077long prctl_get_seccomp(void)
1078{
1079 return current->seccomp.mode;
1080}
1081
Will Drewrye2cfabdf2012-04-12 16:47:57 -05001082/**
Kees Cook3b23dd12014-06-25 15:55:25 -07001083 * seccomp_set_mode_strict: internal function for setting strict seccomp
Will Drewrye2cfabdf2012-04-12 16:47:57 -05001084 *
1085 * Once current->seccomp.mode is non-zero, it may not be changed.
1086 *
1087 * Returns 0 on success or -EINVAL on failure.
1088 */
Kees Cook3b23dd12014-06-25 15:55:25 -07001089static long seccomp_set_mode_strict(void)
Andrea Arcangeli1d9d02f2007-07-15 23:41:32 -07001090{
Kees Cook3b23dd12014-06-25 15:55:25 -07001091 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
Will Drewrye2cfabdf2012-04-12 16:47:57 -05001092 long ret = -EINVAL;
Andrea Arcangeli1d9d02f2007-07-15 23:41:32 -07001093
Kees Cookdbd952122014-06-27 15:18:48 -07001094 spin_lock_irq(&current->sighand->siglock);
1095
Kees Cook1f41b4502014-06-25 15:38:02 -07001096 if (!seccomp_may_assign_mode(seccomp_mode))
Andrea Arcangeli1d9d02f2007-07-15 23:41:32 -07001097 goto out;
1098
Andrea Arcangelicf99aba2007-07-15 23:41:33 -07001099#ifdef TIF_NOTSC
Kees Cook3b23dd12014-06-25 15:55:25 -07001100 disable_TSC();
Andrea Arcangelicf99aba2007-07-15 23:41:33 -07001101#endif
Kees Cook00a02d02018-05-03 14:56:12 -07001102 seccomp_assign_mode(current, seccomp_mode, 0);
Kees Cook3b23dd12014-06-25 15:55:25 -07001103 ret = 0;
1104
1105out:
Kees Cookdbd952122014-06-27 15:18:48 -07001106 spin_unlock_irq(&current->sighand->siglock);
Kees Cook3b23dd12014-06-25 15:55:25 -07001107
1108 return ret;
1109}
1110
Will Drewrye2cfabdf2012-04-12 16:47:57 -05001111#ifdef CONFIG_SECCOMP_FILTER
Tycho Andersena566a902020-09-01 19:40:16 -06001112static void seccomp_notify_detach(struct seccomp_filter *filter)
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001113{
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001114 struct seccomp_knotif *knotif;
1115
Tycho Andersena811dc62019-01-12 11:24:20 -07001116 if (!filter)
Tycho Andersena566a902020-09-01 19:40:16 -06001117 return;
Tycho Andersena811dc62019-01-12 11:24:20 -07001118
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001119 mutex_lock(&filter->notify_lock);
1120
1121 /*
1122 * If this file is being closed because e.g. the task who owned it
1123 * died, let's wake everyone up who was waiting on us.
1124 */
1125 list_for_each_entry(knotif, &filter->notif->notifications, list) {
1126 if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1127 continue;
1128
1129 knotif->state = SECCOMP_NOTIFY_REPLIED;
1130 knotif->error = -ENOSYS;
1131 knotif->val = 0;
1132
Sargun Dhillon7cf97b12020-06-02 18:10:43 -07001133 /*
1134 * We do not need to wake up any pending addfd messages, as
1135 * the notifier will do that for us, as this just looks
1136 * like a standard reply.
1137 */
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001138 complete(&knotif->ready);
1139 }
1140
1141 kfree(filter->notif);
1142 filter->notif = NULL;
1143 mutex_unlock(&filter->notify_lock);
Tycho Andersena566a902020-09-01 19:40:16 -06001144}
1145
1146static int seccomp_notify_release(struct inode *inode, struct file *file)
1147{
1148 struct seccomp_filter *filter = file->private_data;
1149
1150 seccomp_notify_detach(filter);
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001151 __put_seccomp_filter(filter);
1152 return 0;
1153}
1154
Sargun Dhillon9f87dcf2020-06-01 04:25:32 -07001155/* must be called with notif_lock held */
1156static inline struct seccomp_knotif *
1157find_notification(struct seccomp_filter *filter, u64 id)
1158{
1159 struct seccomp_knotif *cur;
1160
1161 lockdep_assert_held(&filter->notify_lock);
1162
1163 list_for_each_entry(cur, &filter->notif->notifications, list) {
1164 if (cur->id == id)
1165 return cur;
1166 }
1167
1168 return NULL;
1169}
1170
1171
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001172static long seccomp_notify_recv(struct seccomp_filter *filter,
1173 void __user *buf)
1174{
1175 struct seccomp_knotif *knotif = NULL, *cur;
1176 struct seccomp_notif unotif;
1177 ssize_t ret;
1178
Sargun Dhillon2882d532019-12-28 22:24:50 -08001179 /* Verify that we're not given garbage to keep struct extensible. */
1180 ret = check_zeroed_user(buf, sizeof(unotif));
1181 if (ret < 0)
1182 return ret;
1183 if (!ret)
1184 return -EINVAL;
1185
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001186 memset(&unotif, 0, sizeof(unotif));
1187
1188 ret = down_interruptible(&filter->notif->request);
1189 if (ret < 0)
1190 return ret;
1191
1192 mutex_lock(&filter->notify_lock);
1193 list_for_each_entry(cur, &filter->notif->notifications, list) {
1194 if (cur->state == SECCOMP_NOTIFY_INIT) {
1195 knotif = cur;
1196 break;
1197 }
1198 }
1199
1200 /*
1201 * If we didn't find a notification, it could be that the task was
1202 * interrupted by a fatal signal between the time we were woken and
1203 * when we were able to acquire the rw lock.
