Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/kernel/seccomp.c |
| 3 | * |
| 4 | * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com> |
| 5 | * |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 6 | * Copyright (C) 2012 Google, Inc. |
| 7 | * Will Drewry <wad@chromium.org> |
| 8 | * |
| 9 | * This defines a simple but solid secure-computing facility. |
| 10 | * |
| 11 | * Mode 1 uses a fixed list of allowed system calls. |
| 12 | * Mode 2 allows user-defined system call filters in the form |
| 13 | * of Berkeley Packet Filters/Linux Socket Filters. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | */ |
| 15 | |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 16 | #include <linux/atomic.h> |
Eric Paris | 85e7bac3 | 2012-01-03 14:23:05 -0500 | [diff] [blame] | 17 | #include <linux/audit.h> |
Roland McGrath | 5b10174 | 2009-02-27 23:25:54 -0800 | [diff] [blame] | 18 | #include <linux/compat.h> |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 19 | #include <linux/sched.h> |
| 20 | #include <linux/seccomp.h> |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 21 | #include <linux/slab.h> |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 22 | #include <linux/syscalls.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 24 | #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 25 | #include <asm/syscall.h> |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 26 | #endif |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 27 | |
| 28 | #ifdef CONFIG_SECCOMP_FILTER |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 29 | #include <linux/filter.h> |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 30 | #include <linux/pid.h> |
Will Drewry | fb0fadf | 2012-04-12 16:48:02 -0500 | [diff] [blame] | 31 | #include <linux/ptrace.h> |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 32 | #include <linux/security.h> |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 33 | #include <linux/tracehook.h> |
| 34 | #include <linux/uaccess.h> |
| 35 | |
| 36 | /** |
| 37 | * struct seccomp_filter - container for seccomp BPF programs |
| 38 | * |
| 39 | * @usage: reference count to manage the object lifetime. |
| 40 | * get/put helpers should be used when accessing an instance |
| 41 | * outside of a lifetime-guarded section. In general, this |
| 42 | * is only needed for handling filters shared across tasks. |
| 43 | * @prev: points to a previously installed, or inherited, filter |
| 44 | * @len: the number of instructions in the program |
Fabian Frederick | 119ce5c | 2014-06-06 14:37:53 -0700 | [diff] [blame] | 45 | * @insnsi: the BPF program instructions to evaluate |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 46 | * |
| 47 | * seccomp_filter objects are organized in a tree linked via the @prev |
| 48 | * pointer. For any task, it appears to be a singly-linked list starting |
| 49 | * with current->seccomp.filter, the most recently attached or inherited filter. |
| 50 | * However, multiple filters may share a @prev node, by way of fork(), which |
| 51 | * results in a unidirectional tree existing in memory. This is similar to |
| 52 | * how namespaces work. |
| 53 | * |
| 54 | * seccomp_filter objects should never be modified after being attached |
| 55 | * to a task_struct (other than @usage). |
| 56 | */ |
| 57 | struct seccomp_filter { |
| 58 | atomic_t usage; |
| 59 | struct seccomp_filter *prev; |
Alexei Starovoitov | 7ae457c | 2014-07-30 20:34:16 -0700 | [diff] [blame] | 60 | struct bpf_prog *prog; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 61 | }; |
| 62 | |
| 63 | /* Limit any path through the tree to 256KB worth of instructions. */ |
| 64 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) |
| 65 | |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 66 | /* |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 67 | * Endianness is explicitly ignored and left for BPF program authors to manage |
| 68 | * as per the specific architecture. |
| 69 | */ |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 70 | static void populate_seccomp_data(struct seccomp_data *sd) |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 71 | { |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 72 | struct task_struct *task = current; |
| 73 | struct pt_regs *regs = task_pt_regs(task); |
Daniel Borkmann | 2eac764 | 2014-04-14 21:02:59 +0200 | [diff] [blame] | 74 | unsigned long args[6]; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 75 | |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 76 | sd->nr = syscall_get_nr(task, regs); |
Linus Torvalds | 0b74717 | 2014-04-12 12:38:53 -0700 | [diff] [blame] | 77 | sd->arch = syscall_get_arch(); |
Daniel Borkmann | 2eac764 | 2014-04-14 21:02:59 +0200 | [diff] [blame] | 78 | syscall_get_arguments(task, regs, 0, 6, args); |
| 79 | sd->args[0] = args[0]; |
| 80 | sd->args[1] = args[1]; |
| 81 | sd->args[2] = args[2]; |
| 82 | sd->args[3] = args[3]; |
| 83 | sd->args[4] = args[4]; |
| 84 | sd->args[5] = args[5]; |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 85 | sd->instruction_pointer = KSTK_EIP(task); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 86 | } |
| 87 | |
| 88 | /** |
| 89 | * seccomp_check_filter - verify seccomp filter code |
| 90 | * @filter: filter to verify |
| 91 | * @flen: length of filter |
| 92 | * |
