Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | /* |
| 3 | * linux/kernel/seccomp.c |
| 4 | * |
| 5 | * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com> |
| 6 | * |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 7 | * 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 15 | */ |
| 16 | |
Kees Cook | 0b5fa22 | 2017-06-26 09:24:00 -0700 | [diff] [blame] | 17 | #include <linux/refcount.h> |
Eric Paris | 85e7bac | 2012-01-03 14:23:05 -0500 | [diff] [blame] | 18 | #include <linux/audit.h> |
Roland McGrath | 5b10174 | 2009-02-27 23:25:54 -0800 | [diff] [blame] | 19 | #include <linux/compat.h> |
Mike Frysinger | b25e671 | 2017-01-19 22:28:57 -0600 | [diff] [blame] | 20 | #include <linux/coredump.h> |
Tyler Hicks | 8e5f1ad | 2017-08-11 04:33:52 +0000 | [diff] [blame] | 21 | #include <linux/kmemleak.h> |
Kees Cook | 87895ae | 2018-05-01 15:07:31 -0700 | [diff] [blame] | 22 | #include <linux/nospec.h> |
| 23 | #include <linux/prctl.h> |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 24 | #include <linux/sched.h> |
Ingo Molnar | 68db0cf | 2017-02-08 18:51:37 +0100 | [diff] [blame] | 25 | #include <linux/sched/task_stack.h> |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 26 | #include <linux/seccomp.h> |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 27 | #include <linux/slab.h> |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 28 | #include <linux/syscalls.h> |
Tyler Hicks | 8e5f1ad | 2017-08-11 04:33:52 +0000 | [diff] [blame] | 29 | #include <linux/sysctl.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | |
Kees Cook | b39e021 | 2022-02-07 20:21:13 -0800 | [diff] [blame] | 31 | /* Not exposed in headers: strictly internal use only. */ |
| 32 | #define SECCOMP_MODE_DEAD (SECCOMP_MODE_FILTER + 1) |
| 33 | |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 34 | #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 35 | #include <asm/syscall.h> |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 36 | #endif |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 37 | |
| 38 | #ifdef CONFIG_SECCOMP_FILTER |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 39 | #include <linux/filter.h> |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 40 | #include <linux/pid.h> |
Will Drewry | fb0fadf | 2012-04-12 16:48:02 -0500 | [diff] [blame] | 41 | #include <linux/ptrace.h> |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 42 | #include <linux/security.h> |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 43 | #include <linux/tracehook.h> |
| 44 | #include <linux/uaccess.h> |
| 45 | |
| 46 | /** |
| 47 | * struct seccomp_filter - container for seccomp BPF programs |
| 48 | * |
| 49 | * @usage: reference count to manage the object lifetime. |
| 50 | * get/put helpers should be used when accessing an instance |
| 51 | * outside of a lifetime-guarded section. In general, this |
| 52 | * is only needed for handling filters shared across tasks. |
Tyler Hicks | e66a399 | 2017-08-11 04:33:56 +0000 | [diff] [blame] | 53 | * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 54 | * @prev: points to a previously installed, or inherited, filter |
Mickaël Salaün | 285fdfc | 2016-09-20 19:39:47 +0200 | [diff] [blame] | 55 | * @prog: the BPF program to evaluate |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 56 | * |
| 57 | * seccomp_filter objects are organized in a tree linked via the @prev |
| 58 | * pointer. For any task, it appears to be a singly-linked list starting |
| 59 | * with current->seccomp.filter, the most recently attached or inherited filter. |
| 60 | * However, multiple filters may share a @prev node, by way of fork(), which |
| 61 | * results in a unidirectional tree existing in memory. This is similar to |
| 62 | * how namespaces work. |
| 63 | * |
| 64 | * seccomp_filter objects should never be modified after being attached |
| 65 | * to a task_struct (other than @usage). |
| 66 | */ |
| 67 | struct seccomp_filter { |
Kees Cook | 0b5fa22 | 2017-06-26 09:24:00 -0700 | [diff] [blame] | 68 | refcount_t usage; |
Tyler Hicks | e66a399 | 2017-08-11 04:33:56 +0000 | [diff] [blame] | 69 | bool log; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 70 | struct seccomp_filter *prev; |
Alexei Starovoitov | 7ae457c | 2014-07-30 20:34:16 -0700 | [diff] [blame] | 71 | struct bpf_prog *prog; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 72 | }; |
| 73 | |
| 74 | /* Limit any path through the tree to 256KB worth of instructions. */ |
| 75 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) |
| 76 | |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 77 | /* |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 78 | * Endianness is explicitly ignored and left for BPF program authors to manage |
| 79 | * as per the specific architecture. |
| 80 | */ |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 81 | static void populate_seccomp_data(struct seccomp_data *sd) |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 82 | { |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 83 | struct task_struct *task = current; |
| 84 | struct pt_regs *regs = task_pt_regs(task); |
Daniel Borkmann | 2eac764 | 2014-04-14 21:02:59 +0200 | [diff] [blame] | 85 | unsigned long args[6]; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 86 | |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 87 | sd->nr = syscall_get_nr(task, regs); |
Linus Torvalds | 0b74717 | 2014-04-12 12:38:53 -0700 | [diff] [blame] | 88 | sd->arch = syscall_get_arch(); |
Daniel Borkmann | 2eac764 | 2014-04-14 21:02:59 +0200 | [diff] [blame] | 89 | syscall_get_arguments(task, regs, 0, 6, args); |
| 90 | sd->args[0] = args[0]; |
| 91 | sd->args[1] = args[1]; |
| 92 | sd->args[2] = args[2]; |
| 93 | sd->args[3] = args[3]; |
| 94 | sd->args[4] = args[4]; |
| 95 | sd->args[5] = args[5]; |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 96 | sd->instruction_pointer = KSTK_EIP(task); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 97 | } |
| 98 | |
| 99 | /** |
| 100 | * seccomp_check_filter - verify seccomp filter code |
| 101 | * @filter: filter to verify |
| 102 | * @flen: length of filter |
| 103 | * |
Alexei Starovoitov | 4df95ff | 2014-07-30 20:34:14 -0700 | [diff] [blame] | 104 | * Takes a previously checked filter (by bpf_check_classic) and |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 105 | * redirects all filter code that loads struct sk_buff data |
| 106 | * and related data through seccomp_bpf_load. It also |
| 107 | * enforces length and alignment checking of those loads. |
| 108 | * |
| 109 | * Returns 0 if the rule set is legal or -EINVAL if not. |
| 110 | */ |
| 111 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) |
| 112 | { |
| 113 | int pc; |
| 114 | for (pc = 0; pc < flen; pc++) { |
| 115 | struct sock_filter *ftest = &filter[pc]; |
| 116 | u16 code = ftest->code; |
| 117 | u32 k = ftest->k; |
| 118 | |
| 119 | switch (code) { |
Daniel Borkmann | 3480593 | 2014-05-29 10:22:50 +0200 | [diff] [blame] | 120 | case BPF_LD | BPF_W | BPF_ABS: |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 121 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 122 | /* 32-bit aligned and not out of bounds. */ |
| 123 | if (k >= sizeof(struct seccomp_data) || k & 3) |
| 124 | return -EINVAL; |
| 125 | continue; |
Daniel Borkmann | 3480593 | 2014-05-29 10:22:50 +0200 | [diff] [blame] | 126 | case BPF_LD | BPF_W | BPF_LEN: |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 127 | ftest->code = BPF_LD | BPF_IMM; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 128 | ftest->k = sizeof(struct seccomp_data); |
| 129 | continue; |
Daniel Borkmann | 3480593 | 2014-05-29 10:22:50 +0200 | [diff] [blame] | 130 | case BPF_LDX | BPF_W | BPF_LEN: |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 131 | ftest->code = BPF_LDX | BPF_IMM; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 132 | ftest->k = sizeof(struct seccomp_data); |
| 133 | continue; |
| 134 | /* Explicitly include allowed calls. */ |
Daniel Borkmann | 3480593 | 2014-05-29 10:22:50 +0200 | [diff] [blame] | 135 | case BPF_RET | BPF_K: |
| 136 | case BPF_RET | BPF_A: |
| 137 | case BPF_ALU | BPF_ADD | BPF_K: |