1204 */
1205 if (!knotif) {
1206 ret = -ENOENT;
1207 goto out;
1208 }
1209
1210 unotif.id = knotif->id;
1211 unotif.pid = task_pid_vnr(knotif->task);
1212 unotif.data = *(knotif->data);
1213
1214 knotif->state = SECCOMP_NOTIFY_SENT;
Christian Brauner76194c42020-06-01 11:50:07 -07001215 wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM);
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001216 ret = 0;
1217out:
1218 mutex_unlock(&filter->notify_lock);
1219
1220 if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1221 ret = -EFAULT;
1222
1223 /*
1224 * Userspace screwed up. To make sure that we keep this
1225 * notification alive, let's reset it back to INIT. It
1226 * may have died when we released the lock, so we need to make
1227 * sure it's still around.
1228 */
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001229 mutex_lock(&filter->notify_lock);
Sargun Dhillon9f87dcf2020-06-01 04:25:32 -07001230 knotif = find_notification(filter, unotif.id);
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001231 if (knotif) {
1232 knotif->state = SECCOMP_NOTIFY_INIT;
1233 up(&filter->notif->request);
1234 }
1235 mutex_unlock(&filter->notify_lock);
1236 }
1237
1238 return ret;
1239}
1240
1241static long seccomp_notify_send(struct seccomp_filter *filter,
1242 void __user *buf)
1243{
1244 struct seccomp_notif_resp resp = {};
Sargun Dhillon9f87dcf2020-06-01 04:25:32 -07001245 struct seccomp_knotif *knotif;
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001246 long ret;
1247
1248 if (copy_from_user(&resp, buf, sizeof(resp)))
1249 return -EFAULT;
1250
Christian Braunerfb3c53862019-09-20 10:30:05 +02001251 if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1252 return -EINVAL;
1253
1254 if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1255 (resp.error || resp.val))
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001256 return -EINVAL;
1257
1258 ret = mutex_lock_interruptible(&filter->notify_lock);
1259 if (ret < 0)
1260 return ret;
1261
Sargun Dhillon9f87dcf2020-06-01 04:25:32 -07001262 knotif = find_notification(filter, resp.id);
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001263 if (!knotif) {
1264 ret = -ENOENT;
1265 goto out;
1266 }
1267
1268 /* Allow exactly one reply. */
1269 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1270 ret = -EINPROGRESS;
1271 goto out;
1272 }
1273
1274 ret = 0;
1275 knotif->state = SECCOMP_NOTIFY_REPLIED;
1276 knotif->error = resp.error;
1277 knotif->val = resp.val;
Christian Braunerfb3c53862019-09-20 10:30:05 +02001278 knotif->flags = resp.flags;
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001279 complete(&knotif->ready);
1280out:
1281 mutex_unlock(&filter->notify_lock);
1282 return ret;
1283}
1284
1285static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1286 void __user *buf)
1287{
Sargun Dhillon9f87dcf2020-06-01 04:25:32 -07001288 struct seccomp_knotif *knotif;
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001289 u64 id;
1290 long ret;
1291
1292 if (copy_from_user(&id, buf, sizeof(id)))
1293 return -EFAULT;
1294
1295 ret = mutex_lock_interruptible(&filter->notify_lock);
1296 if (ret < 0)
1297 return ret;
1298
Sargun Dhillon9f87dcf2020-06-01 04:25:32 -07001299 knotif = find_notification(filter, id);
1300 if (knotif && knotif->state == SECCOMP_NOTIFY_SENT)
1301 ret = 0;
1302 else
1303 ret = -ENOENT;
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001304
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001305 mutex_unlock(&filter->notify_lock);
1306 return ret;
1307}
1308
Sargun Dhillon7cf97b12020-06-02 18:10:43 -07001309static long seccomp_notify_addfd(struct seccomp_filter *filter,
1310 struct seccomp_notif_addfd __user *uaddfd,
1311 unsigned int size)
1312{
1313 struct seccomp_notif_addfd addfd;
1314 struct seccomp_knotif *knotif;
1315 struct seccomp_kaddfd kaddfd;
1316 int ret;
1317
1318 BUILD_BUG_ON(sizeof(addfd) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0);
1319 BUILD_BUG_ON(sizeof(addfd) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST);
1320
1321 if (size < SECCOMP_NOTIFY_ADDFD_SIZE_VER0 || size >= PAGE_SIZE)
1322 return -EINVAL;
1323
1324 ret = copy_struct_from_user(&addfd, sizeof(addfd), uaddfd, size);
1325 if (ret)
1326 return ret;
1327
1328 if (addfd.newfd_flags & ~O_CLOEXEC)
1329 return -EINVAL;
1330
1331 if (addfd.flags & ~SECCOMP_ADDFD_FLAG_SETFD)
1332 return -EINVAL;
1333
1334 if (addfd.newfd && !(addfd.flags & SECCOMP_ADDFD_FLAG_SETFD))
1335 return -EINVAL;
1336
1337 kaddfd.file = fget(addfd.srcfd);
1338 if (!kaddfd.file)
1339 return -EBADF;
1340
1341 kaddfd.flags = addfd.newfd_flags;
1342 kaddfd.fd = (addfd.flags & SECCOMP_ADDFD_FLAG_SETFD) ?