Alexei Starovoitov | 4df95ff | 2014-07-30 20:34:14 -0700 | [diff] [blame] | 93 | * Takes a previously checked filter (by bpf_check_classic) and |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 94 | * redirects all filter code that loads struct sk_buff data |
| 95 | * and related data through seccomp_bpf_load. It also |
| 96 | * enforces length and alignment checking of those loads. |
| 97 | * |
| 98 | * Returns 0 if the rule set is legal or -EINVAL if not. |
| 99 | */ |
| 100 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) |
| 101 | { |
| 102 | int pc; |
| 103 | for (pc = 0; pc < flen; pc++) { |
| 104 | struct sock_filter *ftest = &filter[pc]; |
| 105 | u16 code = ftest->code; |
| 106 | u32 k = ftest->k; |
| 107 | |
| 108 | switch (code) { |
Daniel Borkmann | 3480593 | 2014-05-29 10:22:50 +0200 | [diff] [blame] | 109 | case BPF_LD | BPF_W | BPF_ABS: |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 110 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 111 | /* 32-bit aligned and not out of bounds. */ |
| 112 | if (k >= sizeof(struct seccomp_data) || k & 3) |
| 113 | return -EINVAL; |
| 114 | continue; |
Daniel Borkmann | 3480593 | 2014-05-29 10:22:50 +0200 | [diff] [blame] | 115 | case BPF_LD | BPF_W | BPF_LEN: |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 116 | ftest->code = BPF_LD | BPF_IMM; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 117 | ftest->k = sizeof(struct seccomp_data); |
| 118 | continue; |
Daniel Borkmann | 3480593 | 2014-05-29 10:22:50 +0200 | [diff] [blame] | 119 | case BPF_LDX | BPF_W | BPF_LEN: |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 120 | ftest->code = BPF_LDX | BPF_IMM; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 121 | ftest->k = sizeof(struct seccomp_data); |
| 122 | continue; |
| 123 | /* Explicitly include allowed calls. */ |
Daniel Borkmann | 3480593 | 2014-05-29 10:22:50 +0200 | [diff] [blame] | 124 | case BPF_RET | BPF_K: |
| 125 | case BPF_RET | BPF_A: |
| 126 | case BPF_ALU | BPF_ADD | BPF_K: |
| 127 | case BPF_ALU | BPF_ADD | BPF_X: |
| 128 | case BPF_ALU | BPF_SUB | BPF_K: |
| 129 | case BPF_ALU | BPF_SUB | BPF_X: |
| 130 | case BPF_ALU | BPF_MUL | BPF_K: |
| 131 | case BPF_ALU | BPF_MUL | BPF_X: |
| 132 | case BPF_ALU | BPF_DIV | BPF_K: |
| 133 | case BPF_ALU | BPF_DIV | BPF_X: |
| 134 | case BPF_ALU | BPF_AND | BPF_K: |
| 135 | case BPF_ALU | BPF_AND | BPF_X: |
| 136 | case BPF_ALU | BPF_OR | BPF_K: |
| 137 | case BPF_ALU | BPF_OR | BPF_X: |
| 138 | case BPF_ALU | BPF_XOR | BPF_K: |
| 139 | case BPF_ALU | BPF_XOR | BPF_X: |
| 140 | case BPF_ALU | BPF_LSH | BPF_K: |
| 141 | case BPF_ALU | BPF_LSH | BPF_X: |
| 142 | case BPF_ALU | BPF_RSH | BPF_K: |
| 143 | case BPF_ALU | BPF_RSH | BPF_X: |
| 144 | case BPF_ALU | BPF_NEG: |
| 145 | case BPF_LD | BPF_IMM: |
| 146 | case BPF_LDX | BPF_IMM: |
| 147 | case BPF_MISC | BPF_TAX: |
| 148 | case BPF_MISC | BPF_TXA: |
| 149 | case BPF_LD | BPF_MEM: |
| 150 | case BPF_LDX | BPF_MEM: |
| 151 | case BPF_ST: |
| 152 | case BPF_STX: |
| 153 | case BPF_JMP | BPF_JA: |
| 154 | case BPF_JMP | BPF_JEQ | BPF_K: |
| 155 | case BPF_JMP | BPF_JEQ | BPF_X: |
| 156 | case BPF_JMP | BPF_JGE | BPF_K: |
| 157 | case BPF_JMP | BPF_JGE | BPF_X: |
| 158 | case BPF_JMP | BPF_JGT | BPF_K: |
| 159 | case BPF_JMP | BPF_JGT | BPF_X: |
| 160 | case BPF_JMP | BPF_JSET | BPF_K: |
| 161 | case BPF_JMP | BPF_JSET | BPF_X: |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 162 | continue; |
| 163 | default: |
| 164 | return -EINVAL; |
| 165 | } |
| 166 | } |
| 167 | return 0; |
| 168 | } |
| 169 | |
| 170 | /** |
| 171 | * seccomp_run_filters - evaluates all seccomp filters against @syscall |
| 172 | * @syscall: number of the current system call |
| 173 | * |
| 174 | * Returns valid seccomp BPF response codes. |
| 175 | */ |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 176 | static u32 seccomp_run_filters(struct seccomp_data *sd) |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 177 | { |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 178 | struct seccomp_data sd_local; |
Will Drewry | acf3b2c | 2012-04-12 16:47:59 -0500 | [diff] [blame] | 179 | u32 ret = SECCOMP_RET_ALLOW; |
Pranith Kumar | 8225d38 | 2014-11-21 10:06:01 -0500 | [diff] [blame] | 180 | /* Make sure cross-thread synced filter points somewhere sane. */ |
| 181 | struct seccomp_filter *f = |
| 182 | lockless_dereference(current->seccomp.filter); |
Will Drewry | acf3b2c | 2012-04-12 16:47:59 -0500 | [diff] [blame] | 183 | |
| 184 | /* Ensure unexpected behavior doesn't result in failing open. */ |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 185 | if (unlikely(WARN_ON(f == NULL))) |
Will Drewry | acf3b2c | 2012-04-12 16:47:59 -0500 | [diff] [blame] | 186 | return SECCOMP_RET_KILL; |
| 187 | |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 188 | if (!sd) { |
| 189 | populate_seccomp_data(&sd_local); |
| 190 | sd = &sd_local; |
| 191 | } |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 192 | |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 193 | /* |
| 194 | * All filters in the list are evaluated and the lowest BPF return |
Will Drewry | acf3b2c | 2012-04-12 16:47:59 -0500 | [diff] [blame] | 195 | * value always takes priority (ignoring the DATA). |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 196 | */ |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 197 | for (; f; f = f->prev) { |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 198 | u32 cur_ret = BPF_PROG_RUN(f->prog, (void *)sd); |
Alexei Starovoitov | 8f577ca | 2014-05-13 19:50:47 -0700 | [diff] [blame] | 199 | |
