1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 /* 2 /*
3 * linux/kernel/seccomp.c 3 * linux/kernel/seccomp.c
4 * 4 *
5 * Copyright 2004-2005 Andrea Arcangeli <andr 5 * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com>
6 * 6 *
7 * Copyright (C) 2012 Google, Inc. 7 * Copyright (C) 2012 Google, Inc.
8 * Will Drewry <wad@chromium.org> 8 * Will Drewry <wad@chromium.org>
9 * 9 *
10 * This defines a simple but solid secure-comp 10 * This defines a simple but solid secure-computing facility.
11 * 11 *
12 * Mode 1 uses a fixed list of allowed system 12 * Mode 1 uses a fixed list of allowed system calls.
13 * Mode 2 allows user-defined system call filt 13 * Mode 2 allows user-defined system call filters in the form
14 * of Berkeley Packet Filters/Linux Soc 14 * of Berkeley Packet Filters/Linux Socket Filters.
15 */ 15 */
16 #define pr_fmt(fmt) "seccomp: " fmt 16 #define pr_fmt(fmt) "seccomp: " fmt
17 17
18 #include <linux/refcount.h> 18 #include <linux/refcount.h>
19 #include <linux/audit.h> 19 #include <linux/audit.h>
20 #include <linux/compat.h> 20 #include <linux/compat.h>
21 #include <linux/coredump.h> 21 #include <linux/coredump.h>
22 #include <linux/kmemleak.h> 22 #include <linux/kmemleak.h>
23 #include <linux/nospec.h> 23 #include <linux/nospec.h>
24 #include <linux/prctl.h> 24 #include <linux/prctl.h>
25 #include <linux/sched.h> 25 #include <linux/sched.h>
26 #include <linux/sched/task_stack.h> 26 #include <linux/sched/task_stack.h>
27 #include <linux/seccomp.h> 27 #include <linux/seccomp.h>
28 #include <linux/slab.h> 28 #include <linux/slab.h>
29 #include <linux/syscalls.h> 29 #include <linux/syscalls.h>
30 #include <linux/sysctl.h> 30 #include <linux/sysctl.h>
31 31
32 /* Not exposed in headers: strictly internal u <<
33 #define SECCOMP_MODE_DEAD (SECCOMP_MODE_ <<
34 <<
35 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER 32 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
36 #include <asm/syscall.h> 33 #include <asm/syscall.h>
37 #endif 34 #endif
38 35
39 #ifdef CONFIG_SECCOMP_FILTER 36 #ifdef CONFIG_SECCOMP_FILTER
40 #include <linux/file.h> 37 #include <linux/file.h>
41 #include <linux/filter.h> 38 #include <linux/filter.h>
42 #include <linux/pid.h> 39 #include <linux/pid.h>
43 #include <linux/ptrace.h> 40 #include <linux/ptrace.h>
44 #include <linux/capability.h> 41 #include <linux/capability.h>
>> 42 #include <linux/tracehook.h>
45 #include <linux/uaccess.h> 43 #include <linux/uaccess.h>
46 #include <linux/anon_inodes.h> 44 #include <linux/anon_inodes.h>
47 #include <linux/lockdep.h> 45 #include <linux/lockdep.h>
48 46
49 /* 47 /*
50 * When SECCOMP_IOCTL_NOTIF_ID_VALID was first 48 * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
51 * wrong direction flag in the ioctl number. T 49 * wrong direction flag in the ioctl number. This is the broken one,
52 * which the kernel needs to keep supporting u 50 * which the kernel needs to keep supporting until all userspaces stop
53 * using the wrong command number. 51 * using the wrong command number.
54 */ 52 */
55 #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR 53 #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR SECCOMP_IOR(2, __u64)
56 54
57 enum notify_state { 55 enum notify_state {
58 SECCOMP_NOTIFY_INIT, 56 SECCOMP_NOTIFY_INIT,
59 SECCOMP_NOTIFY_SENT, 57 SECCOMP_NOTIFY_SENT,
60 SECCOMP_NOTIFY_REPLIED, 58 SECCOMP_NOTIFY_REPLIED,
61 }; 59 };
62 60
63 struct seccomp_knotif { 61 struct seccomp_knotif {
64 /* The struct pid of the task whose fi 62 /* The struct pid of the task whose filter triggered the notification */
65 struct task_struct *task; 63 struct task_struct *task;
66 64
67 /* The "cookie" for this request; this 65 /* The "cookie" for this request; this is unique for this filter. */
68 u64 id; 66 u64 id;
69 67
70 /* 68 /*
71 * The seccomp data. This pointer is v 69 * The seccomp data. This pointer is valid the entire time this
72 * notification is active, since it co 70 * notification is active, since it comes from __seccomp_filter which
73 * eclipses the entire lifecycle here. 71 * eclipses the entire lifecycle here.
74 */ 72 */
75 const struct seccomp_data *data; 73 const struct seccomp_data *data;
76 74
77 /* 75 /*
78 * Notification states. When SECCOMP_R 76 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
79 * struct seccomp_knotif is created an 77 * struct seccomp_knotif is created and starts out in INIT. Once the
80 * handler reads the notification off 78 * handler reads the notification off of an FD, it transitions to SENT.
81 * If a signal is received the state t 79 * If a signal is received the state transitions back to INIT and
82 * another message is sent. When the u 80 * another message is sent. When the userspace handler replies, state
83 * transitions to REPLIED. 81 * transitions to REPLIED.
84 */ 82 */
85 enum notify_state state; 83 enum notify_state state;
86 84
87 /* The return values, only valid when 85 /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
88 int error; 86 int error;
89 long val; 87 long val;
90 u32 flags; 88 u32 flags;
91 89
92 /* 90 /*
93 * Signals when this has changed state 91 * Signals when this has changed states, such as the listener
94 * dying, a new seccomp addfd message, 92 * dying, a new seccomp addfd message, or changing to REPLIED
95 */ 93 */
96 struct completion ready; 94 struct completion ready;
97 95
98 struct list_head list; 96 struct list_head list;
99 97
100 /* outstanding addfd requests */ 98 /* outstanding addfd requests */
101 struct list_head addfd; 99 struct list_head addfd;
102 }; 100 };
103 101
104 /** 102 /**
105 * struct seccomp_kaddfd - container for secco 103 * struct seccomp_kaddfd - container for seccomp_addfd ioctl messages
106 * 104 *
107 * @file: A reference to the file to install i 105 * @file: A reference to the file to install in the other task
108 * @fd: The fd number to install it at. If the 106 * @fd: The fd number to install it at. If the fd number is -1, it means the
109 * installing process should allocate the 107 * installing process should allocate the fd as normal.
110 * @flags: The flags for the new file descript 108 * @flags: The flags for the new file descriptor. At the moment, only O_CLOEXEC
111 * is allowed. 109 * is allowed.
112 * @ioctl_flags: The flags used for the seccom <<
113 * @setfd: whether or not SECCOMP_ADDFD_FLAG_S <<
114 * @ret: The return value of the installing pr 110 * @ret: The return value of the installing process. It is set to the fd num
115 * upon success (>= 0). 111 * upon success (>= 0).
116 * @completion: Indicates that the installing 112 * @completion: Indicates that the installing process has completed fd
117 * installation, or gone away (ei 113 * installation, or gone away (either due to successful
118 * reply, or signal) 114 * reply, or signal)
119 * @list: list_head for chaining seccomp_kaddf <<
120 * 115 *
121 */ 116 */
122 struct seccomp_kaddfd { 117 struct seccomp_kaddfd {
123 struct file *file; 118 struct file *file;
124 int fd; 119 int fd;
125 unsigned int flags; 120 unsigned int flags;
126 __u32 ioctl_flags; <<
127 121
128 union { !! 122 /* To only be set on reply */
129 bool setfd; !! 123 int ret;
130 /* To only be set on reply */ <<
131 int ret; <<
132 }; <<
133 struct completion completion; 124 struct completion completion;
134 struct list_head list; 125 struct list_head list;
135 }; 126 };
136 127
137 /** 128 /**
138 * struct notification - container for seccomp 129 * struct notification - container for seccomp userspace notifications. Since
139 * most seccomp filters will not have notifica 130 * most seccomp filters will not have notification listeners attached and this
140 * structure is fairly large, we store the not 131 * structure is fairly large, we store the notification-specific stuff in a
141 * separate structure. 132 * separate structure.
142 * 133 *
143 * @requests: A semaphore that users of this n !! 134 * @request: A semaphore that users of this notification can wait on for
144 * changes. Actual reads and writes !! 135 * changes. Actual reads and writes are still controlled with
145 * filter->notify_lock. !! 136 * filter->notify_lock.
146 * @flags: A set of SECCOMP_USER_NOTIF_FD_* fl <<
147 * @next_id: The id of the next request. 137 * @next_id: The id of the next request.
148 * @notifications: A list of struct seccomp_kn 138 * @notifications: A list of struct seccomp_knotif elements.
149 */ 139 */
150 <<
151 struct notification { 140 struct notification {
152 atomic_t requests; !! 141 struct semaphore request;
153 u32 flags; <<
154 u64 next_id; 142 u64 next_id;
155 struct list_head notifications; 143 struct list_head notifications;
156 }; 144 };
157 145
158 #ifdef SECCOMP_ARCH_NATIVE <<
159 /** <<
160 * struct action_cache - per-filter cache of s <<
161 * arch/syscall pair <<
162 * <<
163 * @allow_native: A bitmap where each bit repr <<
164 * filter will always allow the <<
165 * native architecture. <<
166 * @allow_compat: A bitmap where each bit repr <<
167 * filter will always allow the <<
168 * compat architecture. <<
169 */ <<
170 struct action_cache { <<
171 DECLARE_BITMAP(allow_native, SECCOMP_A <<
172 #ifdef SECCOMP_ARCH_COMPAT <<
173 DECLARE_BITMAP(allow_compat, SECCOMP_A <<
174 #endif <<
175 }; <<
176 #else <<
177 struct action_cache { }; <<
178 <<
179 static inline bool seccomp_cache_check_allow(c <<
180 c <<
181 { <<
182 return false; <<
183 } <<
184 <<
185 static inline void seccomp_cache_prepare(struc <<
186 { <<
187 } <<
188 #endif /* SECCOMP_ARCH_NATIVE */ <<
189 <<
190 /** 146 /**
191 * struct seccomp_filter - container for secco 147 * struct seccomp_filter - container for seccomp BPF programs
192 * 148 *
193 * @refs: Reference count to manage the object 149 * @refs: Reference count to manage the object lifetime.
194 * A filter's reference count is increm 150 * A filter's reference count is incremented for each directly
195 * attached task, once for the dependen 151 * attached task, once for the dependent filter, and if
196 * requested for the user notifier. Whe 152 * requested for the user notifier. When @refs reaches zero,
197 * the filter can be freed. 153 * the filter can be freed.
198 * @users: A filter's @users count is incremen 154 * @users: A filter's @users count is incremented for each directly
199 * attached task (filter installation, 155 * attached task (filter installation, fork(), thread_sync),
200 * and once for the dependent filter ( 156 * and once for the dependent filter (tracked in filter->prev).