| 138 | case BPF_ALU | BPF_ADD | BPF_X: |
| 139 | case BPF_ALU | BPF_SUB | BPF_K: |
| 140 | case BPF_ALU | BPF_SUB | BPF_X: |
| 141 | case BPF_ALU | BPF_MUL | BPF_K: |
| 142 | case BPF_ALU | BPF_MUL | BPF_X: |
| 143 | case BPF_ALU | BPF_DIV | BPF_K: |
| 144 | case BPF_ALU | BPF_DIV | BPF_X: |
| 145 | case BPF_ALU | BPF_AND | BPF_K: |
| 146 | case BPF_ALU | BPF_AND | BPF_X: |
| 147 | case BPF_ALU | BPF_OR | BPF_K: |
| 148 | case BPF_ALU | BPF_OR | BPF_X: |
| 149 | case BPF_ALU | BPF_XOR | BPF_K: |
| 150 | case BPF_ALU | BPF_XOR | BPF_X: |
| 151 | case BPF_ALU | BPF_LSH | BPF_K: |
| 152 | case BPF_ALU | BPF_LSH | BPF_X: |
| 153 | case BPF_ALU | BPF_RSH | BPF_K: |
| 154 | case BPF_ALU | BPF_RSH | BPF_X: |
| 155 | case BPF_ALU | BPF_NEG: |
| 156 | case BPF_LD | BPF_IMM: |
| 157 | case BPF_LDX | BPF_IMM: |
| 158 | case BPF_MISC | BPF_TAX: |
| 159 | case BPF_MISC | BPF_TXA: |
| 160 | case BPF_LD | BPF_MEM: |
| 161 | case BPF_LDX | BPF_MEM: |
| 162 | case BPF_ST: |
| 163 | case BPF_STX: |
| 164 | case BPF_JMP | BPF_JA: |
| 165 | case BPF_JMP | BPF_JEQ | BPF_K: |
| 166 | case BPF_JMP | BPF_JEQ | BPF_X: |
| 167 | case BPF_JMP | BPF_JGE | BPF_K: |
| 168 | case BPF_JMP | BPF_JGE | BPF_X: |
| 169 | case BPF_JMP | BPF_JGT | BPF_K: |
| 170 | case BPF_JMP | BPF_JGT | BPF_X: |
| 171 | case BPF_JMP | BPF_JSET | BPF_K: |
| 172 | case BPF_JMP | BPF_JSET | BPF_X: |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 173 | continue; |
| 174 | default: |
| 175 | return -EINVAL; |
| 176 | } |
| 177 | } |
| 178 | return 0; |
| 179 | } |
| 180 | |
| 181 | /** |
Mickaël Salaün | 285fdfc | 2016-09-20 19:39:47 +0200 | [diff] [blame] | 182 | * seccomp_run_filters - evaluates all seccomp filters against @sd |
| 183 | * @sd: optional seccomp data to be passed to filters |
Kees Cook | deb4de8 | 2017-08-02 15:00:40 -0700 | [diff] [blame] | 184 | * @match: stores struct seccomp_filter that resulted in the return value, |
| 185 | * unless filter returned SECCOMP_RET_ALLOW, in which case it will |
| 186 | * be unchanged. |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 187 | * |
| 188 | * Returns valid seccomp BPF response codes. |
| 189 | */ |
Kees Cook | 0466bdb | 2017-08-11 13:12:11 -0700 | [diff] [blame] | 190 | #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL))) |
Kees Cook | deb4de8 | 2017-08-02 15:00:40 -0700 | [diff] [blame] | 191 | static u32 seccomp_run_filters(const struct seccomp_data *sd, |
| 192 | struct seccomp_filter **match) |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 193 | { |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 194 | struct seccomp_data sd_local; |
Will Drewry | acf3b2c | 2012-04-12 16:47:59 -0500 | [diff] [blame] | 195 | u32 ret = SECCOMP_RET_ALLOW; |
Pranith Kumar | 8225d38 | 2014-11-21 10:06:01 -0500 | [diff] [blame] | 196 | /* Make sure cross-thread synced filter points somewhere sane. */ |
| 197 | struct seccomp_filter *f = |
Will Deacon | 5383f45 | 2017-10-24 11:22:48 +0100 | [diff] [blame] | 198 | READ_ONCE(current->seccomp.filter); |
Will Drewry | acf3b2c | 2012-04-12 16:47:59 -0500 | [diff] [blame] | 199 | |
| 200 | /* Ensure unexpected behavior doesn't result in failing open. */ |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 201 | if (unlikely(WARN_ON(f == NULL))) |
Kees Cook | 4d3b0b0 | 2017-08-11 13:01:39 -0700 | [diff] [blame] | 202 | return SECCOMP_RET_KILL_PROCESS; |
Will Drewry | acf3b2c | 2012-04-12 16:47:59 -0500 | [diff] [blame] | 203 | |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 204 | if (!sd) { |
| 205 | populate_seccomp_data(&sd_local); |
| 206 | sd = &sd_local; |
| 207 | } |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 208 | |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 209 | /* |
| 210 | * All filters in the list are evaluated and the lowest BPF return |
Will Drewry | acf3b2c | 2012-04-12 16:47:59 -0500 | [diff] [blame] | 211 | * value always takes priority (ignoring the DATA). |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 212 | */ |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 213 | for (; f; f = f->prev) { |
Daniel Borkmann | 8857519 | 2016-11-26 01:28:04 +0100 | [diff] [blame] | 214 | u32 cur_ret = BPF_PROG_RUN(f->prog, sd); |
Alexei Starovoitov | 8f577ca | 2014-05-13 19:50:47 -0700 | [diff] [blame] | 215 | |
Kees Cook | 0466bdb | 2017-08-11 13:12:11 -0700 | [diff] [blame] | 216 | if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) { |
Will Drewry | acf3b2c | 2012-04-12 16:47:59 -0500 | [diff] [blame] | 217 | ret = cur_ret; |
Kees Cook | deb4de8 | 2017-08-02 15:00:40 -0700 | [diff] [blame] | 218 | *match = f; |
| 219 | } |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 220 | } |
| 221 | return ret; |
| 222 | } |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 223 | #endif /* CONFIG_SECCOMP_FILTER */ |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 224 | |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 225 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
| 226 | { |
Guenter Roeck | 69f6a34 | 2014-08-10 20:50:30 -0700 | [diff] [blame] | 227 | assert_spin_locked(¤t->sighand->siglock); |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 228 | |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 229 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
| 230 | return false; |
| 231 | |
| 232 | return true; |
| 233 | } |
| 234 | |
Thomas Gleixner | 61dfdc1 | 2018-05-04 15:12:06 +0200 | [diff] [blame] | 235 | void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { } |
Kees Cook | 87895ae | 2018-05-01 15:07:31 -0700 | [diff] [blame] | 236 | |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 237 | static inline void seccomp_assign_mode(struct task_struct *task, |
Kees Cook | 9939db7 | 2018-05-03 14:56:12 -0700 | [diff] [blame] | 238 | unsigned long seccomp_mode, |
| 239 | unsigned long flags) |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 240 | { |
Guenter Roeck | 69f6a34 | 2014-08-10 20:50:30 -0700 | [diff] [blame] | 241 | assert_spin_locked(&task->sighand->siglock); |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 242 | |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 243 | task->seccomp.mode = seccomp_mode; |
| 244 | /* |
| 245 | * Make sure TIF_SECCOMP cannot be set before the mode (and |
| 246 | * filter) is set. |
| 247 | */ |
| 248 | smp_mb__before_atomic(); |
Kees Cook | 9939db7 | 2018-05-03 14:56:12 -0700 | [diff] [blame] | 249 | /* Assume default seccomp processes want spec flaw mitigation. */ |
| 250 | if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0) |
Thomas Gleixner | 61dfdc1 | 2018-05-04 15:12:06 +0200 | [diff] [blame] | 251 | arch_seccomp_spec_mitigate(task); |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 252 | set_tsk_thread_flag(task, TIF_SECCOMP); |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 253 | } |
| 254 | |
| 255 | #ifdef CONFIG_SECCOMP_FILTER |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 256 | /* Returns 1 if the parent is an ancestor of the child. */ |
| 257 | static int is_ancestor(struct seccomp_filter *parent, |
| 258 | struct seccomp_filter *child) |
| 259 | { |
| 260 | /* NULL is the root ancestor. */ |
| 261 | if (parent == NULL) |
| 262 | return 1; |
| 263 | for (; child; child = child->prev) |
| 264 | if (child == parent) |
| 265 | return 1; |
| 266 | return 0; |
| 267 | } |
| 268 | |
| 269 | /** |
| 270 | * seccomp_can_sync_threads: checks if all threads can be synchronized |
| 271 | * |
| 272 | * Expects sighand and cred_guard_mutex locks to be held. |
| 273 | * |
| 274 | * Returns 0 on success, -ve on error, or the pid of a thread which was |
| 275 | * either not in the correct seccomp mode or it did not have an ancestral |
| 276 | * seccomp filter. |
| 277 | */ |
| 278 | static inline pid_t seccomp_can_sync_threads(void) |
| 279 | { |
| 280 | struct task_struct *thread, *caller; |
| 281 | |
| 282 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); |
Guenter Roeck | 69f6a34 | 2014-08-10 20:50:30 -0700 | [diff] [blame] | 283 | assert_spin_locked(¤t->sighand->siglock); |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 284 | |