Will Drewry | acf3b2c | 2012-04-12 16:47:59 -0500 | [diff] [blame] | 200 | if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION)) |
| 201 | ret = cur_ret; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 202 | } |
| 203 | return ret; |
| 204 | } |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 205 | #endif /* CONFIG_SECCOMP_FILTER */ |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 206 | |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 207 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
| 208 | { |
Guenter Roeck | 69f6a34 | 2014-08-10 20:50:30 -0700 | [diff] [blame] | 209 | assert_spin_locked(¤t->sighand->siglock); |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 210 | |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 211 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
| 212 | return false; |
| 213 | |
| 214 | return true; |
| 215 | } |
| 216 | |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 217 | static inline void seccomp_assign_mode(struct task_struct *task, |
| 218 | unsigned long seccomp_mode) |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 219 | { |
Guenter Roeck | 69f6a34 | 2014-08-10 20:50:30 -0700 | [diff] [blame] | 220 | assert_spin_locked(&task->sighand->siglock); |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 221 | |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 222 | task->seccomp.mode = seccomp_mode; |
| 223 | /* |
| 224 | * Make sure TIF_SECCOMP cannot be set before the mode (and |
| 225 | * filter) is set. |
| 226 | */ |
| 227 | smp_mb__before_atomic(); |
| 228 | set_tsk_thread_flag(task, TIF_SECCOMP); |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 229 | } |
| 230 | |
| 231 | #ifdef CONFIG_SECCOMP_FILTER |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 232 | /* Returns 1 if the parent is an ancestor of the child. */ |
| 233 | static int is_ancestor(struct seccomp_filter *parent, |
| 234 | struct seccomp_filter *child) |
| 235 | { |
| 236 | /* NULL is the root ancestor. */ |
| 237 | if (parent == NULL) |
| 238 | return 1; |
| 239 | for (; child; child = child->prev) |
| 240 | if (child == parent) |
| 241 | return 1; |
| 242 | return 0; |
| 243 | } |
| 244 | |
| 245 | /** |
| 246 | * seccomp_can_sync_threads: checks if all threads can be synchronized |
| 247 | * |
| 248 | * Expects sighand and cred_guard_mutex locks to be held. |
| 249 | * |
| 250 | * Returns 0 on success, -ve on error, or the pid of a thread which was |
| 251 | * either not in the correct seccomp mode or it did not have an ancestral |
| 252 | * seccomp filter. |
| 253 | */ |
| 254 | static inline pid_t seccomp_can_sync_threads(void) |
| 255 | { |
| 256 | struct task_struct *thread, *caller; |
| 257 | |
| 258 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); |
Guenter Roeck | 69f6a34 | 2014-08-10 20:50:30 -0700 | [diff] [blame] | 259 | assert_spin_locked(¤t->sighand->siglock); |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 260 | |
| 261 | /* Validate all threads being eligible for synchronization. */ |
| 262 | caller = current; |
| 263 | for_each_thread(caller, thread) { |
| 264 | pid_t failed; |
| 265 | |
| 266 | /* Skip current, since it is initiating the sync. */ |
| 267 | if (thread == caller) |
| 268 | continue; |
| 269 | |
| 270 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || |
| 271 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && |
| 272 | is_ancestor(thread->seccomp.filter, |
| 273 | caller->seccomp.filter))) |
| 274 | continue; |
| 275 | |
| 276 | /* Return the first thread that cannot be synchronized. */ |
| 277 | failed = task_pid_vnr(thread); |
| 278 | /* If the pid cannot be resolved, then return -ESRCH */ |
| 279 | if (unlikely(WARN_ON(failed == 0))) |
| 280 | failed = -ESRCH; |
| 281 | return failed; |
| 282 | } |
| 283 | |
| 284 | return 0; |
| 285 | } |
| 286 | |
| 287 | /** |
| 288 | * seccomp_sync_threads: sets all threads to use current's filter |
| 289 | * |
| 290 | * Expects sighand and cred_guard_mutex locks to be held, and for |
| 291 | * seccomp_can_sync_threads() to have returned success already |
| 292 | * without dropping the locks. |
| 293 | * |
| 294 | */ |
| 295 | static inline void seccomp_sync_threads(void) |
| 296 | { |
| 297 | struct task_struct *thread, *caller; |
| 298 | |
| 299 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); |
Guenter Roeck | 69f6a34 | 2014-08-10 20:50:30 -0700 | [diff] [blame] | 300 | assert_spin_locked(¤t->sighand->siglock); |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 301 | |
| 302 | /* Synchronize all threads. */ |
| 303 | caller = current; |
| 304 | for_each_thread(caller, thread) { |
| 305 | /* Skip current, since it needs no changes. */ |
| 306 | if (thread == caller) |
| 307 | continue; |
| 308 | |
| 309 | /* Get a task reference for the new leaf node. */ |
| 310 | get_seccomp_filter(caller); |
| 311 | /* |
| 312 | * Drop the task reference to the shared ancestor since |
| 313 | * current's path will hold a reference. (This also |
| 314 | * allows a put before the assignment.) |
| 315 | */ |
| 316 | put_seccomp_filter(thread); |
| 317 | smp_store_release(&thread->seccomp.filter, |
| 318 | caller->seccomp.filter); |
| 319 | /* |
| 320 | * Opt the other thread into seccomp if needed. |
| 321 | * As threads are considered to be trust-realm |
| 322 | * equivalent (see ptrace_may_access), it is safe to |
| 323 | * allow one thread to transition the other. |