201 * When it reaches zero it indicates t 157 * When it reaches zero it indicates that no direct or indirect
202 * users of that filter exist. No new 158 * users of that filter exist. No new tasks can get associated with
203 * this filter after reaching 0. The @ 159 * this filter after reaching 0. The @users count is always smaller
204 * or equal to @refs. Hence, reaching 160 * or equal to @refs. Hence, reaching 0 for @users does not mean
205 * the filter can be freed. 161 * the filter can be freed.
206 * @cache: cache of arch/syscall mappings to a <<
207 * @log: true if all actions except for SECCOM 162 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
208 * @wait_killable_recv: Put notifying process <<
209 * notification is receiv <<
210 * @prev: points to a previously installed, or 163 * @prev: points to a previously installed, or inherited, filter
211 * @prog: the BPF program to evaluate 164 * @prog: the BPF program to evaluate
212 * @notif: the struct that holds all notificat 165 * @notif: the struct that holds all notification related information
213 * @notify_lock: A lock for all notification-r 166 * @notify_lock: A lock for all notification-related accesses.
214 * @wqh: A wait queue for poll if a notifier i 167 * @wqh: A wait queue for poll if a notifier is in use.
215 * 168 *
216 * seccomp_filter objects are organized in a t 169 * seccomp_filter objects are organized in a tree linked via the @prev
217 * pointer. For any task, it appears to be a 170 * pointer. For any task, it appears to be a singly-linked list starting
218 * with current->seccomp.filter, the most rece 171 * with current->seccomp.filter, the most recently attached or inherited filter.
219 * However, multiple filters may share a @prev 172 * However, multiple filters may share a @prev node, by way of fork(), which
220 * results in a unidirectional tree existing i 173 * results in a unidirectional tree existing in memory. This is similar to
221 * how namespaces work. 174 * how namespaces work.
222 * 175 *
223 * seccomp_filter objects should never be modi 176 * seccomp_filter objects should never be modified after being attached
224 * to a task_struct (other than @refs). 177 * to a task_struct (other than @refs).
225 */ 178 */
226 struct seccomp_filter { 179 struct seccomp_filter {
227 refcount_t refs; 180 refcount_t refs;
228 refcount_t users; 181 refcount_t users;
229 bool log; 182 bool log;
230 bool wait_killable_recv; <<
231 struct action_cache cache; <<
232 struct seccomp_filter *prev; 183 struct seccomp_filter *prev;
233 struct bpf_prog *prog; 184 struct bpf_prog *prog;
234 struct notification *notif; 185 struct notification *notif;
235 struct mutex notify_lock; 186 struct mutex notify_lock;
236 wait_queue_head_t wqh; 187 wait_queue_head_t wqh;
237 }; 188 };
238 189
239 /* Limit any path through the tree to 256KB wo 190 /* Limit any path through the tree to 256KB worth of instructions. */
240 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof 191 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
241 192
242 /* 193 /*
243 * Endianness is explicitly ignored and left f 194 * Endianness is explicitly ignored and left for BPF program authors to manage
244 * as per the specific architecture. 195 * as per the specific architecture.
245 */ 196 */
246 static void populate_seccomp_data(struct secco 197 static void populate_seccomp_data(struct seccomp_data *sd)
247 { 198 {
248 /* <<
249 * Instead of using current_pt_reg(), <<
250 * to safely fetch "current", so just <<
251 */ <<
252 struct task_struct *task = current; 199 struct task_struct *task = current;
253 struct pt_regs *regs = task_pt_regs(ta 200 struct pt_regs *regs = task_pt_regs(task);
254 unsigned long args[6]; 201 unsigned long args[6];
255 202
256 sd->nr = syscall_get_nr(task, regs); 203 sd->nr = syscall_get_nr(task, regs);
257 sd->arch = syscall_get_arch(task); 204 sd->arch = syscall_get_arch(task);
258 syscall_get_arguments(task, regs, args 205 syscall_get_arguments(task, regs, args);
259 sd->args[0] = args[0]; 206 sd->args[0] = args[0];
260 sd->args[1] = args[1]; 207 sd->args[1] = args[1];
261 sd->args[2] = args[2]; 208 sd->args[2] = args[2];
262 sd->args[3] = args[3]; 209 sd->args[3] = args[3];
263 sd->args[4] = args[4]; 210 sd->args[4] = args[4];
264 sd->args[5] = args[5]; 211 sd->args[5] = args[5];
265 sd->instruction_pointer = KSTK_EIP(tas 212 sd->instruction_pointer = KSTK_EIP(task);
266 } 213 }
267 214
268 /** 215 /**
269 * seccomp_check_filter - verify seccomp 216 * seccomp_check_filter - verify seccomp filter code
270 * @filter: filter to verify 217 * @filter: filter to verify
271 * @flen: length of filter 218 * @flen: length of filter
272 * 219 *
273 * Takes a previously checked filter (by bpf_c 220 * Takes a previously checked filter (by bpf_check_classic) and
274 * redirects all filter code that loads struct 221 * redirects all filter code that loads struct sk_buff data
275 * and related data through seccomp_bpf_load. 222 * and related data through seccomp_bpf_load. It also
276 * enforces length and alignment checking of t 223 * enforces length and alignment checking of those loads.
277 * 224 *
278 * Returns 0 if the rule set is legal or -EINV 225 * Returns 0 if the rule set is legal or -EINVAL if not.
279 */ 226 */
280 static int seccomp_check_filter(struct sock_fi 227 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
281 { 228 {
282 int pc; 229 int pc;
283 for (pc = 0; pc < flen; pc++) { 230 for (pc = 0; pc < flen; pc++) {
284 struct sock_filter *ftest = &f 231 struct sock_filter *ftest = &filter[pc];
285 u16 code = ftest->code; 232 u16 code = ftest->code;
286 u32 k = ftest->k; 233 u32 k = ftest->k;
287 234
288 switch (code) { 235 switch (code) {
289 case BPF_LD | BPF_W | BPF_ABS: 236 case BPF_LD | BPF_W | BPF_ABS:
290 ftest->code = BPF_LDX 237 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
291 /* 32-bit aligned and 238 /* 32-bit aligned and not out of bounds. */
292 if (k >= sizeof(struct 239 if (k >= sizeof(struct seccomp_data) || k & 3)
293 return -EINVAL 240 return -EINVAL;
294 continue; 241 continue;
295 case BPF_LD | BPF_W | BPF_LEN: 242 case BPF_LD | BPF_W | BPF_LEN:
296 ftest->code = BPF_LD | 243 ftest->code = BPF_LD | BPF_IMM;
297 ftest->k = sizeof(stru 244 ftest->k = sizeof(struct seccomp_data);
298 continue; 245 continue;
299 case BPF_LDX | BPF_W | BPF_LEN 246 case BPF_LDX | BPF_W | BPF_LEN:
300 ftest->code = BPF_LDX 247 ftest->code = BPF_LDX | BPF_IMM;
301 ftest->k = sizeof(stru 248 ftest->k = sizeof(struct seccomp_data);
302 continue; 249 continue;
303 /* Explicitly include allowed 250 /* Explicitly include allowed calls. */
304 case BPF_RET | BPF_K: 251 case BPF_RET | BPF_K:
305 case BPF_RET | BPF_A: 252 case BPF_RET | BPF_A:
306 case BPF_ALU | BPF_ADD | BPF_K 253 case BPF_ALU | BPF_ADD | BPF_K:
307 case BPF_ALU | BPF_ADD | BPF_X 254 case BPF_ALU | BPF_ADD | BPF_X:
308 case BPF_ALU | BPF_SUB | BPF_K 255 case BPF_ALU | BPF_SUB | BPF_K:
309 case BPF_ALU | BPF_SUB | BPF_X 256 case BPF_ALU | BPF_SUB | BPF_X:
310 case BPF_ALU | BPF_MUL | BPF_K 257 case BPF_ALU | BPF_MUL | BPF_K:
311 case BPF_ALU | BPF_MUL | BPF_X 258 case BPF_ALU | BPF_MUL | BPF_X:
312 case BPF_ALU | BPF_DIV | BPF_K 259 case BPF_ALU | BPF_DIV | BPF_K:
313 case BPF_ALU | BPF_DIV | BPF_X 260 case BPF_ALU | BPF_DIV | BPF_X:
314 case BPF_ALU | BPF_AND | BPF_K 261 case BPF_ALU | BPF_AND | BPF_K:
315 case BPF_ALU | BPF_AND | BPF_X 262 case BPF_ALU | BPF_AND | BPF_X:
316 case BPF_ALU | BPF_OR | BPF_K: 263 case BPF_ALU | BPF_OR | BPF_K:
317 case BPF_ALU | BPF_OR | BPF_X: 264 case BPF_ALU | BPF_OR | BPF_X:
318 case BPF_ALU | BPF_XOR | BPF_K 265 case BPF_ALU | BPF_XOR | BPF_K:
319 case BPF_ALU | BPF_XOR | BPF_X 266 case BPF_ALU | BPF_XOR | BPF_X:
320 case BPF_ALU | BPF_LSH | BPF_K 267 case BPF_ALU | BPF_LSH | BPF_K:
321 case BPF_ALU | BPF_LSH | BPF_X 268 case BPF_ALU | BPF_LSH | BPF_X:
322 case BPF_ALU | BPF_RSH | BPF_K 269 case BPF_ALU | BPF_RSH | BPF_K:
323 case BPF_ALU | BPF_RSH | BPF_X 270 case BPF_ALU | BPF_RSH | BPF_X:
324 case BPF_ALU | BPF_NEG: 271 case BPF_ALU | BPF_NEG:
325 case BPF_LD | BPF_IMM: 272 case BPF_LD | BPF_IMM:
326 case BPF_LDX | BPF_IMM: 273 case BPF_LDX | BPF_IMM:
327 case BPF_MISC | BPF_TAX: 274 case BPF_MISC | BPF_TAX:
328 case BPF_MISC | BPF_TXA: 275 case BPF_MISC | BPF_TXA:
329 case BPF_LD | BPF_MEM: 276 case BPF_LD | BPF_MEM:
330 case BPF_LDX | BPF_MEM: 277 case BPF_LDX | BPF_MEM:
331 case BPF_ST: 278 case BPF_ST:
332 case BPF_STX: 279 case BPF_STX:
333 case BPF_JMP | BPF_JA: 280 case BPF_JMP | BPF_JA:
334 case BPF_JMP | BPF_JEQ | BPF_K 281 case BPF_JMP | BPF_JEQ | BPF_K:
335 case BPF_JMP | BPF_JEQ | BPF_X 282 case BPF_JMP | BPF_JEQ | BPF_X:
336 case BPF_JMP | BPF_JGE | BPF_K 283 case BPF_JMP | BPF_JGE | BPF_K:
337 case BPF_JMP | BPF_JGE | BPF_X 284 case BPF_JMP | BPF_JGE | BPF_X:
338 case BPF_JMP | BPF_JGT | BPF_K 285 case BPF_JMP | BPF_JGT | BPF_K:
339 case BPF_JMP | BPF_JGT | BPF_X 286 case BPF_JMP | BPF_JGT | BPF_X:
340 case BPF_JMP | BPF_JSET | BPF_ 287 case BPF_JMP | BPF_JSET | BPF_K:
341 case BPF_JMP | BPF_JSET | BPF_ 288 case BPF_JMP | BPF_JSET | BPF_X:
342 continue; 289 continue;
343 default: 290 default:
344 return -EINVAL; 291 return -EINVAL;
345 } 292 }
346 } 293 }
347 return 0; 294 return 0;
348 } 295 }
349 296
350 #ifdef SECCOMP_ARCH_NATIVE <<
351 static inline bool seccomp_cache_check_allow_b <<
352 <<
353 <<
354 { <<
355 if (unlikely(syscall_nr < 0 || syscall <<
356 return false; <<
357 syscall_nr = array_index_nospec(syscal <<
358 <<
359 return test_bit(syscall_nr, bitmap); <<
360 } <<
361 <<
362 /** <<
363 * seccomp_cache_check_allow - lookup seccomp <<
364 * @sfilter: The seccomp filter <<
365 * @sd: The seccomp data to lookup the cache w <<
366 * <<
367 * Returns true if the seccomp_data is cached <<
368 */ <<
369 static inline bool seccomp_cache_check_allow(c <<
370 c <<
371 { <<
372 int syscall_nr = sd->nr; <<
373 const struct action_cache *cache = &sf <<
374 <<
375 #ifndef SECCOMP_ARCH_COMPAT <<
376 /* A native-only architecture doesn't <<
377 return seccomp_cache_check_allow_bitma <<
378 <<
379 <<
380 #else <<
381 if (likely(sd->arch == SECCOMP_ARCH_NA <<
382 return seccomp_cache_check_all <<
383 <<
384 <<
385 if (likely(sd->arch == SECCOMP_ARCH_CO <<
386 return seccomp_cache_check_all <<
387 <<
388 <<
389 #endif /* SECCOMP_ARCH_COMPAT */ <<
390 <<
391 WARN_ON_ONCE(true); <<
392 return false; <<
393 } <<
394 #endif /* SECCOMP_ARCH_NATIVE */ <<
395 <<
396 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCO <<
397 /** 297 /**
398 * seccomp_run_filters - evaluates all seccomp 298 * seccomp_run_filters - evaluates all seccomp filters against @sd
399 * @sd: optional seccomp data to be passed to 299 * @sd: optional seccomp data to be passed to filters
400 * @match: stores struct seccomp_filter that r 300 * @match: stores struct seccomp_filter that resulted in the return value,
401 * unless filter returned SECCOMP_RET_ 301 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
402 * be unchanged. 302 * be unchanged.