| 285 | /* Validate all threads being eligible for synchronization. */ |
| 286 | caller = current; |
| 287 | for_each_thread(caller, thread) { |
| 288 | pid_t failed; |
| 289 | |
| 290 | /* Skip current, since it is initiating the sync. */ |
| 291 | if (thread == caller) |
| 292 | continue; |
| 293 | |
| 294 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || |
| 295 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && |
| 296 | is_ancestor(thread->seccomp.filter, |
| 297 | caller->seccomp.filter))) |
| 298 | continue; |
| 299 | |
| 300 | /* Return the first thread that cannot be synchronized. */ |
| 301 | failed = task_pid_vnr(thread); |
| 302 | /* If the pid cannot be resolved, then return -ESRCH */ |
| 303 | if (unlikely(WARN_ON(failed == 0))) |
| 304 | failed = -ESRCH; |
| 305 | return failed; |
| 306 | } |
| 307 | |
| 308 | return 0; |
| 309 | } |
| 310 | |
| 311 | /** |
| 312 | * seccomp_sync_threads: sets all threads to use current's filter |
| 313 | * |
| 314 | * Expects sighand and cred_guard_mutex locks to be held, and for |
| 315 | * seccomp_can_sync_threads() to have returned success already |
| 316 | * without dropping the locks. |
| 317 | * |
| 318 | */ |
Kees Cook | 9939db7 | 2018-05-03 14:56:12 -0700 | [diff] [blame] | 319 | static inline void seccomp_sync_threads(unsigned long flags) |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 320 | { |
| 321 | struct task_struct *thread, *caller; |
| 322 | |
| 323 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); |
Guenter Roeck | 69f6a34 | 2014-08-10 20:50:30 -0700 | [diff] [blame] | 324 | assert_spin_locked(¤t->sighand->siglock); |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 325 | |
| 326 | /* Synchronize all threads. */ |
| 327 | caller = current; |
| 328 | for_each_thread(caller, thread) { |
| 329 | /* Skip current, since it needs no changes. */ |
| 330 | if (thread == caller) |
| 331 | continue; |
| 332 | |
| 333 | /* Get a task reference for the new leaf node. */ |
| 334 | get_seccomp_filter(caller); |
| 335 | /* |
| 336 | * Drop the task reference to the shared ancestor since |
| 337 | * current's path will hold a reference. (This also |
| 338 | * allows a put before the assignment.) |
| 339 | */ |
| 340 | put_seccomp_filter(thread); |
| 341 | smp_store_release(&thread->seccomp.filter, |
| 342 | caller->seccomp.filter); |
Jann Horn | 103502a | 2015-12-26 06:00:48 +0100 | [diff] [blame] | 343 | |
| 344 | /* |
| 345 | * Don't let an unprivileged task work around |
| 346 | * the no_new_privs restriction by creating |
| 347 | * a thread that sets it up, enters seccomp, |
| 348 | * then dies. |
| 349 | */ |
| 350 | if (task_no_new_privs(caller)) |
| 351 | task_set_no_new_privs(thread); |
| 352 | |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 353 | /* |
| 354 | * Opt the other thread into seccomp if needed. |
| 355 | * As threads are considered to be trust-realm |
| 356 | * equivalent (see ptrace_may_access), it is safe to |
| 357 | * allow one thread to transition the other. |
| 358 | */ |
Jann Horn | 103502a | 2015-12-26 06:00:48 +0100 | [diff] [blame] | 359 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) |
Kees Cook | 9939db7 | 2018-05-03 14:56:12 -0700 | [diff] [blame] | 360 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER, |
| 361 | flags); |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 362 | } |
| 363 | } |
| 364 | |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 365 | /** |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 366 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 367 | * @fprog: BPF program to install |
| 368 | * |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 369 | * Returns filter on success or an ERR_PTR on failure. |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 370 | */ |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 371 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 372 | { |
Daniel Borkmann | ac67eb2 | 2015-05-06 16:12:30 +0200 | [diff] [blame] | 373 | struct seccomp_filter *sfilter; |
| 374 | int ret; |
Masahiro Yamada | 97f2645 | 2016-08-03 13:45:50 -0700 | [diff] [blame] | 375 | const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 376 | |
| 377 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 378 | return ERR_PTR(-EINVAL); |
Nicolas Schichan | d9e12f4 | 2015-05-06 16:12:28 +0200 | [diff] [blame] | 379 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 380 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 381 | |
| 382 | /* |
Fabian Frederick | 119ce5c | 2014-06-06 14:37:53 -0700 | [diff] [blame] | 383 | * Installing a seccomp filter requires that the task has |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 384 | * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. |
| 385 | * This avoids scenarios where unprivileged tasks can affect the |
| 386 | * behavior of privileged children. |
| 387 | */ |
Kees Cook | 1d4457f | 2014-05-21 15:23:46 -0700 | [diff] [blame] | 388 | if (!task_no_new_privs(current) && |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 389 | security_capable_noaudit(current_cred(), current_user_ns(), |
| 390 | CAP_SYS_ADMIN) != 0) |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 391 | return ERR_PTR(-EACCES); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 392 | |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 393 | /* Allocate a new seccomp_filter */ |
Daniel Borkmann | ac67eb2 | 2015-05-06 16:12:30 +0200 | [diff] [blame] | 394 | sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN); |
| 395 | if (!sfilter) |
Nicolas Schichan | d9e12f4 | 2015-05-06 16:12:28 +0200 | [diff] [blame] | 396 | return ERR_PTR(-ENOMEM); |
Daniel Borkmann | ac67eb2 | 2015-05-06 16:12:30 +0200 | [diff] [blame] | 397 | |
| 398 | ret = bpf_prog_create_from_user(&sfilter->prog, fprog, |
Tycho Andersen | f8e529e | 2015-10-27 09:23:59 +0900 | [diff] [blame] | 399 | seccomp_check_filter, save_orig); |
Daniel Borkmann | ac67eb2 | 2015-05-06 16:12:30 +0200 | [diff] [blame] | 400 | if (ret < 0) { |
| 401 | kfree(sfilter); |
| 402 | return ERR_PTR(ret); |
Nicolas Schichan | d9e12f4 | 2015-05-06 16:12:28 +0200 | [diff] [blame] | 403 | } |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 404 | |
Kees Cook | 0b5fa22 | 2017-06-26 09:24:00 -0700 | [diff] [blame] | 405 | refcount_set(&sfilter->usage, 1); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 406 | |
Daniel Borkmann | ac67eb2 | 2015-05-06 16:12:30 +0200 | [diff] [blame] | 407 | return sfilter; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 408 | } |
| 409 | |
| 410 | /** |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 411 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 412 | * @user_filter: pointer to the user data containing a sock_fprog. |
| 413 | * |
| 414 | * Returns 0 on success and non-zero otherwise. |
| 415 | */ |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 416 | static struct seccomp_filter * |
| 417 | seccomp_prepare_user_filter(const char __user *user_filter) |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 418 | { |
| 419 | struct sock_fprog fprog; |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 420 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 421 | |
| 422 | #ifdef CONFIG_COMPAT |
Andy Lutomirski | 5c38065 | 2016-03-22 14:24:52 -0700 | [diff] [blame] | 423 | if (in_compat_syscall()) { |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 424 | struct compat_sock_fprog fprog32; |
| 425 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) |
| 426 | goto out; |
| 427 | fprog.len = fprog32.len; |
| 428 | fprog.filter = compat_ptr(fprog32.filter); |
| 429 | } else /* falls through to the if below. */ |
| 430 | #endif |
| 431 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) |
| 432 | goto out; |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 433 | filter = seccomp_prepare_filter(&fprog); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 434 | out: |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 435 | return filter; |
| 436 | } |
| 437 | |
| 438 | /** |
| 439 | * seccomp_attach_filter: validate and attach filter |
| 440 | * @flags: flags to change filter behavior |