| 324 | */ |
| 325 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) { |
| 326 | /* |
| 327 | * Don't let an unprivileged task work around |
| 328 | * the no_new_privs restriction by creating |
| 329 | * a thread that sets it up, enters seccomp, |
| 330 | * then dies. |
| 331 | */ |
| 332 | if (task_no_new_privs(caller)) |
| 333 | task_set_no_new_privs(thread); |
| 334 | |
| 335 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER); |
| 336 | } |
| 337 | } |
| 338 | } |
| 339 | |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 340 | /** |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 341 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 342 | * @fprog: BPF program to install |
| 343 | * |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 344 | * Returns filter on success or an ERR_PTR on failure. |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 345 | */ |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 346 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 347 | { |
Daniel Borkmann | ac67eb2 | 2015-05-06 16:12:30 +0200 | [diff] [blame] | 348 | struct seccomp_filter *sfilter; |
| 349 | int ret; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 350 | |
| 351 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 352 | return ERR_PTR(-EINVAL); |
Nicolas Schichan | d9e12f42e | 2015-05-06 16:12:28 +0200 | [diff] [blame] | 353 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 354 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 355 | |
| 356 | /* |
Fabian Frederick | 119ce5c | 2014-06-06 14:37:53 -0700 | [diff] [blame] | 357 | * Installing a seccomp filter requires that the task has |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 358 | * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. |
| 359 | * This avoids scenarios where unprivileged tasks can affect the |
| 360 | * behavior of privileged children. |
| 361 | */ |
Kees Cook | 1d4457f | 2014-05-21 15:23:46 -0700 | [diff] [blame] | 362 | if (!task_no_new_privs(current) && |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 363 | security_capable_noaudit(current_cred(), current_user_ns(), |
| 364 | CAP_SYS_ADMIN) != 0) |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 365 | return ERR_PTR(-EACCES); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 366 | |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 367 | /* Allocate a new seccomp_filter */ |
Daniel Borkmann | ac67eb2 | 2015-05-06 16:12:30 +0200 | [diff] [blame] | 368 | sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN); |
| 369 | if (!sfilter) |
Nicolas Schichan | d9e12f42e | 2015-05-06 16:12:28 +0200 | [diff] [blame] | 370 | return ERR_PTR(-ENOMEM); |
Daniel Borkmann | ac67eb2 | 2015-05-06 16:12:30 +0200 | [diff] [blame] | 371 | |
| 372 | ret = bpf_prog_create_from_user(&sfilter->prog, fprog, |
| 373 | seccomp_check_filter); |
| 374 | if (ret < 0) { |
| 375 | kfree(sfilter); |
| 376 | return ERR_PTR(ret); |
Nicolas Schichan | d9e12f42e | 2015-05-06 16:12:28 +0200 | [diff] [blame] | 377 | } |
Alexei Starovoitov | bd4cf0e | 2014-03-28 18:58:25 +0100 | [diff] [blame] | 378 | |
Daniel Borkmann | ac67eb2 | 2015-05-06 16:12:30 +0200 | [diff] [blame] | 379 | atomic_set(&sfilter->usage, 1); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 380 | |
Daniel Borkmann | ac67eb2 | 2015-05-06 16:12:30 +0200 | [diff] [blame] | 381 | return sfilter; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 382 | } |
| 383 | |
| 384 | /** |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 385 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 386 | * @user_filter: pointer to the user data containing a sock_fprog. |
| 387 | * |
| 388 | * Returns 0 on success and non-zero otherwise. |
| 389 | */ |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 390 | static struct seccomp_filter * |
| 391 | seccomp_prepare_user_filter(const char __user *user_filter) |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 392 | { |
| 393 | struct sock_fprog fprog; |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 394 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 395 | |
| 396 | #ifdef CONFIG_COMPAT |
| 397 | if (is_compat_task()) { |
| 398 | struct compat_sock_fprog fprog32; |
| 399 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) |
| 400 | goto out; |
| 401 | fprog.len = fprog32.len; |
| 402 | fprog.filter = compat_ptr(fprog32.filter); |
| 403 | } else /* falls through to the if below. */ |
| 404 | #endif |
| 405 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) |
| 406 | goto out; |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 407 | filter = seccomp_prepare_filter(&fprog); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 408 | out: |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 409 | return filter; |
| 410 | } |
| 411 | |
| 412 | /** |
| 413 | * seccomp_attach_filter: validate and attach filter |
| 414 | * @flags: flags to change filter behavior |
| 415 | * @filter: seccomp filter to add to the current process |
| 416 | * |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 417 | * Caller must be holding current->sighand->siglock lock. |
| 418 | * |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 419 | * Returns 0 on success, -ve on error. |
| 420 | */ |
| 421 | static long seccomp_attach_filter(unsigned int flags, |
| 422 | struct seccomp_filter *filter) |
| 423 | { |
| 424 | unsigned long total_insns; |
| 425 | struct seccomp_filter *walker; |