403 * 303 *
404 * Returns valid seccomp BPF response codes. 304 * Returns valid seccomp BPF response codes.
405 */ 305 */
>> 306 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
406 static u32 seccomp_run_filters(const struct se 307 static u32 seccomp_run_filters(const struct seccomp_data *sd,
407 struct seccomp_ 308 struct seccomp_filter **match)
408 { 309 {
409 u32 ret = SECCOMP_RET_ALLOW; 310 u32 ret = SECCOMP_RET_ALLOW;
410 /* Make sure cross-thread synced filte 311 /* Make sure cross-thread synced filter points somewhere sane. */
411 struct seccomp_filter *f = 312 struct seccomp_filter *f =
412 READ_ONCE(current->sec 313 READ_ONCE(current->seccomp.filter);
413 314
414 /* Ensure unexpected behavior doesn't 315 /* Ensure unexpected behavior doesn't result in failing open. */
415 if (WARN_ON(f == NULL)) 316 if (WARN_ON(f == NULL))
416 return SECCOMP_RET_KILL_PROCES 317 return SECCOMP_RET_KILL_PROCESS;
417 318
418 if (seccomp_cache_check_allow(f, sd)) <<
419 return SECCOMP_RET_ALLOW; <<
420 <<
421 /* 319 /*
422 * All filters in the list are evaluat 320 * All filters in the list are evaluated and the lowest BPF return
423 * value always takes priority (ignori 321 * value always takes priority (ignoring the DATA).
424 */ 322 */
425 for (; f; f = f->prev) { 323 for (; f; f = f->prev) {
426 u32 cur_ret = bpf_prog_run_pin 324 u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
427 325
428 if (ACTION_ONLY(cur_ret) < ACT 326 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
429 ret = cur_ret; 327 ret = cur_ret;
430 *match = f; 328 *match = f;
431 } 329 }
432 } 330 }
433 return ret; 331 return ret;
434 } 332 }
435 #endif /* CONFIG_SECCOMP_FILTER */ 333 #endif /* CONFIG_SECCOMP_FILTER */
436 334
437 static inline bool seccomp_may_assign_mode(uns 335 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
438 { 336 {
439 assert_spin_locked(¤t->sighand-> 337 assert_spin_locked(¤t->sighand->siglock);
440 338
441 if (current->seccomp.mode && current-> 339 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
442 return false; 340 return false;
443 341
444 return true; 342 return true;
445 } 343 }
446 344
447 void __weak arch_seccomp_spec_mitigate(struct 345 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
448 346
449 static inline void seccomp_assign_mode(struct 347 static inline void seccomp_assign_mode(struct task_struct *task,
450 unsigne 348 unsigned long seccomp_mode,
451 unsigne 349 unsigned long flags)
452 { 350 {
453 assert_spin_locked(&task->sighand->sig 351 assert_spin_locked(&task->sighand->siglock);
454 352
455 task->seccomp.mode = seccomp_mode; 353 task->seccomp.mode = seccomp_mode;
456 /* 354 /*
457 * Make sure SYSCALL_WORK_SECCOMP cann !! 355 * Make sure TIF_SECCOMP cannot be set before the mode (and
458 * filter) is set. 356 * filter) is set.
459 */ 357 */
460 smp_mb__before_atomic(); 358 smp_mb__before_atomic();
461 /* Assume default seccomp processes wa 359 /* Assume default seccomp processes want spec flaw mitigation. */
462 if ((flags & SECCOMP_FILTER_FLAG_SPEC_ 360 if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
463 arch_seccomp_spec_mitigate(tas 361 arch_seccomp_spec_mitigate(task);
464 set_task_syscall_work(task, SECCOMP); !! 362 set_tsk_thread_flag(task, TIF_SECCOMP);
465 } 363 }
466 364
467 #ifdef CONFIG_SECCOMP_FILTER 365 #ifdef CONFIG_SECCOMP_FILTER
468 /* Returns 1 if the parent is an ancestor of t 366 /* Returns 1 if the parent is an ancestor of the child. */
469 static int is_ancestor(struct seccomp_filter * 367 static int is_ancestor(struct seccomp_filter *parent,
470 struct seccomp_filter * 368 struct seccomp_filter *child)
471 { 369 {
472 /* NULL is the root ancestor. */ 370 /* NULL is the root ancestor. */
473 if (parent == NULL) 371 if (parent == NULL)
474 return 1; 372 return 1;
475 for (; child; child = child->prev) 373 for (; child; child = child->prev)
476 if (child == parent) 374 if (child == parent)
477 return 1; 375 return 1;
478 return 0; 376 return 0;
479 } 377 }
480 378
481 /** 379 /**
482 * seccomp_can_sync_threads: checks if all thr 380 * seccomp_can_sync_threads: checks if all threads can be synchronized
483 * 381 *
484 * Expects sighand and cred_guard_mutex locks 382 * Expects sighand and cred_guard_mutex locks to be held.
485 * 383 *
486 * Returns 0 on success, -ve on error, or the 384 * Returns 0 on success, -ve on error, or the pid of a thread which was
487 * either not in the correct seccomp mode or d 385 * either not in the correct seccomp mode or did not have an ancestral
488 * seccomp filter. 386 * seccomp filter.
489 */ 387 */
490 static inline pid_t seccomp_can_sync_threads(v 388 static inline pid_t seccomp_can_sync_threads(void)
491 { 389 {
492 struct task_struct *thread, *caller; 390 struct task_struct *thread, *caller;
493 391
494 BUG_ON(!mutex_is_locked(¤t->sign 392 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
495 assert_spin_locked(¤t->sighand-> 393 assert_spin_locked(¤t->sighand->siglock);
496 394
497 /* Validate all threads being eligible 395 /* Validate all threads being eligible for synchronization. */
498 caller = current; 396 caller = current;
499 for_each_thread(caller, thread) { 397 for_each_thread(caller, thread) {
500 pid_t failed; 398 pid_t failed;
501 399
502 /* Skip current, since it is i 400 /* Skip current, since it is initiating the sync. */
503 if (thread == caller) 401 if (thread == caller)
504 continue; 402 continue;
505 /* Skip exited threads. */ <<
506 if (thread->flags & PF_EXITING <<
507 continue; <<
508 403
509 if (thread->seccomp.mode == SE 404 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
510 (thread->seccomp.mode == S 405 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
511 is_ancestor(thread->secco 406 is_ancestor(thread->seccomp.filter,
512 caller->secco 407 caller->seccomp.filter)))
513 continue; 408 continue;
514 409
515 /* Return the first thread tha 410 /* Return the first thread that cannot be synchronized. */
516 failed = task_pid_vnr(thread); 411 failed = task_pid_vnr(thread);
517 /* If the pid cannot be resolv 412 /* If the pid cannot be resolved, then return -ESRCH */
518 if (WARN_ON(failed == 0)) 413 if (WARN_ON(failed == 0))
519 failed = -ESRCH; 414 failed = -ESRCH;
520 return failed; 415 return failed;
521 } 416 }
522 417
523 return 0; 418 return 0;
524 } 419 }
525 420
526 static inline void seccomp_filter_free(struct 421 static inline void seccomp_filter_free(struct seccomp_filter *filter)
527 { 422 {
528 if (filter) { 423 if (filter) {
529 bpf_prog_destroy(filter->prog) 424 bpf_prog_destroy(filter->prog);
530 kfree(filter); 425 kfree(filter);
531 } 426 }
532 } 427 }
533 428
534 static void __seccomp_filter_orphan(struct sec 429 static void __seccomp_filter_orphan(struct seccomp_filter *orig)
535 { 430 {
536 while (orig && refcount_dec_and_test(& 431 while (orig && refcount_dec_and_test(&orig->users)) {
537 if (waitqueue_active(&orig->wq 432 if (waitqueue_active(&orig->wqh))
538 wake_up_poll(&orig->wq 433 wake_up_poll(&orig->wqh, EPOLLHUP);
539 orig = orig->prev; 434 orig = orig->prev;
540 } 435 }
541 } 436 }
542 437
543 static void __put_seccomp_filter(struct seccom 438 static void __put_seccomp_filter(struct seccomp_filter *orig)
544 { 439 {
545 /* Clean up single-reference branches 440 /* Clean up single-reference branches iteratively. */
546 while (orig && refcount_dec_and_test(& 441 while (orig && refcount_dec_and_test(&orig->refs)) {
547 struct seccomp_filter *freeme 442 struct seccomp_filter *freeme = orig;
548 orig = orig->prev; 443 orig = orig->prev;
549 seccomp_filter_free(freeme); 444 seccomp_filter_free(freeme);
550 } 445 }
551 } 446 }
552 447
553 static void __seccomp_filter_release(struct se 448 static void __seccomp_filter_release(struct seccomp_filter *orig)
554 { 449 {
555 /* Notify about any unused filters in 450 /* Notify about any unused filters in the task's former filter tree. */
556 __seccomp_filter_orphan(orig); 451 __seccomp_filter_orphan(orig);
557 /* Finally drop all references to the 452 /* Finally drop all references to the task's former tree. */
558 __put_seccomp_filter(orig); 453 __put_seccomp_filter(orig);
559 } 454 }
560 455
561 /** 456 /**
562 * seccomp_filter_release - Detach the task fr 457 * seccomp_filter_release - Detach the task from its filter tree,
563 * drop its reference 458 * drop its reference count, and notify
564 * about unused filte 459 * about unused filters
565 * 460 *
566 * @tsk: task the filter should be released fr <<
567 * <<
568 * This function should only be called when th 461 * This function should only be called when the task is exiting as
569 * it detaches it from its filter tree. PF_EXI !! 462 * it detaches it from its filter tree. As such, READ_ONCE() and
570 * for the task. !! 463 * barriers are not needed here, as would normally be needed.