| 441 | * @filter: seccomp filter to add to the current process |
| 442 | * |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 443 | * Caller must be holding current->sighand->siglock lock. |
| 444 | * |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 445 | * Returns 0 on success, -ve on error. |
| 446 | */ |
| 447 | static long seccomp_attach_filter(unsigned int flags, |
| 448 | struct seccomp_filter *filter) |
| 449 | { |
| 450 | unsigned long total_insns; |
| 451 | struct seccomp_filter *walker; |
| 452 | |
Guenter Roeck | 69f6a34 | 2014-08-10 20:50:30 -0700 | [diff] [blame] | 453 | assert_spin_locked(¤t->sighand->siglock); |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 454 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 455 | /* Validate resulting filter length. */ |
| 456 | total_insns = filter->prog->len; |
| 457 | for (walker = current->seccomp.filter; walker; walker = walker->prev) |
| 458 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ |
| 459 | if (total_insns > MAX_INSNS_PER_PATH) |
| 460 | return -ENOMEM; |
| 461 | |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 462 | /* If thread sync has been requested, check that it is possible. */ |
| 463 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { |
| 464 | int ret; |
| 465 | |
| 466 | ret = seccomp_can_sync_threads(); |
| 467 | if (ret) |
| 468 | return ret; |
| 469 | } |
| 470 | |
Tyler Hicks | e66a399 | 2017-08-11 04:33:56 +0000 | [diff] [blame] | 471 | /* Set log flag, if present. */ |
| 472 | if (flags & SECCOMP_FILTER_FLAG_LOG) |
| 473 | filter->log = true; |
| 474 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 475 | /* |
| 476 | * If there is an existing filter, make it the prev and don't drop its |
| 477 | * task reference. |
| 478 | */ |
| 479 | filter->prev = current->seccomp.filter; |
| 480 | current->seccomp.filter = filter; |
| 481 | |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 482 | /* Now that the new filter is in place, synchronize to all threads. */ |
| 483 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
Kees Cook | 9939db7 | 2018-05-03 14:56:12 -0700 | [diff] [blame] | 484 | seccomp_sync_threads(flags); |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 485 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 486 | return 0; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 487 | } |
| 488 | |
Colin Ian King | 084f560 | 2017-09-29 14:26:48 +0100 | [diff] [blame] | 489 | static void __get_seccomp_filter(struct seccomp_filter *filter) |
Oleg Nesterov | 66a733e | 2017-09-27 09:25:30 -0600 | [diff] [blame] | 490 | { |
| 491 | /* Reference count is bounded by the number of total processes. */ |
| 492 | refcount_inc(&filter->usage); |
| 493 | } |
| 494 | |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 495 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ |
| 496 | void get_seccomp_filter(struct task_struct *tsk) |
| 497 | { |
| 498 | struct seccomp_filter *orig = tsk->seccomp.filter; |
| 499 | if (!orig) |
| 500 | return; |
Oleg Nesterov | 66a733e | 2017-09-27 09:25:30 -0600 | [diff] [blame] | 501 | __get_seccomp_filter(orig); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 502 | } |
| 503 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 504 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
| 505 | { |
| 506 | if (filter) { |
Daniel Borkmann | bab1899 | 2015-10-02 15:17:33 +0200 | [diff] [blame] | 507 | bpf_prog_destroy(filter->prog); |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 508 | kfree(filter); |
| 509 | } |
| 510 | } |
| 511 | |
Oleg Nesterov | 66a733e | 2017-09-27 09:25:30 -0600 | [diff] [blame] | 512 | static void __put_seccomp_filter(struct seccomp_filter *orig) |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 513 | { |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 514 | /* Clean up single-reference branches iteratively. */ |
Kees Cook | 0b5fa22 | 2017-06-26 09:24:00 -0700 | [diff] [blame] | 515 | while (orig && refcount_dec_and_test(&orig->usage)) { |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 516 | struct seccomp_filter *freeme = orig; |
| 517 | orig = orig->prev; |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 518 | seccomp_filter_free(freeme); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 519 | } |
| 520 | } |
Will Drewry | bb6ea43 | 2012-04-12 16:48:01 -0500 | [diff] [blame] | 521 | |
Oleg Nesterov | 66a733e | 2017-09-27 09:25:30 -0600 | [diff] [blame] | 522 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
| 523 | void put_seccomp_filter(struct task_struct *tsk) |
| 524 | { |
| 525 | __put_seccomp_filter(tsk->seccomp.filter); |
| 526 | } |
| 527 | |
Mike Frysinger | b25e671 | 2017-01-19 22:28:57 -0600 | [diff] [blame] | 528 | static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason) |
| 529 | { |
| 530 | memset(info, 0, sizeof(*info)); |
| 531 | info->si_signo = SIGSYS; |
| 532 | info->si_code = SYS_SECCOMP; |
| 533 | info->si_call_addr = (void __user *)KSTK_EIP(current); |
| 534 | info->si_errno = reason; |
| 535 | info->si_arch = syscall_get_arch(); |
| 536 | info->si_syscall = syscall; |
| 537 | } |
| 538 | |
Will Drewry | bb6ea43 | 2012-04-12 16:48:01 -0500 | [diff] [blame] | 539 | /** |
| 540 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation |
| 541 | * @syscall: syscall number to send to userland |
| 542 | * @reason: filter-supplied reason code to send to userland (via si_errno) |
| 543 | * |
| 544 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. |
| 545 | */ |
| 546 | static void seccomp_send_sigsys(int syscall, int reason) |
| 547 | { |
| 548 | struct siginfo info; |
Mike Frysinger | b25e671 | 2017-01-19 22:28:57 -0600 | [diff] [blame] | 549 | seccomp_init_siginfo(&info, syscall, reason); |
Will Drewry | bb6ea43 | 2012-04-12 16:48:01 -0500 | [diff] [blame] | 550 | force_sig_info(SIGSYS, &info, current); |
| 551 | } |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 552 | #endif /* CONFIG_SECCOMP_FILTER */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 553 | |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 554 | /* For use with seccomp_actions_logged */ |
Kees Cook | 4d3b0b0 | 2017-08-11 13:01:39 -0700 | [diff] [blame] | 555 | #define SECCOMP_LOG_KILL_PROCESS (1 << 0) |
| 556 | #define SECCOMP_LOG_KILL_THREAD (1 << 1) |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 557 | #define SECCOMP_LOG_TRAP (1 << 2) |
| 558 | #define SECCOMP_LOG_ERRNO (1 << 3) |
| 559 | #define SECCOMP_LOG_TRACE (1 << 4) |
Tyler Hicks | 59f5cf4 | 2017-08-11 04:33:57 +0000 | [diff] [blame] | 560 | #define SECCOMP_LOG_LOG (1 << 5) |
| 561 | #define SECCOMP_LOG_ALLOW (1 << 6) |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 562 | |
Kees Cook | 4d3b0b0 | 2017-08-11 13:01:39 -0700 | [diff] [blame] | 563 | static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS | |
| 564 | SECCOMP_LOG_KILL_THREAD | |
Kees Cook | fd76875 | 2017-08-11 12:53:18 -0700 | [diff] [blame] | 565 | SECCOMP_LOG_TRAP | |
| 566 | SECCOMP_LOG_ERRNO | |
| 567 | SECCOMP_LOG_TRACE | |
Tyler Hicks | 59f5cf4 | 2017-08-11 04:33:57 +0000 | [diff] [blame] | 568 | SECCOMP_LOG_LOG; |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 569 | |
Tyler Hicks | e66a399 | 2017-08-11 04:33:56 +0000 | [diff] [blame] | 570 | static inline void seccomp_log(unsigned long syscall, long signr, u32 action, |
| 571 | bool requested) |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 572 | { |
| 573 | bool log = false; |
| 574 | |
| 575 | switch (action) { |
| 576 | case SECCOMP_RET_ALLOW: |
Tyler Hicks | e66a399 | 2017-08-11 04:33:56 +0000 | [diff] [blame] | 577 | break; |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 578 | case SECCOMP_RET_TRAP: |
Tyler Hicks | e66a399 | 2017-08-11 04:33:56 +0000 | [diff] [blame] | 579 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP; |
| 580 | break; |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 581 | case SECCOMP_RET_ERRNO: |
Tyler Hicks | e66a399 | 2017-08-11 04:33:56 +0000 | [diff] [blame] | 582 | log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO; |
| 583 | break; |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 584 | case SECCOMP_RET_TRACE: |