1343 addfd.newfd : -1;
1344 init_completion(&kaddfd.completion);
1345
1346 ret = mutex_lock_interruptible(&filter->notify_lock);
1347 if (ret < 0)
1348 goto out;
1349
1350 knotif = find_notification(filter, addfd.id);
1351 if (!knotif) {
1352 ret = -ENOENT;
1353 goto out_unlock;
1354 }
1355
1356 /*
1357 * We do not want to allow for FD injection to occur before the
1358 * notification has been picked up by a userspace handler, or after
1359 * the notification has been replied to.
1360 */
1361 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1362 ret = -EINPROGRESS;
1363 goto out_unlock;
1364 }
1365
1366 list_add(&kaddfd.list, &knotif->addfd);
1367 complete(&knotif->ready);
1368 mutex_unlock(&filter->notify_lock);
1369
1370 /* Now we wait for it to be processed or be interrupted */
1371 ret = wait_for_completion_interruptible(&kaddfd.completion);
1372 if (ret == 0) {
1373 /*
1374 * We had a successful completion. The other side has already
1375 * removed us from the addfd queue, and
1376 * wait_for_completion_interruptible has a memory barrier upon
1377 * success that lets us read this value directly without
1378 * locking.
1379 */
1380 ret = kaddfd.ret;
1381 goto out;
1382 }
1383
1384 mutex_lock(&filter->notify_lock);
1385 /*
1386 * Even though we were woken up by a signal and not a successful
1387 * completion, a completion may have happened in the mean time.
1388 *
1389 * We need to check again if the addfd request has been handled,
1390 * and if not, we will remove it from the queue.
1391 */
1392 if (list_empty(&kaddfd.list))
1393 ret = kaddfd.ret;
1394 else
1395 list_del(&kaddfd.list);
1396
1397out_unlock:
1398 mutex_unlock(&filter->notify_lock);
1399out:
1400 fput(kaddfd.file);
1401
1402 return ret;
1403}
1404
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001405static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1406 unsigned long arg)
1407{
1408 struct seccomp_filter *filter = file->private_data;
1409 void __user *buf = (void __user *)arg;
1410
Sargun Dhillon7cf97b12020-06-02 18:10:43 -07001411 /* Fixed-size ioctls */
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001412 switch (cmd) {
1413 case SECCOMP_IOCTL_NOTIF_RECV:
1414 return seccomp_notify_recv(filter, buf);
1415 case SECCOMP_IOCTL_NOTIF_SEND:
1416 return seccomp_notify_send(filter, buf);
Kees Cook47e33c052020-06-15 15:42:46 -07001417 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR:
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001418 case SECCOMP_IOCTL_NOTIF_ID_VALID:
1419 return seccomp_notify_id_valid(filter, buf);
Sargun Dhillon7cf97b12020-06-02 18:10:43 -07001420 }
1421
1422 /* Extensible Argument ioctls */
1423#define EA_IOCTL(cmd) ((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1424 switch (EA_IOCTL(cmd)) {
1425 case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD):
1426 return seccomp_notify_addfd(filter, buf, _IOC_SIZE(cmd));
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001427 default:
1428 return -EINVAL;
1429 }
1430}
1431
1432static __poll_t seccomp_notify_poll(struct file *file,
1433 struct poll_table_struct *poll_tab)
1434{
1435 struct seccomp_filter *filter = file->private_data;
1436 __poll_t ret = 0;
1437 struct seccomp_knotif *cur;
1438
Christian Brauner76194c42020-06-01 11:50:07 -07001439 poll_wait(file, &filter->wqh, poll_tab);
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001440
Tycho Andersen319deec2018-12-12 19:46:54 -07001441 if (mutex_lock_interruptible(&filter->notify_lock) < 0)
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001442 return EPOLLERR;
1443
1444 list_for_each_entry(cur, &filter->notif->notifications, list) {
1445 if (cur->state == SECCOMP_NOTIFY_INIT)
1446 ret |= EPOLLIN | EPOLLRDNORM;
1447 if (cur->state == SECCOMP_NOTIFY_SENT)
1448 ret |= EPOLLOUT | EPOLLWRNORM;
1449 if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1450 break;
1451 }
1452
1453 mutex_unlock(&filter->notify_lock);
1454
Christian Brauner99cdb8b2020-05-31 13:50:30 +02001455 if (refcount_read(&filter->users) == 0)
1456 ret |= EPOLLHUP;
1457
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001458 return ret;
1459}
1460
1461static const struct file_operations seccomp_notify_ops = {
1462 .poll = seccomp_notify_poll,
1463 .release = seccomp_notify_release,
1464 .unlocked_ioctl = seccomp_notify_ioctl,
Sven Schnelle3db81af2020-03-10 13:33:32 +01001465 .compat_ioctl = seccomp_notify_ioctl,
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001466};
1467
1468static struct file *init_listener(struct seccomp_filter *filter)
1469{
1470 struct file *ret = ERR_PTR(-EBUSY);
1471 struct seccomp_filter *cur;
1472
1473 for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1474 if (cur->notif)
1475 goto out;
1476 }
1477
1478 ret = ERR_PTR(-ENOMEM);
1479 filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1480 if (!filter->notif)
1481 goto out;
1482
1483 sema_init(&filter->notif->request, 0);
1484 filter->notif->next_id = get_random_u64();
1485 INIT_LIST_HEAD(&filter->notif->notifications);
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001486
1487 ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1488 filter, O_RDWR);
1489 if (IS_ERR(ret))
1490 goto out_notif;
1491
1492 /* The file has a reference to it now */
1493 __get_seccomp_filter(filter);
1494
1495out_notif:
1496 if (IS_ERR(ret))
1497 kfree(filter->notif);
1498out:
1499 return ret;
1500}
1501
Kees Cook3b23dd12014-06-25 15:55:25 -07001502/**
1503 * seccomp_set_mode_filter: internal function for setting seccomp filter
Kees Cook48dc92b2014-06-25 16:08:24 -07001504 * @flags: flags to change filter behavior
Kees Cook3b23dd12014-06-25 15:55:25 -07001505 * @filter: struct sock_fprog containing filter
1506 *
1507 * This function may be called repeatedly to install additional filters.