571 */ 464 */
572 void seccomp_filter_release(struct task_struct 465 void seccomp_filter_release(struct task_struct *tsk)
573 { 466 {
574 struct seccomp_filter *orig; !! 467 struct seccomp_filter *orig = tsk->seccomp.filter;
575 <<
576 if (WARN_ON((tsk->flags & PF_EXITING) <<
577 return; <<
578 468
579 spin_lock_irq(&tsk->sighand->siglock); <<
580 orig = tsk->seccomp.filter; <<
581 /* Detach task from its filter tree. * 469 /* Detach task from its filter tree. */
582 tsk->seccomp.filter = NULL; 470 tsk->seccomp.filter = NULL;
583 spin_unlock_irq(&tsk->sighand->siglock <<
584 __seccomp_filter_release(orig); 471 __seccomp_filter_release(orig);
585 } 472 }
586 473
587 /** 474 /**
588 * seccomp_sync_threads: sets all threads to u 475 * seccomp_sync_threads: sets all threads to use current's filter
589 * 476 *
590 * @flags: SECCOMP_FILTER_FLAG_* flags to set <<
591 * <<
592 * Expects sighand and cred_guard_mutex locks 477 * Expects sighand and cred_guard_mutex locks to be held, and for
593 * seccomp_can_sync_threads() to have returned 478 * seccomp_can_sync_threads() to have returned success already
594 * without dropping the locks. 479 * without dropping the locks.
595 * 480 *
596 */ 481 */
597 static inline void seccomp_sync_threads(unsign 482 static inline void seccomp_sync_threads(unsigned long flags)
598 { 483 {
599 struct task_struct *thread, *caller; 484 struct task_struct *thread, *caller;
600 485
601 BUG_ON(!mutex_is_locked(¤t->sign 486 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
602 assert_spin_locked(¤t->sighand-> 487 assert_spin_locked(¤t->sighand->siglock);
603 488
604 /* Synchronize all threads. */ 489 /* Synchronize all threads. */
605 caller = current; 490 caller = current;
606 for_each_thread(caller, thread) { 491 for_each_thread(caller, thread) {
607 /* Skip current, since it need 492 /* Skip current, since it needs no changes. */
608 if (thread == caller) 493 if (thread == caller)
609 continue; 494 continue;
610 495
611 /* <<
612 * Skip exited threads. seccom <<
613 * been already called for thi <<
614 */ <<
615 if (thread->flags & PF_EXITING <<
616 continue; <<
617 <<
618 /* Get a task reference for th 496 /* Get a task reference for the new leaf node. */
619 get_seccomp_filter(caller); 497 get_seccomp_filter(caller);
620 498
621 /* 499 /*
622 * Drop the task reference to 500 * Drop the task reference to the shared ancestor since
623 * current's path will hold a 501 * current's path will hold a reference. (This also
624 * allows a put before the ass 502 * allows a put before the assignment.)
625 */ 503 */
626 __seccomp_filter_release(threa 504 __seccomp_filter_release(thread->seccomp.filter);
627 505
628 /* Make our new filter tree vi 506 /* Make our new filter tree visible. */
629 smp_store_release(&thread->sec 507 smp_store_release(&thread->seccomp.filter,
630 caller->secc 508 caller->seccomp.filter);
631 atomic_set(&thread->seccomp.fi 509 atomic_set(&thread->seccomp.filter_count,
632 atomic_read(&caller !! 510 atomic_read(&thread->seccomp.filter_count));
633 511
634 /* 512 /*
635 * Don't let an unprivileged t 513 * Don't let an unprivileged task work around
636 * the no_new_privs restrictio 514 * the no_new_privs restriction by creating
637 * a thread that sets it up, e 515 * a thread that sets it up, enters seccomp,
638 * then dies. 516 * then dies.
639 */ 517 */
640 if (task_no_new_privs(caller)) 518 if (task_no_new_privs(caller))
641 task_set_no_new_privs( 519 task_set_no_new_privs(thread);
642 520
643 /* 521 /*
644 * Opt the other thread into s 522 * Opt the other thread into seccomp if needed.
645 * As threads are considered t 523 * As threads are considered to be trust-realm
646 * equivalent (see ptrace_may_ 524 * equivalent (see ptrace_may_access), it is safe to
647 * allow one thread to transit 525 * allow one thread to transition the other.
648 */ 526 */
649 if (thread->seccomp.mode == SE 527 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
650 seccomp_assign_mode(th 528 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
651 fl 529 flags);
652 } 530 }
653 } 531 }
654 532
655 /** 533 /**
656 * seccomp_prepare_filter: Prepares a seccomp 534 * seccomp_prepare_filter: Prepares a seccomp filter for use.
657 * @fprog: BPF program to install 535 * @fprog: BPF program to install
658 * 536 *
659 * Returns filter on success or an ERR_PTR on 537 * Returns filter on success or an ERR_PTR on failure.
660 */ 538 */
661 static struct seccomp_filter *seccomp_prepare_ 539 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
662 { 540 {
663 struct seccomp_filter *sfilter; 541 struct seccomp_filter *sfilter;
664 int ret; 542 int ret;
665 const bool save_orig = !! 543 const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
666 #if defined(CONFIG_CHECKPOINT_RESTORE) || defi <<
667 true; <<
668 #else <<
669 false; <<
670 #endif <<
671 544
672 if (fprog->len == 0 || fprog->len > BP 545 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
673 return ERR_PTR(-EINVAL); 546 return ERR_PTR(-EINVAL);
674 547
675 BUG_ON(INT_MAX / fprog->len < sizeof(s 548 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
676 549
677 /* 550 /*
678 * Installing a seccomp filter require 551 * Installing a seccomp filter requires that the task has
679 * CAP_SYS_ADMIN in its namespace or b 552 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
680 * This avoids scenarios where unprivi 553 * This avoids scenarios where unprivileged tasks can affect the
681 * behavior of privileged children. 554 * behavior of privileged children.
682 */ 555 */
683 if (!task_no_new_privs(current) && 556 if (!task_no_new_privs(current) &&
684 !ns_capable_noaudit(cu 557 !ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN))
685 return ERR_PTR(-EACCES); 558 return ERR_PTR(-EACCES);
686 559
687 /* Allocate a new seccomp_filter */ 560 /* Allocate a new seccomp_filter */
688 sfilter = kzalloc(sizeof(*sfilter), GF 561 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
689 if (!sfilter) 562 if (!sfilter)
690 return ERR_PTR(-ENOMEM); 563 return ERR_PTR(-ENOMEM);
691 564
692 mutex_init(&sfilter->notify_lock); 565 mutex_init(&sfilter->notify_lock);
693 ret = bpf_prog_create_from_user(&sfilt 566 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
694 seccom 567 seccomp_check_filter, save_orig);
695 if (ret < 0) { 568 if (ret < 0) {
696 kfree(sfilter); 569 kfree(sfilter);
697 return ERR_PTR(ret); 570 return ERR_PTR(ret);
698 } 571 }
699 572
700 refcount_set(&sfilter->refs, 1); 573 refcount_set(&sfilter->refs, 1);
701 refcount_set(&sfilter->users, 1); 574 refcount_set(&sfilter->users, 1);
702 init_waitqueue_head(&sfilter->wqh); 575 init_waitqueue_head(&sfilter->wqh);
703 576
704 return sfilter; 577 return sfilter;
705 } 578 }
706 579
707 /** 580 /**
708 * seccomp_prepare_user_filter - prepares a us 581 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
709 * @user_filter: pointer to the user data cont 582 * @user_filter: pointer to the user data containing a sock_fprog.
710 * 583 *
711 * Returns 0 on success and non-zero otherwise 584 * Returns 0 on success and non-zero otherwise.