Tyler Hicks | e66a399 | 2017-08-11 04:33:56 +0000 | [diff] [blame] | 585 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE; |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 586 | break; |
Tyler Hicks | 59f5cf4 | 2017-08-11 04:33:57 +0000 | [diff] [blame] | 587 | case SECCOMP_RET_LOG: |
| 588 | log = seccomp_actions_logged & SECCOMP_LOG_LOG; |
| 589 | break; |
Kees Cook | fd76875 | 2017-08-11 12:53:18 -0700 | [diff] [blame] | 590 | case SECCOMP_RET_KILL_THREAD: |
Kees Cook | fd76875 | 2017-08-11 12:53:18 -0700 | [diff] [blame] | 591 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD; |
Kees Cook | 4d3b0b0 | 2017-08-11 13:01:39 -0700 | [diff] [blame] | 592 | break; |
| 593 | case SECCOMP_RET_KILL_PROCESS: |
| 594 | default: |
| 595 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS; |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 596 | } |
| 597 | |
| 598 | /* |
Kees Cook | fd76875 | 2017-08-11 12:53:18 -0700 | [diff] [blame] | 599 | * Force an audit message to be emitted when the action is RET_KILL_*, |
Tyler Hicks | 59f5cf4 | 2017-08-11 04:33:57 +0000 | [diff] [blame] | 600 | * RET_LOG, or the FILTER_FLAG_LOG bit was set and the action is |
| 601 | * allowed to be logged by the admin. |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 602 | */ |
| 603 | if (log) |
| 604 | return __audit_seccomp(syscall, signr, action); |
| 605 | |
| 606 | /* |
| 607 | * Let the audit subsystem decide if the action should be audited based |
| 608 | * on whether the current task itself is being audited. |
| 609 | */ |
| 610 | return audit_seccomp(syscall, signr, action); |
| 611 | } |
| 612 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | /* |
| 614 | * Secure computing mode 1 allows only read/write/exit/sigreturn. |
| 615 | * To be fully secure this must be combined with rlimit |
| 616 | * to limit the stack allocations too. |
| 617 | */ |
Matt Redfearn | cb4253a | 2016-03-29 09:35:34 +0100 | [diff] [blame] | 618 | static const int mode1_syscalls[] = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 619 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, |
| 620 | 0, /* null terminated */ |
| 621 | }; |
| 622 | |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 623 | static void __secure_computing_strict(int this_syscall) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 624 | { |
Matt Redfearn | cb4253a | 2016-03-29 09:35:34 +0100 | [diff] [blame] | 625 | const int *syscall_whitelist = mode1_syscalls; |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 626 | #ifdef CONFIG_COMPAT |
Andy Lutomirski | 5c38065 | 2016-03-22 14:24:52 -0700 | [diff] [blame] | 627 | if (in_compat_syscall()) |
Matt Redfearn | c983f0e | 2016-03-29 09:35:32 +0100 | [diff] [blame] | 628 | syscall_whitelist = get_compat_mode1_syscalls(); |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 629 | #endif |
| 630 | do { |
| 631 | if (*syscall_whitelist == this_syscall) |
| 632 | return; |
| 633 | } while (*++syscall_whitelist); |
| 634 | |
| 635 | #ifdef SECCOMP_DEBUG |
| 636 | dump_stack(); |
| 637 | #endif |
Kees Cook | b39e021 | 2022-02-07 20:21:13 -0800 | [diff] [blame] | 638 | current->seccomp.mode = SECCOMP_MODE_DEAD; |
Kees Cook | fd76875 | 2017-08-11 12:53:18 -0700 | [diff] [blame] | 639 | seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true); |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 640 | do_exit(SIGKILL); |
| 641 | } |
| 642 | |
| 643 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
| 644 | void secure_computing_strict(int this_syscall) |
| 645 | { |
| 646 | int mode = current->seccomp.mode; |
| 647 | |
Masahiro Yamada | 97f2645 | 2016-08-03 13:45:50 -0700 | [diff] [blame] | 648 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
Tycho Andersen | 13c4a90 | 2015-06-13 09:02:48 -0600 | [diff] [blame] | 649 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
| 650 | return; |
| 651 | |
Kees Cook | 221272f | 2015-06-15 15:29:16 -0700 | [diff] [blame] | 652 | if (mode == SECCOMP_MODE_DISABLED) |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 653 | return; |
| 654 | else if (mode == SECCOMP_MODE_STRICT) |
| 655 | __secure_computing_strict(this_syscall); |
| 656 | else |
| 657 | BUG(); |
| 658 | } |
| 659 | #else |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 660 | |
| 661 | #ifdef CONFIG_SECCOMP_FILTER |
Kees Cook | ce6526e | 2016-06-01 19:29:15 -0700 | [diff] [blame] | 662 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
| 663 | const bool recheck_after_trace) |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 664 | { |
| 665 | u32 filter_ret, action; |
Kees Cook | deb4de8 | 2017-08-02 15:00:40 -0700 | [diff] [blame] | 666 | struct seccomp_filter *match = NULL; |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 667 | int data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 669 | /* |
| 670 | * Make sure that any changes to mode from another thread have |
| 671 | * been seen after TIF_SECCOMP was seen. |
| 672 | */ |
| 673 | rmb(); |
| 674 | |
Kees Cook | deb4de8 | 2017-08-02 15:00:40 -0700 | [diff] [blame] | 675 | filter_ret = seccomp_run_filters(sd, &match); |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 676 | data = filter_ret & SECCOMP_RET_DATA; |
Kees Cook | 0466bdb | 2017-08-11 13:12:11 -0700 | [diff] [blame] | 677 | action = filter_ret & SECCOMP_RET_ACTION_FULL; |
Andy Lutomirski | 87b526d | 2012-10-01 11:40:45 -0700 | [diff] [blame] | 678 | |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 679 | switch (action) { |
| 680 | case SECCOMP_RET_ERRNO: |
Kees Cook | 580c57f | 2015-02-17 13:48:00 -0800 | [diff] [blame] | 681 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
| 682 | if (data > MAX_ERRNO) |
| 683 | data = MAX_ERRNO; |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 684 | syscall_set_return_value(current, task_pt_regs(current), |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 685 | -data, 0); |
| 686 | goto skip; |
| 687 | |
| 688 | case SECCOMP_RET_TRAP: |
| 689 | /* Show the handler the original registers. */ |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 690 | syscall_rollback(current, task_pt_regs(current)); |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 691 | /* Let the filter pass back 16 bits of data. */ |
| 692 | seccomp_send_sigsys(this_syscall, data); |
| 693 | goto skip; |
| 694 | |
| 695 | case SECCOMP_RET_TRACE: |
Kees Cook | ce6526e | 2016-06-01 19:29:15 -0700 | [diff] [blame] | 696 | /* We've been put in this state by the ptracer already. */ |
| 697 | if (recheck_after_trace) |
| 698 | return 0; |
| 699 | |
Kees Cook | 8112c4f | 2016-06-01 16:02:17 -0700 | [diff] [blame] | 700 | /* ENOSYS these calls if there is no tracer attached. */ |
| 701 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { |
| 702 | syscall_set_return_value(current, |
| 703 | task_pt_regs(current), |
| 704 | -ENOSYS, 0); |
| 705 | goto skip; |
| 706 | } |
| 707 | |
| 708 | /* Allow the BPF to provide the event message */ |
| 709 | ptrace_event(PTRACE_EVENT_SECCOMP, data); |
| 710 | /* |
| 711 | * The delivery of a fatal signal during event |
Kees Cook | 485a252 | 2016-08-10 16:28:09 -0700 | [diff] [blame] | 712 | * notification may silently skip tracer notification, |
| 713 | * which could leave us with a potentially unmodified |
| 714 | * syscall that the tracer would have liked to have |
| 715 | * changed. Since the process is about to die, we just |
| 716 | * force the syscall to be skipped and let the signal |
| 717 | * kill the process and correctly handle any tracer exit |
| 718 | * notifications. |
Kees Cook | 8112c4f | 2016-06-01 16:02:17 -0700 | [diff] [blame] | 719 | */ |
| 720 | if (fatal_signal_pending(current)) |
Kees Cook | 485a252 | 2016-08-10 16:28:09 -0700 | [diff] [blame] | 721 | goto skip; |
Kees Cook | 8112c4f | 2016-06-01 16:02:17 -0700 | [diff] [blame] | 722 | /* Check if the tracer forced the syscall to be skipped. */ |
| 723 | this_syscall = syscall_get_nr(current, task_pt_regs(current)); |
| 724 | if (this_syscall < 0) |
| 725 | goto skip; |
| 726 | |
Kees Cook | ce6526e | 2016-06-01 19:29:15 -0700 | [diff] [blame] | 727 | /* |