1508 * Every filter successfully installed will be evaluated (in reverse order)
1509 * for each system call the task makes.
1510 *
1511 * Once current->seccomp.mode is non-zero, it may not be changed.
1512 *
1513 * Returns 0 on success or -EINVAL on failure.
1514 */
Kees Cook48dc92b2014-06-25 16:08:24 -07001515static long seccomp_set_mode_filter(unsigned int flags,
1516 const char __user *filter)
Kees Cook3b23dd12014-06-25 15:55:25 -07001517{
1518 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
Kees Cookc8bee432014-06-27 15:16:33 -07001519 struct seccomp_filter *prepared = NULL;
Kees Cook3b23dd12014-06-25 15:55:25 -07001520 long ret = -EINVAL;
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001521 int listener = -1;
1522 struct file *listener_f = NULL;
Kees Cook3b23dd12014-06-25 15:55:25 -07001523
Kees Cook48dc92b2014-06-25 16:08:24 -07001524 /* Validate flags. */
Kees Cookc2e1f2e2014-06-05 00:23:17 -07001525 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
Kees Cookdbd952122014-06-27 15:18:48 -07001526 return -EINVAL;
Kees Cook48dc92b2014-06-25 16:08:24 -07001527
Tycho Andersen7a0df7f2019-03-06 13:14:13 -07001528 /*
1529 * In the successful case, NEW_LISTENER returns the new listener fd.
1530 * But in the failure case, TSYNC returns the thread that died. If you
1531 * combine these two flags, there's no way to tell whether something
Tycho Andersen51891492020-03-04 11:05:17 -07001532 * succeeded or failed. So, let's disallow this combination if the user
1533 * has not explicitly requested no errors from TSYNC.
Tycho Andersen7a0df7f2019-03-06 13:14:13 -07001534 */
1535 if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
Tycho Andersen51891492020-03-04 11:05:17 -07001536 (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1537 ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
Tycho Andersen7a0df7f2019-03-06 13:14:13 -07001538 return -EINVAL;
1539
Kees Cookc8bee432014-06-27 15:16:33 -07001540 /* Prepare the new filter before holding any locks. */
1541 prepared = seccomp_prepare_user_filter(filter);
1542 if (IS_ERR(prepared))
1543 return PTR_ERR(prepared);
1544
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001545 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1546 listener = get_unused_fd_flags(O_CLOEXEC);
1547 if (listener < 0) {
1548 ret = listener;
1549 goto out_free;
1550 }
1551
1552 listener_f = init_listener(prepared);
1553 if (IS_ERR(listener_f)) {
1554 put_unused_fd(listener);
1555 ret = PTR_ERR(listener_f);
1556 goto out_free;
1557 }
1558 }
1559
Kees Cookc2e1f2e2014-06-05 00:23:17 -07001560 /*
1561 * Make sure we cannot change seccomp or nnp state via TSYNC
1562 * while another thread is in the middle of calling exec.