| 426 | |
Guenter Roeck | 69f6a34 | 2014-08-10 20:50:30 -0700 | [diff] [blame] | 427 | assert_spin_locked(¤t->sighand->siglock); |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 428 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 429 | /* Validate resulting filter length. */ |
| 430 | total_insns = filter->prog->len; |
| 431 | for (walker = current->seccomp.filter; walker; walker = walker->prev) |
| 432 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ |
| 433 | if (total_insns > MAX_INSNS_PER_PATH) |
| 434 | return -ENOMEM; |
| 435 | |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 436 | /* If thread sync has been requested, check that it is possible. */ |
| 437 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { |
| 438 | int ret; |
| 439 | |
| 440 | ret = seccomp_can_sync_threads(); |
| 441 | if (ret) |
| 442 | return ret; |
| 443 | } |
| 444 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 445 | /* |
| 446 | * If there is an existing filter, make it the prev and don't drop its |
| 447 | * task reference. |
| 448 | */ |
| 449 | filter->prev = current->seccomp.filter; |
| 450 | current->seccomp.filter = filter; |
| 451 | |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 452 | /* Now that the new filter is in place, synchronize to all threads. */ |
| 453 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
| 454 | seccomp_sync_threads(); |
| 455 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 456 | return 0; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 457 | } |
| 458 | |
| 459 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ |
| 460 | void get_seccomp_filter(struct task_struct *tsk) |
| 461 | { |
| 462 | struct seccomp_filter *orig = tsk->seccomp.filter; |
| 463 | if (!orig) |
| 464 | return; |
| 465 | /* Reference count is bounded by the number of total processes. */ |
| 466 | atomic_inc(&orig->usage); |
| 467 | } |
| 468 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 469 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
| 470 | { |
| 471 | if (filter) { |
Linus Torvalds | ae045e2 | 2014-08-06 09:38:14 -0700 | [diff] [blame] | 472 | bpf_prog_free(filter->prog); |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 473 | kfree(filter); |
| 474 | } |
| 475 | } |
| 476 | |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 477 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
| 478 | void put_seccomp_filter(struct task_struct *tsk) |
| 479 | { |
| 480 | struct seccomp_filter *orig = tsk->seccomp.filter; |
| 481 | /* Clean up single-reference branches iteratively. */ |
| 482 | while (orig && atomic_dec_and_test(&orig->usage)) { |
| 483 | struct seccomp_filter *freeme = orig; |
| 484 | orig = orig->prev; |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 485 | seccomp_filter_free(freeme); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 486 | } |
| 487 | } |
Will Drewry | bb6ea43 | 2012-04-12 16:48:01 -0500 | [diff] [blame] | 488 | |
| 489 | /** |
| 490 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation |
| 491 | * @syscall: syscall number to send to userland |
| 492 | * @reason: filter-supplied reason code to send to userland (via si_errno) |
| 493 | * |
| 494 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. |
| 495 | */ |
| 496 | static void seccomp_send_sigsys(int syscall, int reason) |
| 497 | { |
| 498 | struct siginfo info; |
| 499 | memset(&info, 0, sizeof(info)); |
| 500 | info.si_signo = SIGSYS; |
| 501 | info.si_code = SYS_SECCOMP; |
| 502 | info.si_call_addr = (void __user *)KSTK_EIP(current); |
| 503 | info.si_errno = reason; |
Eric Paris | 5e937a9 | 2014-03-11 12:48:43 -0400 | [diff] [blame] | 504 | info.si_arch = syscall_get_arch(); |
Will Drewry | bb6ea43 | 2012-04-12 16:48:01 -0500 | [diff] [blame] | 505 | info.si_syscall = syscall; |
| 506 | force_sig_info(SIGSYS, &info, current); |
| 507 | } |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 508 | #endif /* CONFIG_SECCOMP_FILTER */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 509 | |
| 510 | /* |
| 511 | * Secure computing mode 1 allows only read/write/exit/sigreturn. |
| 512 | * To be fully secure this must be combined with rlimit |
| 513 | * to limit the stack allocations too. |
| 514 | */ |
| 515 | static int mode1_syscalls[] = { |
| 516 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, |
| 517 | 0, /* null terminated */ |
| 518 | }; |
| 519 | |
Roland McGrath | 5b10174 | 2009-02-27 23:25:54 -0800 | [diff] [blame] | 520 | #ifdef CONFIG_COMPAT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 521 | static int mode1_syscalls_32[] = { |
| 522 | __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32, |
| 523 | 0, /* null terminated */ |
| 524 | }; |
| 525 | #endif |
| 526 | |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 527 | static void __secure_computing_strict(int this_syscall) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 528 | { |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 529 | int *syscall_whitelist = mode1_syscalls; |
| 530 | #ifdef CONFIG_COMPAT |
| 531 | if (is_compat_task()) |
| 532 | syscall_whitelist = mode1_syscalls_32; |
| 533 | #endif |
| 534 | do { |
| 535 | if (*syscall_whitelist == this_syscall) |
| 536 | return; |
| 537 | } while (*++syscall_whitelist); |
| 538 | |
| 539 | #ifdef SECCOMP_DEBUG |
| 540 | dump_stack(); |
| 541 | #endif |
| 542 | audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL); |