| 728 | * Recheck the syscall, since it may have changed. This |
| 729 | * intentionally uses a NULL struct seccomp_data to force |
| 730 | * a reload of all registers. This does not goto skip since |
| 731 | * a skip would have already been reported. |
| 732 | */ |
| 733 | if (__seccomp_filter(this_syscall, NULL, true)) |
| 734 | return -1; |
| 735 | |
Kees Cook | 8112c4f | 2016-06-01 16:02:17 -0700 | [diff] [blame] | 736 | return 0; |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 737 | |
Tyler Hicks | 59f5cf4 | 2017-08-11 04:33:57 +0000 | [diff] [blame] | 738 | case SECCOMP_RET_LOG: |
| 739 | seccomp_log(this_syscall, 0, action, true); |
| 740 | return 0; |
| 741 | |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 742 | case SECCOMP_RET_ALLOW: |
Kees Cook | deb4de8 | 2017-08-02 15:00:40 -0700 | [diff] [blame] | 743 | /* |
| 744 | * Note that the "match" filter will always be NULL for |
| 745 | * this action since SECCOMP_RET_ALLOW is the starting |
| 746 | * state in seccomp_run_filters(). |
| 747 | */ |
Kees Cook | 8112c4f | 2016-06-01 16:02:17 -0700 | [diff] [blame] | 748 | return 0; |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 749 | |
Kees Cook | fd76875 | 2017-08-11 12:53:18 -0700 | [diff] [blame] | 750 | case SECCOMP_RET_KILL_THREAD: |
Kees Cook | 4d3b0b0 | 2017-08-11 13:01:39 -0700 | [diff] [blame] | 751 | case SECCOMP_RET_KILL_PROCESS: |
Kees Cook | 131b635 | 2017-02-23 09:24:24 -0800 | [diff] [blame] | 752 | default: |
Kees Cook | b39e021 | 2022-02-07 20:21:13 -0800 | [diff] [blame] | 753 | current->seccomp.mode = SECCOMP_MODE_DEAD; |
Tyler Hicks | e66a399 | 2017-08-11 04:33:56 +0000 | [diff] [blame] | 754 | seccomp_log(this_syscall, SIGSYS, action, true); |
Kees Cook | d7276e3 | 2017-02-07 15:18:51 -0800 | [diff] [blame] | 755 | /* Dump core only if this is the last remaining thread. */ |
Kees Cook | 4d3b0b0 | 2017-08-11 13:01:39 -0700 | [diff] [blame] | 756 | if (action == SECCOMP_RET_KILL_PROCESS || |
| 757 | get_nr_threads(current) == 1) { |
Kees Cook | 131b635 | 2017-02-23 09:24:24 -0800 | [diff] [blame] | 758 | siginfo_t info; |
| 759 | |
Kees Cook | d7276e3 | 2017-02-07 15:18:51 -0800 | [diff] [blame] | 760 | /* Show the original registers in the dump. */ |
| 761 | syscall_rollback(current, task_pt_regs(current)); |
| 762 | /* Trigger a manual coredump since do_exit skips it. */ |
| 763 | seccomp_init_siginfo(&info, this_syscall, data); |
| 764 | do_coredump(&info); |
| 765 | } |
Kees Cook | 4d3b0b0 | 2017-08-11 13:01:39 -0700 | [diff] [blame] | 766 | if (action == SECCOMP_RET_KILL_PROCESS) |
| 767 | do_group_exit(SIGSYS); |
| 768 | else |
| 769 | do_exit(SIGSYS); |
Will Drewry | 8156b45 | 2012-04-17 14:48:58 -0500 | [diff] [blame] | 770 | } |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 771 | |
| 772 | unreachable(); |
| 773 | |
| 774 | skip: |
Tyler Hicks | e66a399 | 2017-08-11 04:33:56 +0000 | [diff] [blame] | 775 | seccomp_log(this_syscall, 0, action, match ? match->log : false); |
Kees Cook | 8112c4f | 2016-06-01 16:02:17 -0700 | [diff] [blame] | 776 | return -1; |
| 777 | } |
| 778 | #else |
Kees Cook | ce6526e | 2016-06-01 19:29:15 -0700 | [diff] [blame] | 779 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
| 780 | const bool recheck_after_trace) |
Kees Cook | 8112c4f | 2016-06-01 16:02:17 -0700 | [diff] [blame] | 781 | { |
| 782 | BUG(); |
Paul Cercueil | 44ae925 | 2021-01-11 17:28:39 +0000 | [diff] [blame] | 783 | |
| 784 | return -1; |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 785 | } |
| 786 | #endif |
| 787 | |
Kees Cook | 8112c4f | 2016-06-01 16:02:17 -0700 | [diff] [blame] | 788 | int __secure_computing(const struct seccomp_data *sd) |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 789 | { |
| 790 | int mode = current->seccomp.mode; |
Kees Cook | 8112c4f | 2016-06-01 16:02:17 -0700 | [diff] [blame] | 791 | int this_syscall; |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 792 | |
Masahiro Yamada | 97f2645 | 2016-08-03 13:45:50 -0700 | [diff] [blame] | 793 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
Tycho Andersen | 13c4a90 | 2015-06-13 09:02:48 -0600 | [diff] [blame] | 794 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
Kees Cook | 8112c4f | 2016-06-01 16:02:17 -0700 | [diff] [blame] | 795 | return 0; |
| 796 | |
| 797 | this_syscall = sd ? sd->nr : |
| 798 | syscall_get_nr(current, task_pt_regs(current)); |
Tycho Andersen | 13c4a90 | 2015-06-13 09:02:48 -0600 | [diff] [blame] | 799 | |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 800 | switch (mode) { |
| 801 | case SECCOMP_MODE_STRICT: |
| 802 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
Kees Cook | 8112c4f | 2016-06-01 16:02:17 -0700 | [diff] [blame] | 803 | return 0; |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 804 | case SECCOMP_MODE_FILTER: |
Kees Cook | ce6526e | 2016-06-01 19:29:15 -0700 | [diff] [blame] | 805 | return __seccomp_filter(this_syscall, sd, false); |
Kees Cook | b39e021 | 2022-02-07 20:21:13 -0800 | [diff] [blame] | 806 | /* Surviving SECCOMP_RET_KILL_* must be proactively impossible. */ |
| 807 | case SECCOMP_MODE_DEAD: |
| 808 | WARN_ON_ONCE(1); |
| 809 | do_exit(SIGKILL); |
| 810 | return -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 811 | default: |
| 812 | BUG(); |
| 813 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 814 | } |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 815 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 816 | |
| 817 | long prctl_get_seccomp(void) |
| 818 | { |
| 819 | return current->seccomp.mode; |
| 820 | } |
| 821 | |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 822 | /** |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 823 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 824 | * |
| 825 | * Once current->seccomp.mode is non-zero, it may not be changed. |
| 826 | * |
| 827 | * Returns 0 on success or -EINVAL on failure. |
| 828 | */ |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 829 | static long seccomp_set_mode_strict(void) |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 830 | { |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 831 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 832 | long ret = -EINVAL; |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 833 | |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 834 | spin_lock_irq(¤t->sighand->siglock); |
| 835 | |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 836 | if (!seccomp_may_assign_mode(seccomp_mode)) |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 837 | goto out; |
| 838 | |
Andrea Arcangeli | cf99aba | 2007-07-15 23:41:33 -0700 | [diff] [blame] | 839 | #ifdef TIF_NOTSC |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 840 | disable_TSC(); |
Andrea Arcangeli | cf99aba | 2007-07-15 23:41:33 -0700 | [diff] [blame] | 841 | #endif |
Kees Cook | 9939db7 | 2018-05-03 14:56:12 -0700 | [diff] [blame] | 842 | seccomp_assign_mode(current, seccomp_mode, 0); |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 843 | ret = 0; |
| 844 | |
| 845 | out: |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 846 | spin_unlock_irq(¤t->sighand->siglock); |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 847 | |
| 848 | return ret; |
| 849 | } |
| 850 | |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 851 | #ifdef CONFIG_SECCOMP_FILTER |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 852 | /** |
| 853 | * seccomp_set_mode_filter: internal function for setting seccomp filter |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 854 | * @flags: flags to change filter behavior |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 855 | * @filter: struct sock_fprog containing filter |
| 856 | * |
| 857 | * This function may be called repeatedly to install additional filters. |
| 858 | * Every filter successfully installed will be evaluated (in reverse order) |
| 859 | * for each system call the task makes. |
| 860 | * |
| 861 | * Once current->seccomp.mode is non-zero, it may not be changed. |
| 862 | * |
| 863 | * Returns 0 on success or -EINVAL on failure. |
| 864 | */ |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 865 | static long seccomp_set_mode_filter(unsigned int flags, |