712 */ 585 */
713 static struct seccomp_filter * 586 static struct seccomp_filter *
714 seccomp_prepare_user_filter(const char __user 587 seccomp_prepare_user_filter(const char __user *user_filter)
715 { 588 {
716 struct sock_fprog fprog; 589 struct sock_fprog fprog;
717 struct seccomp_filter *filter = ERR_PT 590 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
718 591
719 #ifdef CONFIG_COMPAT 592 #ifdef CONFIG_COMPAT
720 if (in_compat_syscall()) { 593 if (in_compat_syscall()) {
721 struct compat_sock_fprog fprog 594 struct compat_sock_fprog fprog32;
722 if (copy_from_user(&fprog32, u 595 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
723 goto out; 596 goto out;
724 fprog.len = fprog32.len; 597 fprog.len = fprog32.len;
725 fprog.filter = compat_ptr(fpro 598 fprog.filter = compat_ptr(fprog32.filter);
726 } else /* falls through to the if belo 599 } else /* falls through to the if below. */
727 #endif 600 #endif
728 if (copy_from_user(&fprog, user_filter 601 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
729 goto out; 602 goto out;
730 filter = seccomp_prepare_filter(&fprog 603 filter = seccomp_prepare_filter(&fprog);
731 out: 604 out:
732 return filter; 605 return filter;
733 } 606 }
734 607
735 #ifdef SECCOMP_ARCH_NATIVE <<
736 /** <<
737 * seccomp_is_const_allow - check if filter is <<
738 * @fprog: The BPF programs <<
739 * @sd: The seccomp data to check against, onl <<
740 * number are considered constant. <<
741 */ <<
742 static bool seccomp_is_const_allow(struct sock <<
743 struct secc <<
744 { <<
745 unsigned int reg_value = 0; <<
746 unsigned int pc; <<
747 bool op_res; <<
748 <<
749 if (WARN_ON_ONCE(!fprog)) <<
750 return false; <<
751 <<
752 for (pc = 0; pc < fprog->len; pc++) { <<
753 struct sock_filter *insn = &fp <<
754 u16 code = insn->code; <<
755 u32 k = insn->k; <<
756 <<
757 switch (code) { <<
758 case BPF_LD | BPF_W | BPF_ABS: <<
759 switch (k) { <<
760 case offsetof(struct s <<
761 reg_value = sd <<
762 break; <<
763 case offsetof(struct s <<
764 reg_value = sd <<
765 break; <<
766 default: <<
767 /* can't optim <<
768 return false; <<
769 } <<
770 break; <<
771 case BPF_RET | BPF_K: <<
772 /* reached return with <<
773 return k == SECCOMP_RE <<
774 case BPF_JMP | BPF_JA: <<
775 pc += insn->k; <<
776 break; <<
777 case BPF_JMP | BPF_JEQ | BPF_K <<
778 case BPF_JMP | BPF_JGE | BPF_K <<
779 case BPF_JMP | BPF_JGT | BPF_K <<
780 case BPF_JMP | BPF_JSET | BPF_ <<
781 switch (BPF_OP(code)) <<
782 case BPF_JEQ: <<
783 op_res = reg_v <<
784 break; <<
785 case BPF_JGE: <<
786 op_res = reg_v <<
787 break; <<
788 case BPF_JGT: <<
789 op_res = reg_v <<
790 break; <<
791 case BPF_JSET: <<
792 op_res = !!(re <<
793 break; <<
794 default: <<
795 /* can't optim <<
796 return false; <<
797 } <<
798 <<
799 pc += op_res ? insn->j <<
800 break; <<
801 case BPF_ALU | BPF_AND | BPF_K <<
802 reg_value &= k; <<
803 break; <<
804 default: <<
805 /* can't optimize (unk <<
806 return false; <<
807 } <<
808 } <<
809 <<
810 /* ran off the end of the filter?! */ <<
811 WARN_ON(1); <<
812 return false; <<
813 } <<
814 <<
815 static void seccomp_cache_prepare_bitmap(struc <<
816 void <<
817 size_ <<
818 { <<
819 struct sock_fprog_kern *fprog = sfilte <<
820 struct seccomp_data sd; <<
821 int nr; <<
822 <<
823 if (bitmap_prev) { <<
824 /* The new filter must be as r <<
825 bitmap_copy(bitmap, bitmap_pre <<
826 } else { <<
827 /* Before any filters, all sys <<
828 bitmap_fill(bitmap, bitmap_siz <<
829 } <<
830 <<
831 for (nr = 0; nr < bitmap_size; nr++) { <<
832 /* No bitmap change: not a cac <<
833 if (!test_bit(nr, bitmap)) <<
834 continue; <<
835 <<
836 sd.nr = nr; <<
837 sd.arch = arch; <<
838 <<
839 /* No bitmap change: continue <<
840 if (seccomp_is_const_allow(fpr <<
841 continue; <<
842 <<
843 /* <<
844 * Not a cacheable action: alw <<
845 * atomic clear_bit() not need <<
846 */ <<
847 __clear_bit(nr, bitmap); <<
848 } <<
849 } <<
850 <<
851 /** <<
852 * seccomp_cache_prepare - emulate the filter <<
853 * @sfilter: The seccomp filter <<
854 * <<
855 * Returns 0 if successful or -errno if error <<
856 */ <<
857 static void seccomp_cache_prepare(struct secco <<
858 { <<
859 struct action_cache *cache = &sfilter- <<
860 const struct action_cache *cache_prev <<
861 sfilter->prev ? &sfilter->prev <<
862 <<
863 seccomp_cache_prepare_bitmap(sfilter, <<
864 cache_pre <<
865 SECCOMP_A <<
866 SECCOMP_A <<
867 <<
868 #ifdef SECCOMP_ARCH_COMPAT <<
869 seccomp_cache_prepare_bitmap(sfilter, <<
870 cache_pre <<
871 SECCOMP_A <<
872 SECCOMP_A <<
873 #endif /* SECCOMP_ARCH_COMPAT */ <<
874 } <<
875 #endif /* SECCOMP_ARCH_NATIVE */ <<
876 <<
877 /** 608 /**
878 * seccomp_attach_filter: validate and attach 609 * seccomp_attach_filter: validate and attach filter
879 * @flags: flags to change filter behavior 610 * @flags: flags to change filter behavior
880 * @filter: seccomp filter to add to the curre 611 * @filter: seccomp filter to add to the current process
881 * 612 *
882 * Caller must be holding current->sighand->si 613 * Caller must be holding current->sighand->siglock lock.
883 * 614 *
884 * Returns 0 on success, -ve on error, or 615 * Returns 0 on success, -ve on error, or
885 * - in TSYNC mode: the pid of a thread whic 616 * - in TSYNC mode: the pid of a thread which was either not in the correct
886 * seccomp mode or did not have an ancestr 617 * seccomp mode or did not have an ancestral seccomp filter
887 * - in NEW_LISTENER mode: the fd of the new 618 * - in NEW_LISTENER mode: the fd of the new listener
888 */ 619 */
889 static long seccomp_attach_filter(unsigned int 620 static long seccomp_attach_filter(unsigned int flags,
890 struct secco 621 struct seccomp_filter *filter)
891 { 622 {
892 unsigned long total_insns; 623 unsigned long total_insns;
893 struct seccomp_filter *walker; 624 struct seccomp_filter *walker;
894 625
895 assert_spin_locked(¤t->sighand-> 626 assert_spin_locked(¤t->sighand->siglock);
896 627
897 /* Validate resulting filter length. * 628 /* Validate resulting filter length. */
898 total_insns = filter->prog->len; 629 total_insns = filter->prog->len;
899 for (walker = current->seccomp.filter; 630 for (walker = current->seccomp.filter; walker; walker = walker->prev)
900 total_insns += walker->prog->l 631 total_insns += walker->prog->len + 4; /* 4 instr penalty */
901 if (total_insns > MAX_INSNS_PER_PATH) 632 if (total_insns > MAX_INSNS_PER_PATH)
902 return -ENOMEM; 633 return -ENOMEM;
903 634
904 /* If thread sync has been requested, 635 /* If thread sync has been requested, check that it is possible. */
905 if (flags & SECCOMP_FILTER_FLAG_TSYNC) 636 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
906 int ret; 637 int ret;
907 638
908 ret = seccomp_can_sync_threads 639 ret = seccomp_can_sync_threads();
909 if (ret) { 640 if (ret) {
910 if (flags & SECCOMP_FI 641 if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
911 return -ESRCH; 642 return -ESRCH;
912 else 643 else
913 return ret; 644 return ret;
914 } 645 }
915 } 646 }
916 647
917 /* Set log flag, if present. */ 648 /* Set log flag, if present. */
918 if (flags & SECCOMP_FILTER_FLAG_LOG) 649 if (flags & SECCOMP_FILTER_FLAG_LOG)
919 filter->log = true; 650 filter->log = true;
920 651
921 /* Set wait killable flag, if present. <<
922 if (flags & SECCOMP_FILTER_FLAG_WAIT_K <<
923 filter->wait_killable_recv = t <<
924 <<
925 /* 652 /*
926 * If there is an existing filter, mak 653 * If there is an existing filter, make it the prev and don't drop its
927 * task reference. 654 * task reference.
928 */ 655 */
929 filter->prev = current->seccomp.filter 656 filter->prev = current->seccomp.filter;
930 seccomp_cache_prepare(filter); <<
931 current->seccomp.filter = filter; 657 current->seccomp.filter = filter;
932 atomic_inc(¤t->seccomp.filter_co 658 atomic_inc(¤t->seccomp.filter_count);
933 659
934 /* Now that the new filter is in place 660 /* Now that the new filter is in place, synchronize to all threads. */
935 if (flags & SECCOMP_FILTER_FLAG_TSYNC) 661 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
936 seccomp_sync_threads(flags); 662 seccomp_sync_threads(flags);
937 663
938 return 0; 664 return 0;
939 } 665 }
940 666
941 static void __get_seccomp_filter(struct seccom 667 static void __get_seccomp_filter(struct seccomp_filter *filter)
942 { 668 {
943 refcount_inc(&filter->refs); 669 refcount_inc(&filter->refs);
944 } 670 }
945 671
946 /* get_seccomp_filter - increments the referen 672 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
947 void get_seccomp_filter(struct task_struct *ts 673 void get_seccomp_filter(struct task_struct *tsk)
948 { 674 {
949 struct seccomp_filter *orig = tsk->sec 675 struct seccomp_filter *orig = tsk->seccomp.filter;
950 if (!orig) 676 if (!orig)
951 return; 677 return;
952 __get_seccomp_filter(orig); 678 __get_seccomp_filter(orig);
953 refcount_inc(&orig->users); 679 refcount_inc(&orig->users);
954 } 680 }
955 681
>> 682 static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason)
>> 683 {
>> 684 clear_siginfo(info);
>> 685 info->si_signo = SIGSYS;
>> 686 info->si_code = SYS_SECCOMP;
>> 687 info->si_call_addr = (void __user *)KSTK_EIP(current);
>> 688 info->si_errno = reason;
>> 689 info->si_arch = syscall_get_arch(current);
>> 690 info->si_syscall = syscall;
>> 691 }
>> 692
>> 693 /**
>> 694 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
>> 695 * @syscall: syscall number to send to userland
>> 696 * @reason: filter-supplied reason code to send to userland (via si_errno)
>> 697 *
>> 698 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
>> 699 */
>> 700 static void seccomp_send_sigsys(int syscall, int reason)
>> 701 {
>> 702 struct kernel_siginfo info;
>> 703 seccomp_init_siginfo(&info, syscall, reason);
>> 704 force_sig_info(&info);
>> 705 }
956 #endif /* CONFIG_SECCOMP_FILTER */ 706 #endif /* CONFIG_SECCOMP_FILTER */
957 707
958 /* For use with seccomp_actions_logged */ 708 /* For use with seccomp_actions_logged */
959 #define SECCOMP_LOG_KILL_PROCESS (1 << 709 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
960 #define SECCOMP_LOG_KILL_THREAD (1 << 710 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
961 #define SECCOMP_LOG_TRAP (1 << 711 #define SECCOMP_LOG_TRAP (1 << 2)
962 #define SECCOMP_LOG_ERRNO (1 << 712 #define SECCOMP_LOG_ERRNO (1 << 3)
963 #define SECCOMP_LOG_TRACE (1 << 713 #define SECCOMP_LOG_TRACE (1 << 4)
964 #define SECCOMP_LOG_LOG (1 << 714 #define SECCOMP_LOG_LOG (1 << 5)
965 #define SECCOMP_LOG_ALLOW (1 << 715 #define SECCOMP_LOG_ALLOW (1 << 6)
966 #define SECCOMP_LOG_USER_NOTIF (1 << 716 #define SECCOMP_LOG_USER_NOTIF (1 << 7)
967 717
968 static u32 seccomp_actions_logged = SECCOMP_LO 718 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
969 SECCOMP_LO 719 SECCOMP_LOG_KILL_THREAD |
970 SECCOMP_LO 720 SECCOMP_LOG_TRAP |
971 SECCOMP_LO 721 SECCOMP_LOG_ERRNO |
972 SECCOMP_LO 722 SECCOMP_LOG_USER_NOTIF |
973 SECCOMP_LO 723 SECCOMP_LOG_TRACE |
974 SECCOMP_LO 724 SECCOMP_LOG_LOG;
975 725
976 static inline void seccomp_log(unsigned long s 726 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
977 bool requested) 727 bool requested)
978 { 728 {
979 bool log = false; 729 bool log = false;
980 730
981 switch (action) { 731 switch (action) {
982 case SECCOMP_RET_ALLOW: 732 case SECCOMP_RET_ALLOW:
983 break; 733 break;
984 case SECCOMP_RET_TRAP: 734 case SECCOMP_RET_TRAP:
985 log = requested && seccomp_act 735 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
986 break; 736 break;
987 case SECCOMP_RET_ERRNO: 737 case SECCOMP_RET_ERRNO:
988 log = requested && seccomp_act 738 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
989 break; 739 break;
990 case SECCOMP_RET_TRACE: 740 case SECCOMP_RET_TRACE:
991 log = requested && seccomp_act 741 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
992 break; 742 break;
993 case SECCOMP_RET_USER_NOTIF: 743 case SECCOMP_RET_USER_NOTIF:
994 log = requested && seccomp_act 744 log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
995 break; 745 break;
996 case SECCOMP_RET_LOG: 746 case SECCOMP_RET_LOG:
997 log = seccomp_actions_logged & 747 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
998 break; 748 break;
999 case SECCOMP_RET_KILL_THREAD: 749 case SECCOMP_RET_KILL_THREAD:
1000 log = seccomp_actions_logged 750 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
1001 break; 751 break;
1002 case SECCOMP_RET_KILL_PROCESS: 752 case SECCOMP_RET_KILL_PROCESS:
1003 default: 753 default:
1004 log = seccomp_actions_logged 754 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
1005 } 755 }
1006 756
1007 /* 757 /*
1008 * Emit an audit message when the act 758 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
1009 * FILTER_FLAG_LOG bit was set. The a 759 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
1010 * any action from being logged by re 760 * any action from being logged by removing the action name from the
1011 * seccomp_actions_logged sysctl. 761 * seccomp_actions_logged sysctl.