| 543 | do_exit(SIGKILL); |
| 544 | } |
| 545 | |
| 546 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
| 547 | void secure_computing_strict(int this_syscall) |
| 548 | { |
| 549 | int mode = current->seccomp.mode; |
| 550 | |
Tycho Andersen | 13c4a90 | 2015-06-13 09:02:48 -0600 | [diff] [blame] | 551 | if (config_enabled(CONFIG_CHECKPOINT_RESTORE) && |
| 552 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
| 553 | return; |
| 554 | |
Kees Cook | 221272f | 2015-06-15 15:29:16 -0700 | [diff] [blame] | 555 | if (mode == SECCOMP_MODE_DISABLED) |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 556 | return; |
| 557 | else if (mode == SECCOMP_MODE_STRICT) |
| 558 | __secure_computing_strict(this_syscall); |
| 559 | else |
| 560 | BUG(); |
| 561 | } |
| 562 | #else |
| 563 | int __secure_computing(void) |
| 564 | { |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 565 | u32 phase1_result = seccomp_phase1(NULL); |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 566 | |
| 567 | if (likely(phase1_result == SECCOMP_PHASE1_OK)) |
| 568 | return 0; |
| 569 | else if (likely(phase1_result == SECCOMP_PHASE1_SKIP)) |
| 570 | return -1; |
| 571 | else |
| 572 | return seccomp_phase2(phase1_result); |
| 573 | } |
| 574 | |
| 575 | #ifdef CONFIG_SECCOMP_FILTER |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 576 | static u32 __seccomp_phase1_filter(int this_syscall, struct seccomp_data *sd) |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 577 | { |
| 578 | u32 filter_ret, action; |
| 579 | int data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 580 | |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 581 | /* |
| 582 | * Make sure that any changes to mode from another thread have |
| 583 | * been seen after TIF_SECCOMP was seen. |
| 584 | */ |
| 585 | rmb(); |
| 586 | |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 587 | filter_ret = seccomp_run_filters(sd); |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 588 | data = filter_ret & SECCOMP_RET_DATA; |
| 589 | action = filter_ret & SECCOMP_RET_ACTION; |
Andy Lutomirski | 87b526d | 2012-10-01 11:40:45 -0700 | [diff] [blame] | 590 | |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 591 | switch (action) { |
| 592 | case SECCOMP_RET_ERRNO: |
Kees Cook | 580c57f | 2015-02-17 13:48:00 -0800 | [diff] [blame] | 593 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
| 594 | if (data > MAX_ERRNO) |
| 595 | data = MAX_ERRNO; |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 596 | syscall_set_return_value(current, task_pt_regs(current), |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 597 | -data, 0); |
| 598 | goto skip; |
| 599 | |
| 600 | case SECCOMP_RET_TRAP: |
| 601 | /* Show the handler the original registers. */ |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 602 | syscall_rollback(current, task_pt_regs(current)); |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 603 | /* Let the filter pass back 16 bits of data. */ |
| 604 | seccomp_send_sigsys(this_syscall, data); |
| 605 | goto skip; |
| 606 | |
| 607 | case SECCOMP_RET_TRACE: |
| 608 | return filter_ret; /* Save the rest for phase 2. */ |
| 609 | |
| 610 | case SECCOMP_RET_ALLOW: |
| 611 | return SECCOMP_PHASE1_OK; |
| 612 | |
| 613 | case SECCOMP_RET_KILL: |
| 614 | default: |
| 615 | audit_seccomp(this_syscall, SIGSYS, action); |
| 616 | do_exit(SIGSYS); |
Will Drewry | 8156b45 | 2012-04-17 14:48:58 -0500 | [diff] [blame] | 617 | } |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 618 | |
| 619 | unreachable(); |
| 620 | |
| 621 | skip: |
| 622 | audit_seccomp(this_syscall, 0, action); |
| 623 | return SECCOMP_PHASE1_SKIP; |
| 624 | } |
| 625 | #endif |
| 626 | |
| 627 | /** |
| 628 | * seccomp_phase1() - run fast path seccomp checks on the current syscall |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 629 | * @arg sd: The seccomp_data or NULL |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 630 | * |
| 631 | * This only reads pt_regs via the syscall_xyz helpers. The only change |
| 632 | * it will make to pt_regs is via syscall_set_return_value, and it will |
| 633 | * only do that if it returns SECCOMP_PHASE1_SKIP. |
| 634 | * |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 635 | * If sd is provided, it will not read pt_regs at all. |
| 636 | * |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 637 | * It may also call do_exit or force a signal; these actions must be |
| 638 | * safe. |
| 639 | * |
| 640 | * If it returns SECCOMP_PHASE1_OK, the syscall passes checks and should |
| 641 | * be processed normally. |
| 642 | * |
| 643 | * If it returns SECCOMP_PHASE1_SKIP, then the syscall should not be |
| 644 | * invoked. In this case, seccomp_phase1 will have set the return value |
| 645 | * using syscall_set_return_value. |
| 646 | * |
| 647 | * If it returns anything else, then the return value should be passed |
| 648 | * to seccomp_phase2 from a context in which ptrace hooks are safe. |
| 649 | */ |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 650 | u32 seccomp_phase1(struct seccomp_data *sd) |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 651 | { |
| 652 | int mode = current->seccomp.mode; |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 653 | int this_syscall = sd ? sd->nr : |
| 654 | syscall_get_nr(current, task_pt_regs(current)); |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 655 | |