| 866 | const char __user *filter) |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 867 | { |
| 868 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 869 | struct seccomp_filter *prepared = NULL; |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 870 | long ret = -EINVAL; |
| 871 | |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 872 | /* Validate flags. */ |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 873 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 874 | return -EINVAL; |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 875 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 876 | /* Prepare the new filter before holding any locks. */ |
| 877 | prepared = seccomp_prepare_user_filter(filter); |
| 878 | if (IS_ERR(prepared)) |
| 879 | return PTR_ERR(prepared); |
| 880 | |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 881 | /* |
| 882 | * Make sure we cannot change seccomp or nnp state via TSYNC |
| 883 | * while another thread is in the middle of calling exec. |
| 884 | */ |
| 885 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && |
| 886 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) |
| 887 | goto out_free; |
| 888 | |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 889 | spin_lock_irq(¤t->sighand->siglock); |
| 890 | |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 891 | if (!seccomp_may_assign_mode(seccomp_mode)) |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 892 | goto out; |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 893 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 894 | ret = seccomp_attach_filter(flags, prepared); |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 895 | if (ret) |
| 896 | goto out; |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 897 | /* Do not free the successfully attached filter. */ |
| 898 | prepared = NULL; |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 899 | |
Kees Cook | 9939db7 | 2018-05-03 14:56:12 -0700 | [diff] [blame] | 900 | seccomp_assign_mode(current, seccomp_mode, flags); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 901 | out: |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 902 | spin_unlock_irq(¤t->sighand->siglock); |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 903 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
| 904 | mutex_unlock(¤t->signal->cred_guard_mutex); |
| 905 | out_free: |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 906 | seccomp_filter_free(prepared); |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 907 | return ret; |
| 908 | } |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 909 | #else |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 910 | static inline long seccomp_set_mode_filter(unsigned int flags, |
| 911 | const char __user *filter) |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 912 | { |
| 913 | return -EINVAL; |
| 914 | } |
| 915 | #endif |
Kees Cook | d78ab02 | 2014-05-21 15:02:11 -0700 | [diff] [blame] | 916 | |
Tyler Hicks | d612b1f | 2017-08-11 04:33:53 +0000 | [diff] [blame] | 917 | static long seccomp_get_action_avail(const char __user *uaction) |
| 918 | { |
| 919 | u32 action; |
| 920 | |
| 921 | if (copy_from_user(&action, uaction, sizeof(action))) |
| 922 | return -EFAULT; |
| 923 | |
| 924 | switch (action) { |
Kees Cook | 0466bdb | 2017-08-11 13:12:11 -0700 | [diff] [blame] | 925 | case SECCOMP_RET_KILL_PROCESS: |
Kees Cook | fd76875 | 2017-08-11 12:53:18 -0700 | [diff] [blame] | 926 | case SECCOMP_RET_KILL_THREAD: |
Tyler Hicks | d612b1f | 2017-08-11 04:33:53 +0000 | [diff] [blame] | 927 | case SECCOMP_RET_TRAP: |
| 928 | case SECCOMP_RET_ERRNO: |
| 929 | case SECCOMP_RET_TRACE: |
Tyler Hicks | 59f5cf4 | 2017-08-11 04:33:57 +0000 | [diff] [blame] | 930 | case SECCOMP_RET_LOG: |
Tyler Hicks | d612b1f | 2017-08-11 04:33:53 +0000 | [diff] [blame] | 931 | case SECCOMP_RET_ALLOW: |
| 932 | break; |
| 933 | default: |
| 934 | return -EOPNOTSUPP; |
| 935 | } |
| 936 | |
| 937 | return 0; |
| 938 | } |
| 939 | |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 940 | /* Common entry point for both prctl and syscall. */ |
| 941 | static long do_seccomp(unsigned int op, unsigned int flags, |
| 942 | const char __user *uargs) |
| 943 | { |
| 944 | switch (op) { |
| 945 | case SECCOMP_SET_MODE_STRICT: |
| 946 | if (flags != 0 || uargs != NULL) |
| 947 | return -EINVAL; |
| 948 | return seccomp_set_mode_strict(); |
| 949 | case SECCOMP_SET_MODE_FILTER: |
| 950 | return seccomp_set_mode_filter(flags, uargs); |
Tyler Hicks | d612b1f | 2017-08-11 04:33:53 +0000 | [diff] [blame] | 951 | case SECCOMP_GET_ACTION_AVAIL: |
| 952 | if (flags != 0) |
| 953 | return -EINVAL; |
| 954 | |
| 955 | return seccomp_get_action_avail(uargs); |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 956 | default: |
| 957 | return -EINVAL; |
| 958 | } |
| 959 | } |
| 960 | |
| 961 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, |
| 962 | const char __user *, uargs) |
| 963 | { |
| 964 | return do_seccomp(op, flags, uargs); |
| 965 | } |
| 966 | |
Kees Cook | d78ab02 | 2014-05-21 15:02:11 -0700 | [diff] [blame] | 967 | /** |
| 968 | * prctl_set_seccomp: configures current->seccomp.mode |
| 969 | * @seccomp_mode: requested mode to use |
| 970 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER |
| 971 | * |
| 972 | * Returns 0 on success or -EINVAL on failure. |
| 973 | */ |
| 974 | long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter) |
| 975 | { |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 976 | unsigned int op; |
| 977 | char __user *uargs; |
| 978 | |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 979 | switch (seccomp_mode) { |
| 980 | case SECCOMP_MODE_STRICT: |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 981 | op = SECCOMP_SET_MODE_STRICT; |
| 982 | /* |
| 983 | * Setting strict mode through prctl always ignored filter, |
| 984 | * so make sure it is always NULL here to pass the internal |
| 985 | * check in do_seccomp(). |
| 986 | */ |
| 987 | uargs = NULL; |
| 988 | break; |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 989 | case SECCOMP_MODE_FILTER: |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 990 | op = SECCOMP_SET_MODE_FILTER; |
| 991 | uargs = filter; |
| 992 | break; |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 993 | default: |
| 994 | return -EINVAL; |
| 995 | } |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 996 | |
| 997 | /* prctl interface doesn't have flags, so they are always zero. */ |
| 998 | return do_seccomp(op, 0, uargs); |
Kees Cook | d78ab02 | 2014-05-21 15:02:11 -0700 | [diff] [blame] | 999 | } |
Tycho Andersen | f8e529e | 2015-10-27 09:23:59 +0900 | [diff] [blame] | 1000 | |
| 1001 | #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE) |
| 1002 | long seccomp_get_filter(struct task_struct *task, unsigned long filter_off, |
| 1003 | void __user *data) |
| 1004 | { |
| 1005 | struct seccomp_filter *filter; |
| 1006 | struct sock_fprog_kern *fprog; |
| 1007 | long ret; |
| 1008 | unsigned long count = 0; |
| 1009 | |
| 1010 | if (!capable(CAP_SYS_ADMIN) || |
| 1011 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { |
| 1012 | return -EACCES; |
| 1013 | } |
| 1014 | |
| 1015 | spin_lock_irq(&task->sighand->siglock); |
| 1016 | if (task->seccomp.mode != SECCOMP_MODE_FILTER) { |
| 1017 | ret = -EINVAL; |
| 1018 | goto out; |
| 1019 | } |
| 1020 | |
| 1021 | filter = task->seccomp.filter; |
| 1022 | while (filter) { |
| 1023 | filter = filter->prev; |
| 1024 | count++; |
| 1025 | } |
| 1026 | |
| 1027 | if (filter_off >= count) { |
| 1028 | ret = -ENOENT; |
| 1029 | goto out; |
| 1030 | } |
| 1031 | count -= filter_off; |
| 1032 | |
| 1033 | filter = task->seccomp.filter; |
| 1034 | while (filter && count > 1) { |
| 1035 | filter = filter->prev; |
| 1036 | count--; |
| 1037 | } |
| 1038 | |
| 1039 | if (WARN_ON(count != 1 || !filter)) { |
| 1040 | /* The filter tree shouldn't shrink while we're using it. */ |
| 1041 | ret = -ENOENT; |
| 1042 | goto out; |
| 1043 | } |
| 1044 | |
| 1045 | fprog = filter->prog->orig_prog; |