1563 */
1564 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1565 mutex_lock_killable(&current->signal->cred_guard_mutex))
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001566 goto out_put_fd;
Kees Cookc2e1f2e2014-06-05 00:23:17 -07001567
Kees Cookdbd952122014-06-27 15:18:48 -07001568 spin_lock_irq(&current->sighand->siglock);
1569
Kees Cook3b23dd12014-06-25 15:55:25 -07001570 if (!seccomp_may_assign_mode(seccomp_mode))
Will Drewrye2cfabdf2012-04-12 16:47:57 -05001571 goto out;
Kees Cook3b23dd12014-06-25 15:55:25 -07001572
Kees Cookc8bee432014-06-27 15:16:33 -07001573 ret = seccomp_attach_filter(flags, prepared);
Kees Cook3b23dd12014-06-25 15:55:25 -07001574 if (ret)
1575 goto out;
Kees Cookc8bee432014-06-27 15:16:33 -07001576 /* Do not free the successfully attached filter. */
1577 prepared = NULL;
Andrea Arcangeli1d9d02f2007-07-15 23:41:32 -07001578
Kees Cook00a02d02018-05-03 14:56:12 -07001579 seccomp_assign_mode(current, seccomp_mode, flags);
Will Drewrye2cfabdf2012-04-12 16:47:57 -05001580out:
Kees Cookdbd952122014-06-27 15:18:48 -07001581 spin_unlock_irq(&current->sighand->siglock);
Kees Cookc2e1f2e2014-06-05 00:23:17 -07001582 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
1583 mutex_unlock(&current->signal->cred_guard_mutex);
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001584out_put_fd:
1585 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
Tycho Andersen7a0df7f2019-03-06 13:14:13 -07001586 if (ret) {
Tycho Andersena811dc62019-01-12 11:24:20 -07001587 listener_f->private_data = NULL;
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001588 fput(listener_f);
1589 put_unused_fd(listener);
Tycho Andersena566a902020-09-01 19:40:16 -06001590 seccomp_notify_detach(prepared);
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001591 } else {
1592 fd_install(listener, listener_f);
1593 ret = listener;
1594 }
1595 }
Kees Cookc2e1f2e2014-06-05 00:23:17 -07001596out_free:
Kees Cookc8bee432014-06-27 15:16:33 -07001597 seccomp_filter_free(prepared);
Andrea Arcangeli1d9d02f2007-07-15 23:41:32 -07001598 return ret;
1599}
Kees Cook3b23dd12014-06-25 15:55:25 -07001600#else
Kees Cook48dc92b2014-06-25 16:08:24 -07001601static inline long seccomp_set_mode_filter(unsigned int flags,
1602 const char __user *filter)
Kees Cook3b23dd12014-06-25 15:55:25 -07001603{
1604 return -EINVAL;
1605}
1606#endif
Kees Cookd78ab022014-05-21 15:02:11 -07001607
Tyler Hicksd612b1f2017-08-11 04:33:53 +00001608static long seccomp_get_action_avail(const char __user *uaction)
1609{
1610 u32 action;
1611
1612 if (copy_from_user(&action, uaction, sizeof(action)))
1613 return -EFAULT;
1614
1615 switch (action) {
Kees Cook0466bdb2017-08-11 13:12:11 -07001616 case SECCOMP_RET_KILL_PROCESS:
Kees Cookfd768752017-08-11 12:53:18 -07001617 case SECCOMP_RET_KILL_THREAD:
Tyler Hicksd612b1f2017-08-11 04:33:53 +00001618 case SECCOMP_RET_TRAP:
1619 case SECCOMP_RET_ERRNO:
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001620 case SECCOMP_RET_USER_NOTIF:
Tyler Hicksd612b1f2017-08-11 04:33:53 +00001621 case SECCOMP_RET_TRACE:
Tyler Hicks59f5cf42017-08-11 04:33:57 +00001622 case SECCOMP_RET_LOG:
Tyler Hicksd612b1f2017-08-11 04:33:53 +00001623 case SECCOMP_RET_ALLOW:
1624 break;
1625 default:
1626 return -EOPNOTSUPP;
1627 }
1628
1629 return 0;
1630}
1631
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001632static long seccomp_get_notif_sizes(void __user *usizes)
1633{
1634 struct seccomp_notif_sizes sizes = {
1635 .seccomp_notif = sizeof(struct seccomp_notif),
1636 .seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
1637 .seccomp_data = sizeof(struct seccomp_data),
1638 };
1639
1640 if (copy_to_user(usizes, &sizes, sizeof(sizes)))
1641 return -EFAULT;
1642
1643 return 0;
1644}
1645
Kees Cook48dc92b2014-06-25 16:08:24 -07001646/* Common entry point for both prctl and syscall. */
1647static long do_seccomp(unsigned int op, unsigned int flags,
Tycho Andersena5662e42018-12-09 11:24:12 -07001648 void __user *uargs)
Kees Cook48dc92b2014-06-25 16:08:24 -07001649{
1650 switch (op) {
1651 case SECCOMP_SET_MODE_STRICT:
1652 if (flags != 0 || uargs != NULL)
1653 return -EINVAL;
1654 return seccomp_set_mode_strict();
1655 case SECCOMP_SET_MODE_FILTER:
1656 return seccomp_set_mode_filter(flags, uargs);
Tyler Hicksd612b1f2017-08-11 04:33:53 +00001657 case SECCOMP_GET_ACTION_AVAIL:
1658 if (flags != 0)
1659 return -EINVAL;
1660
1661 return seccomp_get_action_avail(uargs);
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001662 case SECCOMP_GET_NOTIF_SIZES:
1663 if (flags != 0)
1664 return -EINVAL;
1665
1666 return seccomp_get_notif_sizes(uargs);
Kees Cook48dc92b2014-06-25 16:08:24 -07001667 default:
1668 return -EINVAL;
1669 }
1670}
1671
1672SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
Tycho Andersena5662e42018-12-09 11:24:12 -07001673 void __user *, uargs)
Kees Cook48dc92b2014-06-25 16:08:24 -07001674{
1675 return do_seccomp(op, flags, uargs);
1676}
1677
Kees Cookd78ab022014-05-21 15:02:11 -07001678/**
1679 * prctl_set_seccomp: configures current->seccomp.mode
1680 * @seccomp_mode: requested mode to use
1681 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1682 *
1683 * Returns 0 on success or -EINVAL on failure.