1012 */ 762 */
1013 if (!log) 763 if (!log)
1014 return; 764 return;
1015 765
1016 audit_seccomp(syscall, signr, action) 766 audit_seccomp(syscall, signr, action);
1017 } 767 }
1018 768
1019 /* 769 /*
1020 * Secure computing mode 1 allows only read/w 770 * Secure computing mode 1 allows only read/write/exit/sigreturn.
1021 * To be fully secure this must be combined w 771 * To be fully secure this must be combined with rlimit
1022 * to limit the stack allocations too. 772 * to limit the stack allocations too.
1023 */ 773 */
1024 static const int mode1_syscalls[] = { 774 static const int mode1_syscalls[] = {
1025 __NR_seccomp_read, __NR_seccomp_write 775 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
1026 -1, /* negative terminated */ 776 -1, /* negative terminated */
1027 }; 777 };
1028 778
1029 static void __secure_computing_strict(int thi 779 static void __secure_computing_strict(int this_syscall)
1030 { 780 {
1031 const int *allowed_syscalls = mode1_s 781 const int *allowed_syscalls = mode1_syscalls;
1032 #ifdef CONFIG_COMPAT 782 #ifdef CONFIG_COMPAT
1033 if (in_compat_syscall()) 783 if (in_compat_syscall())
1034 allowed_syscalls = get_compat 784 allowed_syscalls = get_compat_mode1_syscalls();
1035 #endif 785 #endif
1036 do { 786 do {
1037 if (*allowed_syscalls == this 787 if (*allowed_syscalls == this_syscall)
1038 return; 788 return;
1039 } while (*++allowed_syscalls != -1); 789 } while (*++allowed_syscalls != -1);
1040 790
1041 #ifdef SECCOMP_DEBUG 791 #ifdef SECCOMP_DEBUG
1042 dump_stack(); 792 dump_stack();
1043 #endif 793 #endif
1044 current->seccomp.mode = SECCOMP_MODE_ <<
1045 seccomp_log(this_syscall, SIGKILL, SE 794 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
1046 do_exit(SIGKILL); 795 do_exit(SIGKILL);
1047 } 796 }
1048 797
1049 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER 798 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
1050 void secure_computing_strict(int this_syscall 799 void secure_computing_strict(int this_syscall)
1051 { 800 {
1052 int mode = current->seccomp.mode; 801 int mode = current->seccomp.mode;
1053 802
1054 if (IS_ENABLED(CONFIG_CHECKPOINT_REST 803 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1055 unlikely(current->ptrace & PT_SUS 804 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1056 return; 805 return;
1057 806
1058 if (mode == SECCOMP_MODE_DISABLED) 807 if (mode == SECCOMP_MODE_DISABLED)
1059 return; 808 return;
1060 else if (mode == SECCOMP_MODE_STRICT) 809 else if (mode == SECCOMP_MODE_STRICT)
1061 __secure_computing_strict(thi 810 __secure_computing_strict(this_syscall);
1062 else 811 else
1063 BUG(); 812 BUG();
1064 } 813 }
1065 #else 814 #else
1066 815
1067 #ifdef CONFIG_SECCOMP_FILTER 816 #ifdef CONFIG_SECCOMP_FILTER
1068 static u64 seccomp_next_notify_id(struct secc 817 static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
1069 { 818 {
1070 /* 819 /*
1071 * Note: overflow is ok here, the id 820 * Note: overflow is ok here, the id just needs to be unique per
1072 * filter. 821 * filter.
1073 */ 822 */
1074 lockdep_assert_held(&filter->notify_l 823 lockdep_assert_held(&filter->notify_lock);
1075 return filter->notif->next_id++; 824 return filter->notif->next_id++;
1076 } 825 }
1077 826
1078 static void seccomp_handle_addfd(struct secco !! 827 static void seccomp_handle_addfd(struct seccomp_kaddfd *addfd)
1079 { 828 {
1080 int fd; <<
1081 <<
1082 /* 829 /*
1083 * Remove the notification, and reset 830 * Remove the notification, and reset the list pointers, indicating
1084 * that it has been handled. 831 * that it has been handled.
1085 */ 832 */
1086 list_del_init(&addfd->list); 833 list_del_init(&addfd->list);
1087 if (!addfd->setfd) !! 834 addfd->ret = receive_fd_replace(addfd->fd, addfd->file, addfd->flags);
1088 fd = receive_fd(addfd->file, <<
1089 else <<
1090 fd = receive_fd_replace(addfd <<
1091 addfd->ret = fd; <<
1092 <<
1093 if (addfd->ioctl_flags & SECCOMP_ADDF <<
1094 /* If we fail reset and retur <<
1095 if (fd < 0) { <<
1096 n->state = SECCOMP_NO <<
1097 } else { <<
1098 /* Return the FD we j <<
1099 n->flags = 0; <<
1100 n->error = 0; <<
1101 n->val = fd; <<
1102 } <<
1103 } <<
1104 <<
1105 /* <<
1106 * Mark the notification as completed <<
1107 * might be invalidated and we can't <<
1108 */ <<
1109 complete(&addfd->completion); 835 complete(&addfd->completion);
1110 } 836 }
1111 837
1112 static bool should_sleep_killable(struct secc <<
1113 struct secc <<
1114 { <<
1115 return match->wait_killable_recv && n <<
1116 } <<
1117 <<
1118 static int seccomp_do_user_notification(int t 838 static int seccomp_do_user_notification(int this_syscall,
1119 struc 839 struct seccomp_filter *match,
1120 const 840 const struct seccomp_data *sd)
1121 { 841 {
1122 int err; 842 int err;
1123 u32 flags = 0; 843 u32 flags = 0;
1124 long ret = 0; 844 long ret = 0;
1125 struct seccomp_knotif n = {}; 845 struct seccomp_knotif n = {};
1126 struct seccomp_kaddfd *addfd, *tmp; 846 struct seccomp_kaddfd *addfd, *tmp;
1127 847
1128 mutex_lock(&match->notify_lock); 848 mutex_lock(&match->notify_lock);
1129 err = -ENOSYS; 849 err = -ENOSYS;
1130 if (!match->notif) 850 if (!match->notif)
1131 goto out; 851 goto out;
1132 852
1133 n.task = current; 853 n.task = current;
1134 n.state = SECCOMP_NOTIFY_INIT; 854 n.state = SECCOMP_NOTIFY_INIT;
1135 n.data = sd; 855 n.data = sd;
1136 n.id = seccomp_next_notify_id(match); 856 n.id = seccomp_next_notify_id(match);
1137 init_completion(&n.ready); 857 init_completion(&n.ready);
1138 list_add_tail(&n.list, &match->notif- !! 858 list_add(&n.list, &match->notif->notifications);
1139 INIT_LIST_HEAD(&n.addfd); 859 INIT_LIST_HEAD(&n.addfd);
1140 860
1141 atomic_inc(&match->notif->requests); !! 861 up(&match->notif->request);
1142 if (match->notif->flags & SECCOMP_USE !! 862 wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM);
1143 wake_up_poll_on_current_cpu(& !! 863 mutex_unlock(&match->notify_lock);
1144 else <<
1145 wake_up_poll(&match->wqh, EPO <<
1146 864
1147 /* 865 /*
1148 * This is where we wait for a reply 866 * This is where we wait for a reply from userspace.
1149 */ 867 */
1150 do { !! 868 wait:
1151 bool wait_killable = should_s !! 869 err = wait_for_completion_interruptible(&n.ready);
1152 !! 870 mutex_lock(&match->notify_lock);
1153 mutex_unlock(&match->notify_l !! 871 if (err == 0) {
1154 if (wait_killable) !! 872 /* Check if we were woken up by a addfd message */
1155 err = wait_for_comple <<
1156 else <<
1157 err = wait_for_comple <<
1158 mutex_lock(&match->notify_loc <<
1159 <<
1160 if (err != 0) { <<
1161 /* <<
1162 * Check to see if th <<
1163 * whether we should <<
1164 */ <<
1165 if (!wait_killable && <<
1166 continue; <<
1167 <<
1168 goto interrupted; <<
1169 } <<
1170 <<
1171 addfd = list_first_entry_or_n 873 addfd = list_first_entry_or_null(&n.addfd,
1172 874 struct seccomp_kaddfd, list);
1173 /* Check if we were woken up !! 875 if (addfd && n.state != SECCOMP_NOTIFY_REPLIED) {
1174 if (addfd) !! 876 seccomp_handle_addfd(addfd);
1175 seccomp_handle_addfd( !! 877 mutex_unlock(&match->notify_lock);
1176 !! 878 goto wait;
1177 } while (n.state != SECCOMP_NOTIFY_R !! 879 }
1178 !! 880 ret = n.val;
1179 ret = n.val; !! 881 err = n.error;
1180 err = n.error; !! 882 flags = n.flags;
1181 flags = n.flags; !! 883 }
1182 884
1183 interrupted: <<
1184 /* If there were any pending addfd ca 885 /* If there were any pending addfd calls, clear them out */
1185 list_for_each_entry_safe(addfd, tmp, 886 list_for_each_entry_safe(addfd, tmp, &n.addfd, list) {
1186 /* The process went away befo 887 /* The process went away before we got a chance to handle it */
1187 addfd->ret = -ESRCH; 888 addfd->ret = -ESRCH;
1188 list_del_init(&addfd->list); 889 list_del_init(&addfd->list);
1189 complete(&addfd->completion); 890 complete(&addfd->completion);
1190 } 891 }
1191 892
1192 /* 893 /*
1193 * Note that it's possible the listen 894 * Note that it's possible the listener died in between the time when
1194 * we were notified of a response (or 895 * we were notified of a response (or a signal) and when we were able to
1195 * re-acquire the lock, so only delet 896 * re-acquire the lock, so only delete from the list if the
1196 * notification actually exists. 897 * notification actually exists.
1197 * 898 *
1198 * Also note that this test is only v 899 * Also note that this test is only valid because there's no way to
1199 * *reattach* to a notifier right now 900 * *reattach* to a notifier right now. If one is added, we'll need to
1200 * keep track of the notif itself and 901 * keep track of the notif itself and make sure they match here.