| 1046 | if (!fprog) { |
Mickaël Salaün | 470bf1f | 2016-03-24 02:46:33 +0100 | [diff] [blame] | 1047 | /* This must be a new non-cBPF filter, since we save |
Tycho Andersen | f8e529e | 2015-10-27 09:23:59 +0900 | [diff] [blame] | 1048 | * every cBPF filter's orig_prog above when |
| 1049 | * CONFIG_CHECKPOINT_RESTORE is enabled. |
| 1050 | */ |
| 1051 | ret = -EMEDIUMTYPE; |
| 1052 | goto out; |
| 1053 | } |
| 1054 | |
| 1055 | ret = fprog->len; |
| 1056 | if (!data) |
| 1057 | goto out; |
| 1058 | |
Oleg Nesterov | 66a733e | 2017-09-27 09:25:30 -0600 | [diff] [blame] | 1059 | __get_seccomp_filter(filter); |
Tycho Andersen | f8e529e | 2015-10-27 09:23:59 +0900 | [diff] [blame] | 1060 | spin_unlock_irq(&task->sighand->siglock); |
| 1061 | |
| 1062 | if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog))) |
| 1063 | ret = -EFAULT; |
| 1064 | |
Oleg Nesterov | 66a733e | 2017-09-27 09:25:30 -0600 | [diff] [blame] | 1065 | __put_seccomp_filter(filter); |
Tycho Andersen | f8e529e | 2015-10-27 09:23:59 +0900 | [diff] [blame] | 1066 | return ret; |
| 1067 | |
| 1068 | out: |
| 1069 | spin_unlock_irq(&task->sighand->siglock); |
| 1070 | return ret; |
| 1071 | } |
| 1072 | #endif |
Tyler Hicks | 8e5f1ad | 2017-08-11 04:33:52 +0000 | [diff] [blame] | 1073 | |
| 1074 | #ifdef CONFIG_SYSCTL |
| 1075 | |
| 1076 | /* Human readable action names for friendly sysctl interaction */ |
Kees Cook | 0466bdb | 2017-08-11 13:12:11 -0700 | [diff] [blame] | 1077 | #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process" |
Kees Cook | fd76875 | 2017-08-11 12:53:18 -0700 | [diff] [blame] | 1078 | #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread" |
Tyler Hicks | 8e5f1ad | 2017-08-11 04:33:52 +0000 | [diff] [blame] | 1079 | #define SECCOMP_RET_TRAP_NAME "trap" |
| 1080 | #define SECCOMP_RET_ERRNO_NAME "errno" |
| 1081 | #define SECCOMP_RET_TRACE_NAME "trace" |
Tyler Hicks | 59f5cf4 | 2017-08-11 04:33:57 +0000 | [diff] [blame] | 1082 | #define SECCOMP_RET_LOG_NAME "log" |
Tyler Hicks | 8e5f1ad | 2017-08-11 04:33:52 +0000 | [diff] [blame] | 1083 | #define SECCOMP_RET_ALLOW_NAME "allow" |
| 1084 | |
Kees Cook | fd76875 | 2017-08-11 12:53:18 -0700 | [diff] [blame] | 1085 | static const char seccomp_actions_avail[] = |
Kees Cook | 0466bdb | 2017-08-11 13:12:11 -0700 | [diff] [blame] | 1086 | SECCOMP_RET_KILL_PROCESS_NAME " " |
Kees Cook | fd76875 | 2017-08-11 12:53:18 -0700 | [diff] [blame] | 1087 | SECCOMP_RET_KILL_THREAD_NAME " " |
| 1088 | SECCOMP_RET_TRAP_NAME " " |
| 1089 | SECCOMP_RET_ERRNO_NAME " " |
| 1090 | SECCOMP_RET_TRACE_NAME " " |
| 1091 | SECCOMP_RET_LOG_NAME " " |
| 1092 | SECCOMP_RET_ALLOW_NAME; |
Tyler Hicks | 8e5f1ad | 2017-08-11 04:33:52 +0000 | [diff] [blame] | 1093 | |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 1094 | struct seccomp_log_name { |
| 1095 | u32 log; |
| 1096 | const char *name; |
| 1097 | }; |
| 1098 | |
| 1099 | static const struct seccomp_log_name seccomp_log_names[] = { |
Kees Cook | 0466bdb | 2017-08-11 13:12:11 -0700 | [diff] [blame] | 1100 | { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME }, |
Kees Cook | fd76875 | 2017-08-11 12:53:18 -0700 | [diff] [blame] | 1101 | { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME }, |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 1102 | { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME }, |
| 1103 | { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME }, |
| 1104 | { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME }, |
Tyler Hicks | 59f5cf4 | 2017-08-11 04:33:57 +0000 | [diff] [blame] | 1105 | { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME }, |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 1106 | { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME }, |
| 1107 | { } |
| 1108 | }; |
| 1109 | |
| 1110 | static bool seccomp_names_from_actions_logged(char *names, size_t size, |
| 1111 | u32 actions_logged) |
| 1112 | { |
| 1113 | const struct seccomp_log_name *cur; |
| 1114 | bool append_space = false; |
| 1115 | |
| 1116 | for (cur = seccomp_log_names; cur->name && size; cur++) { |
| 1117 | ssize_t ret; |
| 1118 | |
| 1119 | if (!(actions_logged & cur->log)) |
| 1120 | continue; |
| 1121 | |
| 1122 | if (append_space) { |
| 1123 | ret = strscpy(names, " ", size); |
| 1124 | if (ret < 0) |
| 1125 | return false; |
| 1126 | |
| 1127 | names += ret; |
| 1128 | size -= ret; |
| 1129 | } else |
| 1130 | append_space = true; |
| 1131 | |
| 1132 | ret = strscpy(names, cur->name, size); |
| 1133 | if (ret < 0) |
| 1134 | return false; |
| 1135 | |
| 1136 | names += ret; |
| 1137 | size -= ret; |
| 1138 | } |
| 1139 | |
| 1140 | return true; |
| 1141 | } |
| 1142 | |
| 1143 | static bool seccomp_action_logged_from_name(u32 *action_logged, |
| 1144 | const char *name) |
| 1145 | { |
| 1146 | const struct seccomp_log_name *cur; |
| 1147 | |
| 1148 | for (cur = seccomp_log_names; cur->name; cur++) { |
| 1149 | if (!strcmp(cur->name, name)) { |
| 1150 | *action_logged = cur->log; |
| 1151 | return true; |
| 1152 | } |
| 1153 | } |
| 1154 | |
| 1155 | return false; |
| 1156 | } |
| 1157 | |
| 1158 | static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names) |
| 1159 | { |
| 1160 | char *name; |
| 1161 | |
| 1162 | *actions_logged = 0; |
| 1163 | while ((name = strsep(&names, " ")) && *name) { |
| 1164 | u32 action_logged = 0; |
| 1165 | |
| 1166 | if (!seccomp_action_logged_from_name(&action_logged, name)) |
| 1167 | return false; |
| 1168 | |
| 1169 | *actions_logged |= action_logged; |
| 1170 | } |
| 1171 | |
| 1172 | return true; |
| 1173 | } |
| 1174 | |
| 1175 | static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write, |
| 1176 | void __user *buffer, size_t *lenp, |
| 1177 | loff_t *ppos) |
| 1178 | { |
| 1179 | char names[sizeof(seccomp_actions_avail)]; |
| 1180 | struct ctl_table table; |
| 1181 | int ret; |
| 1182 | |
| 1183 | if (write && !capable(CAP_SYS_ADMIN)) |
| 1184 | return -EPERM; |
| 1185 | |
| 1186 | memset(names, 0, sizeof(names)); |
| 1187 | |
| 1188 | if (!write) { |
| 1189 | if (!seccomp_names_from_actions_logged(names, sizeof(names), |
| 1190 | seccomp_actions_logged)) |
| 1191 | return -EINVAL; |
| 1192 | } |
| 1193 | |
| 1194 | table = *ro_table; |
| 1195 | table.data = names; |
| 1196 | table.maxlen = sizeof(names); |
| 1197 | ret = proc_dostring(&table, write, buffer, lenp, ppos); |
| 1198 | if (ret) |
| 1199 | return ret; |
| 1200 | |
| 1201 | if (write) { |
| 1202 | u32 actions_logged; |
| 1203 | |
| 1204 | if (!seccomp_actions_logged_from_names(&actions_logged, |
| 1205 | table.data)) |
| 1206 | return -EINVAL; |
| 1207 | |
| 1208 | if (actions_logged & SECCOMP_LOG_ALLOW) |
| 1209 | return -EINVAL; |
| 1210 | |
| 1211 | seccomp_actions_logged = actions_logged; |
| 1212 | } |
| 1213 | |
| 1214 | return 0; |
| 1215 | } |
| 1216 | |
Tyler Hicks | 8e5f1ad | 2017-08-11 04:33:52 +0000 | [diff] [blame] | 1217 | static struct ctl_path seccomp_sysctl_path[] = { |
| 1218 | { .procname = "kernel", }, |
| 1219 | { .procname = "seccomp", }, |
| 1220 | { } |
| 1221 | }; |
| 1222 | |
| 1223 | static struct ctl_table seccomp_sysctl_table[] = { |
| 1224 | { |
| 1225 | .procname = "actions_avail", |
| 1226 | .data = (void *) &seccomp_actions_avail, |
| 1227 | .maxlen = sizeof(seccomp_actions_avail), |
| 1228 | .mode = 0444, |
| 1229 | .proc_handler = proc_dostring, |
| 1230 | }, |
Tyler Hicks | 0ddec0f | 2017-08-11 04:33:54 +0000 | [diff] [blame] | 1231 | { |
| 1232 | .procname = "actions_logged", |
| 1233 | .mode = 0644, |
| 1234 | .proc_handler = seccomp_actions_logged_handler, |
| 1235 | }, |
Tyler Hicks | 8e5f1ad | 2017-08-11 04:33:52 +0000 | [diff] [blame] | 1236 | { } |
| 1237 | }; |
| 1238 | |
| 1239 | static int __init seccomp_sysctl_init(void) |
| 1240 | { |
| 1241 | struct ctl_table_header *hdr; |
| 1242 | |
| 1243 | hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table); |
| 1244 | if (!hdr) |
| 1245 | pr_warn("seccomp: sysctl registration failed\n"); |
| 1246 | else |
| 1247 | kmemleak_not_leak(hdr); |
| 1248 | |
| 1249 | return 0; |
| 1250 | } |
| 1251 | |
| 1252 | device_initcall(seccomp_sysctl_init) |
| 1253 | |
| 1254 | #endif /* CONFIG_SYSCTL */ |