Tycho Andersen | 13c4a90 | 2015-06-13 09:02:48 -0600 | [diff] [blame] | 656 | if (config_enabled(CONFIG_CHECKPOINT_RESTORE) && |
| 657 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
| 658 | return SECCOMP_PHASE1_OK; |
| 659 | |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 660 | switch (mode) { |
| 661 | case SECCOMP_MODE_STRICT: |
| 662 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
| 663 | return SECCOMP_PHASE1_OK; |
| 664 | #ifdef CONFIG_SECCOMP_FILTER |
| 665 | case SECCOMP_MODE_FILTER: |
Andy Lutomirski | d39bd00 | 2014-07-21 18:49:16 -0700 | [diff] [blame] | 666 | return __seccomp_phase1_filter(this_syscall, sd); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 667 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | default: |
| 669 | BUG(); |
| 670 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 671 | } |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 672 | |
Andy Lutomirski | 13aa72f | 2014-07-21 18:49:15 -0700 | [diff] [blame] | 673 | /** |
| 674 | * seccomp_phase2() - finish slow path seccomp work for the current syscall |
| 675 | * @phase1_result: The return value from seccomp_phase1() |
| 676 | * |
| 677 | * This must be called from a context in which ptrace hooks can be used. |
| 678 | * |
| 679 | * Returns 0 if the syscall should be processed or -1 to skip the syscall. |
| 680 | */ |
| 681 | int seccomp_phase2(u32 phase1_result) |
| 682 | { |
| 683 | struct pt_regs *regs = task_pt_regs(current); |
| 684 | u32 action = phase1_result & SECCOMP_RET_ACTION; |
| 685 | int data = phase1_result & SECCOMP_RET_DATA; |
| 686 | |
| 687 | BUG_ON(action != SECCOMP_RET_TRACE); |
| 688 | |
| 689 | audit_seccomp(syscall_get_nr(current, regs), 0, action); |
| 690 | |
| 691 | /* Skip these calls if there is no tracer. */ |
| 692 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { |
| 693 | syscall_set_return_value(current, regs, |
| 694 | -ENOSYS, 0); |
| 695 | return -1; |
| 696 | } |
| 697 | |
| 698 | /* Allow the BPF to provide the event message */ |
| 699 | ptrace_event(PTRACE_EVENT_SECCOMP, data); |
| 700 | /* |
| 701 | * The delivery of a fatal signal during event |
| 702 | * notification may silently skip tracer notification. |
| 703 | * Terminating the task now avoids executing a system |
| 704 | * call that may not be intended. |
| 705 | */ |
| 706 | if (fatal_signal_pending(current)) |
| 707 | do_exit(SIGSYS); |
| 708 | if (syscall_get_nr(current, regs) < 0) |
| 709 | return -1; /* Explicit request to skip. */ |
| 710 | |
| 711 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 712 | } |
Andy Lutomirski | a4412fc | 2014-07-21 18:49:14 -0700 | [diff] [blame] | 713 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 714 | |
| 715 | long prctl_get_seccomp(void) |
| 716 | { |
| 717 | return current->seccomp.mode; |
| 718 | } |
| 719 | |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 720 | /** |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 721 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 722 | * |
| 723 | * Once current->seccomp.mode is non-zero, it may not be changed. |
| 724 | * |
| 725 | * Returns 0 on success or -EINVAL on failure. |
| 726 | */ |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 727 | static long seccomp_set_mode_strict(void) |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 728 | { |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 729 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 730 | long ret = -EINVAL; |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 731 | |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 732 | spin_lock_irq(¤t->sighand->siglock); |
| 733 | |
Kees Cook | 1f41b450 | 2014-06-25 15:38:02 -0700 | [diff] [blame] | 734 | if (!seccomp_may_assign_mode(seccomp_mode)) |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 735 | goto out; |
| 736 | |
Andrea Arcangeli | cf99aba | 2007-07-15 23:41:33 -0700 | [diff] [blame] | 737 | #ifdef TIF_NOTSC |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 738 | disable_TSC(); |
Andrea Arcangeli | cf99aba | 2007-07-15 23:41:33 -0700 | [diff] [blame] | 739 | #endif |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 740 | seccomp_assign_mode(current, seccomp_mode); |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 741 | ret = 0; |
| 742 | |
| 743 | out: |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 744 | spin_unlock_irq(¤t->sighand->siglock); |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 745 | |
| 746 | return ret; |
| 747 | } |
| 748 | |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 749 | #ifdef CONFIG_SECCOMP_FILTER |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 750 | /** |
| 751 | * seccomp_set_mode_filter: internal function for setting seccomp filter |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 752 | * @flags: flags to change filter behavior |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 753 | * @filter: struct sock_fprog containing filter |
| 754 | * |
| 755 | * This function may be called repeatedly to install additional filters. |
| 756 | * Every filter successfully installed will be evaluated (in reverse order) |
| 757 | * for each system call the task makes. |
| 758 | * |
| 759 | * Once current->seccomp.mode is non-zero, it may not be changed. |
| 760 | * |
| 761 | * Returns 0 on success or -EINVAL on failure. |
| 762 | */ |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 763 | static long seccomp_set_mode_filter(unsigned int flags, |