1684 */
Tycho Andersena5662e42018-12-09 11:24:12 -07001685long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
Kees Cookd78ab022014-05-21 15:02:11 -07001686{
Kees Cook48dc92b2014-06-25 16:08:24 -07001687 unsigned int op;
Tycho Andersena5662e42018-12-09 11:24:12 -07001688 void __user *uargs;
Kees Cook48dc92b2014-06-25 16:08:24 -07001689
Kees Cook3b23dd12014-06-25 15:55:25 -07001690 switch (seccomp_mode) {
1691 case SECCOMP_MODE_STRICT:
Kees Cook48dc92b2014-06-25 16:08:24 -07001692 op = SECCOMP_SET_MODE_STRICT;
1693 /*
1694 * Setting strict mode through prctl always ignored filter,
1695 * so make sure it is always NULL here to pass the internal
1696 * check in do_seccomp().
1697 */
1698 uargs = NULL;
1699 break;
Kees Cook3b23dd12014-06-25 15:55:25 -07001700 case SECCOMP_MODE_FILTER:
Kees Cook48dc92b2014-06-25 16:08:24 -07001701 op = SECCOMP_SET_MODE_FILTER;
1702 uargs = filter;
1703 break;
Kees Cook3b23dd12014-06-25 15:55:25 -07001704 default:
1705 return -EINVAL;
1706 }
Kees Cook48dc92b2014-06-25 16:08:24 -07001707
1708 /* prctl interface doesn't have flags, so they are always zero. */
1709 return do_seccomp(op, 0, uargs);
Kees Cookd78ab022014-05-21 15:02:11 -07001710}
Tycho Andersenf8e529ed2015-10-27 09:23:59 +09001711
1712#if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
Tycho Andersenf06eae82017-10-11 09:39:20 -06001713static struct seccomp_filter *get_nth_filter(struct task_struct *task,
1714 unsigned long filter_off)
1715{
1716 struct seccomp_filter *orig, *filter;
1717 unsigned long count;
1718
1719 /*
1720 * Note: this is only correct because the caller should be the (ptrace)
1721 * tracer of the task, otherwise lock_task_sighand is needed.
1722 */
1723 spin_lock_irq(&task->sighand->siglock);
1724
1725 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1726 spin_unlock_irq(&task->sighand->siglock);
1727 return ERR_PTR(-EINVAL);
1728 }
1729
1730 orig = task->seccomp.filter;
1731 __get_seccomp_filter(orig);
1732 spin_unlock_irq(&task->sighand->siglock);
1733
1734 count = 0;
1735 for (filter = orig; filter; filter = filter->prev)
1736 count++;
1737
1738 if (filter_off >= count) {
1739 filter = ERR_PTR(-ENOENT);
1740 goto out;
1741 }
1742
1743 count -= filter_off;
1744 for (filter = orig; filter && count > 1; filter = filter->prev)
1745 count--;
1746
1747 if (WARN_ON(count != 1 || !filter)) {
1748 filter = ERR_PTR(-ENOENT);
1749 goto out;
1750 }
1751
1752 __get_seccomp_filter(filter);
1753
1754out:
1755 __put_seccomp_filter(orig);
1756 return filter;
1757}
1758
Tycho Andersenf8e529ed2015-10-27 09:23:59 +09001759long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1760 void __user *data)
1761{
1762 struct seccomp_filter *filter;
1763 struct sock_fprog_kern *fprog;
1764 long ret;
Tycho Andersenf8e529ed2015-10-27 09:23:59 +09001765
1766 if (!capable(CAP_SYS_ADMIN) ||
1767 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1768 return -EACCES;
1769 }
1770
Tycho Andersenf06eae82017-10-11 09:39:20 -06001771 filter = get_nth_filter(task, filter_off);
1772 if (IS_ERR(filter))
1773 return PTR_ERR(filter);
Tycho Andersenf8e529ed2015-10-27 09:23:59 +09001774
1775 fprog = filter->prog->orig_prog;
1776 if (!fprog) {
Mickaël Salaün470bf1f2016-03-24 02:46:33 +01001777 /* This must be a new non-cBPF filter, since we save
Tycho Andersenf8e529ed2015-10-27 09:23:59 +09001778 * every cBPF filter's orig_prog above when
1779 * CONFIG_CHECKPOINT_RESTORE is enabled.