1201 */ 902 */
1202 if (match->notif) 903 if (match->notif)
1203 list_del(&n.list); 904 list_del(&n.list);
1204 out: 905 out:
1205 mutex_unlock(&match->notify_lock); 906 mutex_unlock(&match->notify_lock);
1206 907
1207 /* Userspace requests to continue the 908 /* Userspace requests to continue the syscall. */
1208 if (flags & SECCOMP_USER_NOTIF_FLAG_C 909 if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1209 return 0; 910 return 0;
1210 911
1211 syscall_set_return_value(current, cur !! 912 syscall_set_return_value(current, task_pt_regs(current),
1212 err, ret); 913 err, ret);
1213 return -1; 914 return -1;
1214 } 915 }
1215 916
1216 static int __seccomp_filter(int this_syscall, 917 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1217 const bool rechec 918 const bool recheck_after_trace)
1218 { 919 {
1219 u32 filter_ret, action; 920 u32 filter_ret, action;
1220 struct seccomp_filter *match = NULL; 921 struct seccomp_filter *match = NULL;
1221 int data; 922 int data;
1222 struct seccomp_data sd_local; 923 struct seccomp_data sd_local;
1223 924
1224 /* 925 /*
1225 * Make sure that any changes to mode 926 * Make sure that any changes to mode from another thread have
1226 * been seen after SYSCALL_WORK_SECCO !! 927 * been seen after TIF_SECCOMP was seen.
1227 */ 928 */
1228 smp_rmb(); !! 929 rmb();
1229 930
1230 if (!sd) { 931 if (!sd) {
1231 populate_seccomp_data(&sd_loc 932 populate_seccomp_data(&sd_local);
1232 sd = &sd_local; 933 sd = &sd_local;
1233 } 934 }
1234 935
1235 filter_ret = seccomp_run_filters(sd, 936 filter_ret = seccomp_run_filters(sd, &match);
1236 data = filter_ret & SECCOMP_RET_DATA; 937 data = filter_ret & SECCOMP_RET_DATA;
1237 action = filter_ret & SECCOMP_RET_ACT 938 action = filter_ret & SECCOMP_RET_ACTION_FULL;
1238 939
1239 switch (action) { 940 switch (action) {
1240 case SECCOMP_RET_ERRNO: 941 case SECCOMP_RET_ERRNO:
1241 /* Set low-order bits as an e 942 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
1242 if (data > MAX_ERRNO) 943 if (data > MAX_ERRNO)
1243 data = MAX_ERRNO; 944 data = MAX_ERRNO;
1244 syscall_set_return_value(curr !! 945 syscall_set_return_value(current, task_pt_regs(current),
1245 -dat 946 -data, 0);
1246 goto skip; 947 goto skip;
1247 948
1248 case SECCOMP_RET_TRAP: 949 case SECCOMP_RET_TRAP:
1249 /* Show the handler the origi 950 /* Show the handler the original registers. */
1250 syscall_rollback(current, cur !! 951 syscall_rollback(current, task_pt_regs(current));
1251 /* Let the filter pass back 1 952 /* Let the filter pass back 16 bits of data. */
1252 force_sig_seccomp(this_syscal !! 953 seccomp_send_sigsys(this_syscall, data);
1253 goto skip; 954 goto skip;
1254 955
1255 case SECCOMP_RET_TRACE: 956 case SECCOMP_RET_TRACE:
1256 /* We've been put in this sta 957 /* We've been put in this state by the ptracer already. */
1257 if (recheck_after_trace) 958 if (recheck_after_trace)
1258 return 0; 959 return 0;
1259 960
1260 /* ENOSYS these calls if ther 961 /* ENOSYS these calls if there is no tracer attached. */
1261 if (!ptrace_event_enabled(cur 962 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
1262 syscall_set_return_va 963 syscall_set_return_value(current,
1263 !! 964 task_pt_regs(current),
1264 965 -ENOSYS, 0);
1265 goto skip; 966 goto skip;
1266 } 967 }
1267 968
1268 /* Allow the BPF to provide t 969 /* Allow the BPF to provide the event message */
1269 ptrace_event(PTRACE_EVENT_SEC 970 ptrace_event(PTRACE_EVENT_SECCOMP, data);
1270 /* 971 /*
1271 * The delivery of a fatal si 972 * The delivery of a fatal signal during event
1272 * notification may silently 973 * notification may silently skip tracer notification,
1273 * which could leave us with 974 * which could leave us with a potentially unmodified
1274 * syscall that the tracer wo 975 * syscall that the tracer would have liked to have
1275 * changed. Since the process 976 * changed. Since the process is about to die, we just
1276 * force the syscall to be sk 977 * force the syscall to be skipped and let the signal
1277 * kill the process and corre 978 * kill the process and correctly handle any tracer exit
1278 * notifications. 979 * notifications.
1279 */ 980 */
1280 if (fatal_signal_pending(curr 981 if (fatal_signal_pending(current))
1281 goto skip; 982 goto skip;
1282 /* Check if the tracer forced 983 /* Check if the tracer forced the syscall to be skipped. */
1283 this_syscall = syscall_get_nr !! 984 this_syscall = syscall_get_nr(current, task_pt_regs(current));
1284 if (this_syscall < 0) 985 if (this_syscall < 0)
1285 goto skip; 986 goto skip;
1286 987
1287 /* 988 /*
1288 * Recheck the syscall, since 989 * Recheck the syscall, since it may have changed. This
1289 * intentionally uses a NULL 990 * intentionally uses a NULL struct seccomp_data to force
1290 * a reload of all registers. 991 * a reload of all registers. This does not goto skip since
1291 * a skip would have already 992 * a skip would have already been reported.
1292 */ 993 */
1293 if (__seccomp_filter(this_sys 994 if (__seccomp_filter(this_syscall, NULL, true))
1294 return -1; 995 return -1;
1295 996
1296 return 0; 997 return 0;
1297 998
1298 case SECCOMP_RET_USER_NOTIF: 999 case SECCOMP_RET_USER_NOTIF:
1299 if (seccomp_do_user_notificat 1000 if (seccomp_do_user_notification(this_syscall, match, sd))
1300 goto skip; 1001 goto skip;
1301 1002
1302 return 0; 1003 return 0;
1303 1004
1304 case SECCOMP_RET_LOG: 1005 case SECCOMP_RET_LOG:
1305 seccomp_log(this_syscall, 0, 1006 seccomp_log(this_syscall, 0, action, true);
1306 return 0; 1007 return 0;
1307 1008
1308 case SECCOMP_RET_ALLOW: 1009 case SECCOMP_RET_ALLOW:
1309 /* 1010 /*
1310 * Note that the "match" filt 1011 * Note that the "match" filter will always be NULL for
1311 * this action since SECCOMP_ 1012 * this action since SECCOMP_RET_ALLOW is the starting
1312 * state in seccomp_run_filte 1013 * state in seccomp_run_filters().
1313 */ 1014 */
1314 return 0; 1015 return 0;
1315 1016
1316 case SECCOMP_RET_KILL_THREAD: 1017 case SECCOMP_RET_KILL_THREAD:
1317 case SECCOMP_RET_KILL_PROCESS: 1018 case SECCOMP_RET_KILL_PROCESS:
1318 default: 1019 default:
1319 current->seccomp.mode = SECCO <<
1320 seccomp_log(this_syscall, SIG 1020 seccomp_log(this_syscall, SIGSYS, action, true);
1321 /* Dump core only if this is 1021 /* Dump core only if this is the last remaining thread. */
1322 if (action != SECCOMP_RET_KIL !! 1022 if (action == SECCOMP_RET_KILL_PROCESS ||
1323 (atomic_read(¤t->si !! 1023 get_nr_threads(current) == 1) {
>> 1024 kernel_siginfo_t info;
>> 1025
1324 /* Show the original 1026 /* Show the original registers in the dump. */
1325 syscall_rollback(curr !! 1027 syscall_rollback(current, task_pt_regs(current));
1326 /* Trigger a coredump !! 1028 /* Trigger a manual coredump since do_exit skips it. */
1327 force_sig_seccomp(thi !! 1029 seccomp_init_siginfo(&info, this_syscall, data);
1328 } else { !! 1030 do_coredump(&info);
1329 do_exit(SIGSYS); <<
1330 } 1031 }
1331 return -1; /* skip the syscal !! 1032 if (action == SECCOMP_RET_KILL_PROCESS)
>> 1033 do_group_exit(SIGSYS);
>> 1034 else
>> 1035 do_exit(SIGSYS);
1332 } 1036 }
1333 1037
1334 unreachable(); 1038 unreachable();
1335 1039
1336 skip: 1040 skip:
1337 seccomp_log(this_syscall, 0, action, 1041 seccomp_log(this_syscall, 0, action, match ? match->log : false);
1338 return -1; 1042 return -1;
1339 } 1043 }
1340 #else 1044 #else
1341 static int __seccomp_filter(int this_syscall, 1045 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1342 const bool rechec 1046 const bool recheck_after_trace)
1343 { 1047 {
1344 BUG(); 1048 BUG();
1345 <<
1346 return -1; <<
1347 } 1049 }
1348 #endif 1050 #endif
1349 1051
1350 int __secure_computing(const struct seccomp_d 1052 int __secure_computing(const struct seccomp_data *sd)
1351 { 1053 {
1352 int mode = current->seccomp.mode; 1054 int mode = current->seccomp.mode;
1353 int this_syscall; 1055 int this_syscall;
1354 1056
1355 if (IS_ENABLED(CONFIG_CHECKPOINT_REST 1057 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1356 unlikely(current->ptrace & PT_SUS 1058 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1357 return 0; 1059 return 0;
1358 1060
1359 this_syscall = sd ? sd->nr : 1061 this_syscall = sd ? sd->nr :
1360 syscall_get_nr(current, curre !! 1062 syscall_get_nr(current, task_pt_regs(current));
1361 1063
1362 switch (mode) { 1064 switch (mode) {
1363 case SECCOMP_MODE_STRICT: 1065 case SECCOMP_MODE_STRICT:
1364 __secure_computing_strict(thi 1066 __secure_computing_strict(this_syscall); /* may call do_exit */
1365 return 0; 1067 return 0;
1366 case SECCOMP_MODE_FILTER: 1068 case SECCOMP_MODE_FILTER:
1367 return __seccomp_filter(this_ 1069 return __seccomp_filter(this_syscall, sd, false);
1368 /* Surviving SECCOMP_RET_KILL_* must <<
1369 case SECCOMP_MODE_DEAD: <<
1370 WARN_ON_ONCE(1); <<
1371 do_exit(SIGKILL); <<
1372 return -1; <<
1373 default: 1070 default:
1374 BUG(); 1071 BUG();
1375 } 1072 }
1376 } 1073 }
1377 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ 1074 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1378 1075
1379 long prctl_get_seccomp(void) 1076 long prctl_get_seccomp(void)
1380 { 1077 {
1381 return current->seccomp.mode; 1078 return current->seccomp.mode;
1382 } 1079 }
1383 1080
1384 /** 1081 /**
1385 * seccomp_set_mode_strict: internal function 1082 * seccomp_set_mode_strict: internal function for setting strict seccomp
1386 * 1083 *
1387 * Once current->seccomp.mode is non-zero, it 1084 * Once current->seccomp.mode is non-zero, it may not be changed.