| 764 | const char __user *filter) |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 765 | { |
| 766 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 767 | struct seccomp_filter *prepared = NULL; |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 768 | long ret = -EINVAL; |
| 769 | |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 770 | /* Validate flags. */ |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 771 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 772 | return -EINVAL; |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 773 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 774 | /* Prepare the new filter before holding any locks. */ |
| 775 | prepared = seccomp_prepare_user_filter(filter); |
| 776 | if (IS_ERR(prepared)) |
| 777 | return PTR_ERR(prepared); |
| 778 | |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 779 | /* |
| 780 | * Make sure we cannot change seccomp or nnp state via TSYNC |
| 781 | * while another thread is in the middle of calling exec. |
| 782 | */ |
| 783 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && |
| 784 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) |
| 785 | goto out_free; |
| 786 | |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 787 | spin_lock_irq(¤t->sighand->siglock); |
| 788 | |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 789 | if (!seccomp_may_assign_mode(seccomp_mode)) |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 790 | goto out; |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 791 | |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 792 | ret = seccomp_attach_filter(flags, prepared); |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 793 | if (ret) |
| 794 | goto out; |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 795 | /* Do not free the successfully attached filter. */ |
| 796 | prepared = NULL; |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 797 | |
Kees Cook | 3ba2530 | 2014-06-27 15:01:35 -0700 | [diff] [blame] | 798 | seccomp_assign_mode(current, seccomp_mode); |
Will Drewry | e2cfabdf | 2012-04-12 16:47:57 -0500 | [diff] [blame] | 799 | out: |
Kees Cook | dbd95212 | 2014-06-27 15:18:48 -0700 | [diff] [blame] | 800 | spin_unlock_irq(¤t->sighand->siglock); |
Kees Cook | c2e1f2e | 2014-06-05 00:23:17 -0700 | [diff] [blame] | 801 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
| 802 | mutex_unlock(¤t->signal->cred_guard_mutex); |
| 803 | out_free: |
Kees Cook | c8bee43 | 2014-06-27 15:16:33 -0700 | [diff] [blame] | 804 | seccomp_filter_free(prepared); |
Andrea Arcangeli | 1d9d02f | 2007-07-15 23:41:32 -0700 | [diff] [blame] | 805 | return ret; |
| 806 | } |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 807 | #else |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 808 | static inline long seccomp_set_mode_filter(unsigned int flags, |
| 809 | const char __user *filter) |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 810 | { |
| 811 | return -EINVAL; |
| 812 | } |
| 813 | #endif |
Kees Cook | d78ab02 | 2014-05-21 15:02:11 -0700 | [diff] [blame] | 814 | |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 815 | /* Common entry point for both prctl and syscall. */ |
| 816 | static long do_seccomp(unsigned int op, unsigned int flags, |
| 817 | const char __user *uargs) |
| 818 | { |
| 819 | switch (op) { |
| 820 | case SECCOMP_SET_MODE_STRICT: |
| 821 | if (flags != 0 || uargs != NULL) |
| 822 | return -EINVAL; |
| 823 | return seccomp_set_mode_strict(); |
| 824 | case SECCOMP_SET_MODE_FILTER: |
| 825 | return seccomp_set_mode_filter(flags, uargs); |
| 826 | default: |
| 827 | return -EINVAL; |
| 828 | } |
| 829 | } |
| 830 | |
| 831 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, |
| 832 | const char __user *, uargs) |
| 833 | { |
| 834 | return do_seccomp(op, flags, uargs); |
| 835 | } |
| 836 | |
Kees Cook | d78ab02 | 2014-05-21 15:02:11 -0700 | [diff] [blame] | 837 | /** |
| 838 | * prctl_set_seccomp: configures current->seccomp.mode |
| 839 | * @seccomp_mode: requested mode to use |
| 840 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER |
| 841 | * |
| 842 | * Returns 0 on success or -EINVAL on failure. |
| 843 | */ |
| 844 | long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter) |
| 845 | { |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 846 | unsigned int op; |
| 847 | char __user *uargs; |
| 848 | |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 849 | switch (seccomp_mode) { |
| 850 | case SECCOMP_MODE_STRICT: |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 851 | op = SECCOMP_SET_MODE_STRICT; |
| 852 | /* |
| 853 | * Setting strict mode through prctl always ignored filter, |
| 854 | * so make sure it is always NULL here to pass the internal |
| 855 | * check in do_seccomp(). |
| 856 | */ |
| 857 | uargs = NULL; |
| 858 | break; |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 859 | case SECCOMP_MODE_FILTER: |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 860 | op = SECCOMP_SET_MODE_FILTER; |
| 861 | uargs = filter; |
| 862 | break; |
Kees Cook | 3b23dd1 | 2014-06-25 15:55:25 -0700 | [diff] [blame] | 863 | default: |
| 864 | return -EINVAL; |
| 865 | } |
Kees Cook | 48dc92b | 2014-06-25 16:08:24 -0700 | [diff] [blame] | 866 | |
| 867 | /* prctl interface doesn't have flags, so they are always zero. */ |
| 868 | return do_seccomp(op, 0, uargs); |
Kees Cook | d78ab02 | 2014-05-21 15:02:11 -0700 | [diff] [blame] | 869 | } |