1780 */
1781 ret = -EMEDIUMTYPE;
1782 goto out;
1783 }
1784
1785 ret = fprog->len;
1786 if (!data)
1787 goto out;
1788
Tycho Andersenf8e529ed2015-10-27 09:23:59 +09001789 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1790 ret = -EFAULT;
1791
Tycho Andersenf8e529ed2015-10-27 09:23:59 +09001792out:
Tyler Hicks8e5f1ad2017-08-11 04:33:52 +00001793 __put_seccomp_filter(filter);
1794 return ret;
Tycho Andersenf8e529ed2015-10-27 09:23:59 +09001795}
Tycho Andersenf8e529ed2015-10-27 09:23:59 +09001796
Tycho Andersen26500472017-10-11 09:39:21 -06001797long seccomp_get_metadata(struct task_struct *task,
1798 unsigned long size, void __user *data)
1799{
1800 long ret;
1801 struct seccomp_filter *filter;
1802 struct seccomp_metadata kmd = {};
1803
1804 if (!capable(CAP_SYS_ADMIN) ||
1805 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1806 return -EACCES;
1807 }
1808
1809 size = min_t(unsigned long, size, sizeof(kmd));
1810
Tycho Andersen63bb0042018-02-20 19:47:46 -07001811 if (size < sizeof(kmd.filter_off))
1812 return -EINVAL;
1813
1814 if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
Tycho Andersen26500472017-10-11 09:39:21 -06001815 return -EFAULT;
1816
1817 filter = get_nth_filter(task, kmd.filter_off);
1818 if (IS_ERR(filter))
1819 return PTR_ERR(filter);
1820
Tycho Andersen26500472017-10-11 09:39:21 -06001821 if (filter->log)
1822 kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1823
1824 ret = size;
1825 if (copy_to_user(data, &kmd, size))
1826 ret = -EFAULT;
1827
1828 __put_seccomp_filter(filter);
Tycho Andersenf8e529ed2015-10-27 09:23:59 +09001829 return ret;
1830}
1831#endif
Tyler Hicks8e5f1ad2017-08-11 04:33:52 +00001832
1833#ifdef CONFIG_SYSCTL
1834
1835/* Human readable action names for friendly sysctl interaction */
Kees Cook0466bdb2017-08-11 13:12:11 -07001836#define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
Kees Cookfd768752017-08-11 12:53:18 -07001837#define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
Tyler Hicks8e5f1ad2017-08-11 04:33:52 +00001838#define SECCOMP_RET_TRAP_NAME "trap"
1839#define SECCOMP_RET_ERRNO_NAME "errno"
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001840#define SECCOMP_RET_USER_NOTIF_NAME "user_notif"
Tyler Hicks8e5f1ad2017-08-11 04:33:52 +00001841#define SECCOMP_RET_TRACE_NAME "trace"
Tyler Hicks59f5cf42017-08-11 04:33:57 +00001842#define SECCOMP_RET_LOG_NAME "log"
Tyler Hicks8e5f1ad2017-08-11 04:33:52 +00001843#define SECCOMP_RET_ALLOW_NAME "allow"
1844
Kees Cookfd768752017-08-11 12:53:18 -07001845static const char seccomp_actions_avail[] =
Kees Cook0466bdb2017-08-11 13:12:11 -07001846 SECCOMP_RET_KILL_PROCESS_NAME " "
Kees Cookfd768752017-08-11 12:53:18 -07001847 SECCOMP_RET_KILL_THREAD_NAME " "
1848 SECCOMP_RET_TRAP_NAME " "
1849 SECCOMP_RET_ERRNO_NAME " "
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001850 SECCOMP_RET_USER_NOTIF_NAME " "
Kees Cookfd768752017-08-11 12:53:18 -07001851 SECCOMP_RET_TRACE_NAME " "
1852 SECCOMP_RET_LOG_NAME " "
1853 SECCOMP_RET_ALLOW_NAME;
Tyler Hicks8e5f1ad2017-08-11 04:33:52 +00001854
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001855struct seccomp_log_name {
1856 u32 log;
1857 const char *name;
1858};
1859
1860static const struct seccomp_log_name seccomp_log_names[] = {
Kees Cook0466bdb2017-08-11 13:12:11 -07001861 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
Kees Cookfd768752017-08-11 12:53:18 -07001862 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001863 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1864 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
Tycho Andersen6a21cc52018-12-09 11:24:13 -07001865 { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001866 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
Tyler Hicks59f5cf42017-08-11 04:33:57 +00001867 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001868 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1869 { }
1870};
1871
1872static bool seccomp_names_from_actions_logged(char *names, size_t size,
Tyler Hicksbeb44ac2018-05-04 01:08:13 +00001873 u32 actions_logged,
1874 const char *sep)
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001875{
1876 const struct seccomp_log_name *cur;
Tyler Hicksbeb44ac2018-05-04 01:08:13 +00001877 bool append_sep = false;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001878
1879 for (cur = seccomp_log_names; cur->name && size; cur++) {
1880 ssize_t ret;
1881
1882 if (!(actions_logged & cur->log))
1883 continue;
1884
Tyler Hicksbeb44ac2018-05-04 01:08:13 +00001885 if (append_sep) {
1886 ret = strscpy(names, sep, size);
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001887 if (ret < 0)
1888 return false;
1889
1890 names += ret;
1891 size -= ret;
1892 } else
Tyler Hicksbeb44ac2018-05-04 01:08:13 +00001893 append_sep = true;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001894
1895 ret = strscpy(names, cur->name, size);
1896 if (ret < 0)
1897 return false;
1898
1899 names += ret;
1900 size -= ret;
1901 }
1902
1903 return true;
1904}
1905
1906static bool seccomp_action_logged_from_name(u32 *action_logged,
1907 const char *name)
1908{
1909 const struct seccomp_log_name *cur;
1910
1911 for (cur = seccomp_log_names; cur->name; cur++) {
1912 if (!strcmp(cur->name, name)) {
1913 *action_logged = cur->log;
1914 return true;
1915 }
1916 }
1917
1918 return false;
1919}
1920
1921static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1922{
1923 char *name;
1924
1925 *actions_logged = 0;
1926 while ((name = strsep(&names, " ")) && *name) {
1927 u32 action_logged = 0;
1928
1929 if (!seccomp_action_logged_from_name(&action_logged, name))
1930 return false;
1931
1932 *actions_logged |= action_logged;
1933 }
1934
1935 return true;
1936}
1937
Tyler Hicksd013db02018-05-04 01:08:12 +00001938static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1939 size_t *lenp, loff_t *ppos)
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001940{
1941 char names[sizeof(seccomp_actions_avail)];
1942 struct ctl_table table;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001943
1944 memset(names, 0, sizeof(names));
1945
Tyler Hicksd013db02018-05-04 01:08:12 +00001946 if (!seccomp_names_from_actions_logged(names, sizeof(names),
Tyler Hicksbeb44ac2018-05-04 01:08:13 +00001947 seccomp_actions_logged, " "))
Tyler Hicksd013db02018-05-04 01:08:12 +00001948 return -EINVAL;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001949
1950 table = *ro_table;
1951 table.data = names;
1952 table.maxlen = sizeof(names);
Tyler Hicksd013db02018-05-04 01:08:12 +00001953 return proc_dostring(&table, 0, buffer, lenp, ppos);
1954}
1955
1956static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer,
Tyler Hicksea6eca72018-05-04 01:08:14 +00001957 size_t *lenp, loff_t *ppos, u32 *actions_logged)
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001958{
1959 char names[sizeof(seccomp_actions_avail)];
1960 struct ctl_table table;
1961 int ret;
1962
Tyler Hicksd013db02018-05-04 01:08:12 +00001963 if (!capable(CAP_SYS_ADMIN))
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001964 return -EPERM;
1965
1966 memset(names, 0, sizeof(names));
1967
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001968 table = *ro_table;
1969 table.data = names;
1970 table.maxlen = sizeof(names);
Tyler Hicksd013db02018-05-04 01:08:12 +00001971 ret = proc_dostring(&table, 1, buffer, lenp, ppos);
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001972 if (ret)
1973 return ret;
1974
Tyler Hicksea6eca72018-05-04 01:08:14 +00001975 if (!seccomp_actions_logged_from_names(actions_logged, table.data))
Tyler Hicksd013db02018-05-04 01:08:12 +00001976 return -EINVAL;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001977
Tyler Hicksea6eca72018-05-04 01:08:14 +00001978 if (*actions_logged & SECCOMP_LOG_ALLOW)
Tyler Hicksd013db02018-05-04 01:08:12 +00001979 return -EINVAL;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001980
Tyler Hicksea6eca72018-05-04 01:08:14 +00001981 seccomp_actions_logged = *actions_logged;
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00001982 return 0;
1983}
1984
Tyler Hicksea6eca72018-05-04 01:08:14 +00001985static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
1986 int ret)
1987{
1988 char names[sizeof(seccomp_actions_avail)];
1989 char old_names[sizeof(seccomp_actions_avail)];
1990 const char *new = names;
1991 const char *old = old_names;
1992
1993 if (!audit_enabled)
1994 return;
1995
1996 memset(names, 0, sizeof(names));
1997 memset(old_names, 0, sizeof(old_names));
1998
1999 if (ret)
2000 new = "?";
2001 else if (!actions_logged)
2002 new = "(none)";
2003 else if (!seccomp_names_from_actions_logged(names, sizeof(names),
2004 actions_logged, ","))
2005 new = "?";
2006
2007 if (!old_actions_logged)
2008 old = "(none)";
2009 else if (!seccomp_names_from_actions_logged(old_names,
2010 sizeof(old_names),
2011 old_actions_logged, ","))
2012 old = "?";
2013
2014 return audit_seccomp_actions_logged(new, old, !ret);
2015}
2016
Tyler Hicksd013db02018-05-04 01:08:12 +00002017static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
Christoph Hellwig32927392020-04-24 08:43:38 +02002018 void *buffer, size_t *lenp,
Tyler Hicksd013db02018-05-04 01:08:12 +00002019 loff_t *ppos)
2020{
Tyler Hicksea6eca72018-05-04 01:08:14 +00002021 int ret;
2022
2023 if (write) {
2024 u32 actions_logged = 0;
2025 u32 old_actions_logged = seccomp_actions_logged;
2026
2027 ret = write_actions_logged(ro_table, buffer, lenp, ppos,
2028 &actions_logged);
2029 audit_actions_logged(actions_logged, old_actions_logged, ret);
2030 } else
2031 ret = read_actions_logged(ro_table, buffer, lenp, ppos);
2032
2033 return ret;
Tyler Hicksd013db02018-05-04 01:08:12 +00002034}
2035
Tyler Hicks8e5f1ad2017-08-11 04:33:52 +00002036static struct ctl_path seccomp_sysctl_path[] = {
2037 { .procname = "kernel", },
2038 { .procname = "seccomp", },
2039 { }
2040};
2041
2042static struct ctl_table seccomp_sysctl_table[] = {
2043 {
2044 .procname = "actions_avail",
2045 .data = (void *) &seccomp_actions_avail,
2046 .maxlen = sizeof(seccomp_actions_avail),
2047 .mode = 0444,
2048 .proc_handler = proc_dostring,
2049 },
Tyler Hicks0ddec0f2017-08-11 04:33:54 +00002050 {
2051 .procname = "actions_logged",
2052 .mode = 0644,
2053 .proc_handler = seccomp_actions_logged_handler,
2054 },
Tyler Hicks8e5f1ad2017-08-11 04:33:52 +00002055 { }
2056};
2057
2058static int __init seccomp_sysctl_init(void)
2059{
2060 struct ctl_table_header *hdr;
2061
2062 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
2063 if (!hdr)
Kees Cooke68f9d42020-06-15 22:02:56 -07002064 pr_warn("sysctl registration failed\n");
Tyler Hicks8e5f1ad2017-08-11 04:33:52 +00002065 else
2066 kmemleak_not_leak(hdr);
2067
2068 return 0;
2069}
2070
2071device_initcall(seccomp_sysctl_init)
2072
2073#endif /* CONFIG_SYSCTL */