1388 * 1085 *
1389 * Returns 0 on success or -EINVAL on failure 1086 * Returns 0 on success or -EINVAL on failure.
1390 */ 1087 */
1391 static long seccomp_set_mode_strict(void) 1088 static long seccomp_set_mode_strict(void)
1392 { 1089 {
1393 const unsigned long seccomp_mode = SE 1090 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
1394 long ret = -EINVAL; 1091 long ret = -EINVAL;
1395 1092
1396 spin_lock_irq(¤t->sighand->sigl 1093 spin_lock_irq(¤t->sighand->siglock);
1397 1094
1398 if (!seccomp_may_assign_mode(seccomp_ 1095 if (!seccomp_may_assign_mode(seccomp_mode))
1399 goto out; 1096 goto out;
1400 1097
1401 #ifdef TIF_NOTSC 1098 #ifdef TIF_NOTSC
1402 disable_TSC(); 1099 disable_TSC();
1403 #endif 1100 #endif
1404 seccomp_assign_mode(current, seccomp_ 1101 seccomp_assign_mode(current, seccomp_mode, 0);
1405 ret = 0; 1102 ret = 0;
1406 1103
1407 out: 1104 out:
1408 spin_unlock_irq(¤t->sighand->si 1105 spin_unlock_irq(¤t->sighand->siglock);
1409 1106
1410 return ret; 1107 return ret;
1411 } 1108 }
1412 1109
1413 #ifdef CONFIG_SECCOMP_FILTER 1110 #ifdef CONFIG_SECCOMP_FILTER
1414 static void seccomp_notify_free(struct seccom 1111 static void seccomp_notify_free(struct seccomp_filter *filter)
1415 { 1112 {
1416 kfree(filter->notif); 1113 kfree(filter->notif);
1417 filter->notif = NULL; 1114 filter->notif = NULL;
1418 } 1115 }
1419 1116
1420 static void seccomp_notify_detach(struct secc 1117 static void seccomp_notify_detach(struct seccomp_filter *filter)
1421 { 1118 {
1422 struct seccomp_knotif *knotif; 1119 struct seccomp_knotif *knotif;
1423 1120
1424 if (!filter) 1121 if (!filter)
1425 return; 1122 return;
1426 1123
1427 mutex_lock(&filter->notify_lock); 1124 mutex_lock(&filter->notify_lock);
1428 1125
1429 /* 1126 /*
1430 * If this file is being closed becau 1127 * If this file is being closed because e.g. the task who owned it
1431 * died, let's wake everyone up who w 1128 * died, let's wake everyone up who was waiting on us.
1432 */ 1129 */
1433 list_for_each_entry(knotif, &filter-> 1130 list_for_each_entry(knotif, &filter->notif->notifications, list) {
1434 if (knotif->state == SECCOMP_ 1131 if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1435 continue; 1132 continue;
1436 1133
1437 knotif->state = SECCOMP_NOTIF 1134 knotif->state = SECCOMP_NOTIFY_REPLIED;
1438 knotif->error = -ENOSYS; 1135 knotif->error = -ENOSYS;
1439 knotif->val = 0; 1136 knotif->val = 0;
1440 1137
1441 /* 1138 /*
1442 * We do not need to wake up 1139 * We do not need to wake up any pending addfd messages, as
1443 * the notifier will do that 1140 * the notifier will do that for us, as this just looks
1444 * like a standard reply. 1141 * like a standard reply.
1445 */ 1142 */
1446 complete(&knotif->ready); 1143 complete(&knotif->ready);
1447 } 1144 }
1448 1145
1449 seccomp_notify_free(filter); 1146 seccomp_notify_free(filter);
1450 mutex_unlock(&filter->notify_lock); 1147 mutex_unlock(&filter->notify_lock);
1451 } 1148 }
1452 1149
1453 static int seccomp_notify_release(struct inod 1150 static int seccomp_notify_release(struct inode *inode, struct file *file)
1454 { 1151 {
1455 struct seccomp_filter *filter = file- 1152 struct seccomp_filter *filter = file->private_data;
1456 1153
1457 seccomp_notify_detach(filter); 1154 seccomp_notify_detach(filter);
1458 __put_seccomp_filter(filter); 1155 __put_seccomp_filter(filter);
1459 return 0; 1156 return 0;
1460 } 1157 }
1461 1158
1462 /* must be called with notif_lock held */ 1159 /* must be called with notif_lock held */
1463 static inline struct seccomp_knotif * 1160 static inline struct seccomp_knotif *
1464 find_notification(struct seccomp_filter *filt 1161 find_notification(struct seccomp_filter *filter, u64 id)
1465 { 1162 {
1466 struct seccomp_knotif *cur; 1163 struct seccomp_knotif *cur;
1467 1164
1468 lockdep_assert_held(&filter->notify_l 1165 lockdep_assert_held(&filter->notify_lock);
1469 1166
1470 list_for_each_entry(cur, &filter->not 1167 list_for_each_entry(cur, &filter->notif->notifications, list) {
1471 if (cur->id == id) 1168 if (cur->id == id)
1472 return cur; 1169 return cur;
1473 } 1170 }
1474 1171
1475 return NULL; 1172 return NULL;
1476 } 1173 }
1477 1174
1478 static int recv_wake_function(wait_queue_entr <<
1479 void *key) <<
1480 { <<
1481 /* Avoid a wakeup if event not intere <<
1482 if (key && !(key_to_poll(key) & (EPOL <<
1483 return 0; <<
1484 return autoremove_wake_function(wait, <<
1485 } <<
1486 <<
1487 static int recv_wait_event(struct seccomp_fil <<
1488 { <<
1489 DEFINE_WAIT_FUNC(wait, recv_wake_func <<
1490 int ret; <<
1491 <<
1492 if (refcount_read(&filter->users) == <<
1493 return 0; <<
1494 <<
1495 if (atomic_dec_if_positive(&filter->n <<
1496 return 0; <<
1497 <<
1498 for (;;) { <<
1499 ret = prepare_to_wait_event(& <<
1500 <<
1501 if (atomic_dec_if_positive(&f <<
1502 break; <<
1503 if (refcount_read(&filter->us <<
1504 break; <<
1505 <<
1506 if (ret) <<
1507 return ret; <<
1508 <<
1509 schedule(); <<
1510 } <<
1511 finish_wait(&filter->wqh, &wait); <<
1512 return 0; <<
1513 } <<
1514 1175
1515 static long seccomp_notify_recv(struct seccom 1176 static long seccomp_notify_recv(struct seccomp_filter *filter,
1516 void __user * 1177 void __user *buf)
1517 { 1178 {
1518 struct seccomp_knotif *knotif = NULL, 1179 struct seccomp_knotif *knotif = NULL, *cur;
1519 struct seccomp_notif unotif; 1180 struct seccomp_notif unotif;
1520 ssize_t ret; 1181 ssize_t ret;
1521 1182
1522 /* Verify that we're not given garbag 1183 /* Verify that we're not given garbage to keep struct extensible. */
1523 ret = check_zeroed_user(buf, sizeof(u 1184 ret = check_zeroed_user(buf, sizeof(unotif));
1524 if (ret < 0) 1185 if (ret < 0)
1525 return ret; 1186 return ret;
1526 if (!ret) 1187 if (!ret)
1527 return -EINVAL; 1188 return -EINVAL;
1528 1189
1529 memset(&unotif, 0, sizeof(unotif)); 1190 memset(&unotif, 0, sizeof(unotif));
1530 1191
1531 ret = recv_wait_event(filter); !! 1192 ret = down_interruptible(&filter->notif->request);
1532 if (ret < 0) 1193 if (ret < 0)
1533 return ret; 1194 return ret;
1534 1195
1535 mutex_lock(&filter->notify_lock); 1196 mutex_lock(&filter->notify_lock);
1536 list_for_each_entry(cur, &filter->not 1197 list_for_each_entry(cur, &filter->notif->notifications, list) {
1537 if (cur->state == SECCOMP_NOT 1198 if (cur->state == SECCOMP_NOTIFY_INIT) {
1538 knotif = cur; 1199 knotif = cur;
1539 break; 1200 break;
1540 } 1201 }
1541 } 1202 }
1542 1203
1543 /* 1204 /*
1544 * If we didn't find a notification, 1205 * If we didn't find a notification, it could be that the task was
1545 * interrupted by a fatal signal betw 1206 * interrupted by a fatal signal between the time we were woken and
1546 * when we were able to acquire the r 1207 * when we were able to acquire the rw lock.
1547 */ 1208 */
1548 if (!knotif) { 1209 if (!knotif) {
1549 ret = -ENOENT; 1210 ret = -ENOENT;
1550 goto out; 1211 goto out;
1551 } 1212 }
1552 1213
1553 unotif.id = knotif->id; 1214 unotif.id = knotif->id;
1554 unotif.pid = task_pid_vnr(knotif->tas 1215 unotif.pid = task_pid_vnr(knotif->task);
1555 unotif.data = *(knotif->data); 1216 unotif.data = *(knotif->data);
1556 1217
1557 knotif->state = SECCOMP_NOTIFY_SENT; 1218 knotif->state = SECCOMP_NOTIFY_SENT;
1558 wake_up_poll(&filter->wqh, EPOLLOUT | 1219 wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM);
1559 ret = 0; 1220 ret = 0;
1560 out: 1221 out:
1561 mutex_unlock(&filter->notify_lock); 1222 mutex_unlock(&filter->notify_lock);
1562 1223
1563 if (ret == 0 && copy_to_user(buf, &un 1224 if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1564 ret = -EFAULT; 1225 ret = -EFAULT;
1565 1226
1566 /* 1227 /*
1567 * Userspace screwed up. To m 1228 * Userspace screwed up. To make sure that we keep this
1568 * notification alive, let's 1229 * notification alive, let's reset it back to INIT. It
1569 * may have died when we rele 1230 * may have died when we released the lock, so we need to make
1570 * sure it's still around. 1231 * sure it's still around.
1571 */ 1232 */
1572 mutex_lock(&filter->notify_lo 1233 mutex_lock(&filter->notify_lock);
1573 knotif = find_notification(fi 1234 knotif = find_notification(filter, unotif.id);
1574 if (knotif) { 1235 if (knotif) {
1575 /* Reset the process <<
1576 if (should_sleep_kill <<
1577 complete(&kno <<
1578 knotif->state = SECCO 1236 knotif->state = SECCOMP_NOTIFY_INIT;
1579 atomic_inc(&filter->n !! 1237 up(&filter->notif->request);
1580 wake_up_poll(&filter- <<
1581 } 1238 }
1582 mutex_unlock(&filter->notify_ 1239 mutex_unlock(&filter->notify_lock);
1583 } 1240 }
1584 1241
1585 return ret; 1242 return ret;
1586 } 1243 }
1587 1244
1588 static long seccomp_notify_send(struct seccom 1245 static long seccomp_notify_send(struct seccomp_filter *filter,
1589 void __user * 1246 void __user *buf)
1590 { 1247 {
1591 struct seccomp_notif_resp resp = {}; 1248 struct seccomp_notif_resp resp = {};
1592 struct seccomp_knotif *knotif; 1249 struct seccomp_knotif *knotif;
1593 long ret; 1250 long ret;
1594 1251
1595 if (copy_from_user(&resp, buf, sizeof 1252 if (copy_from_user(&resp, buf, sizeof(resp)))
1596 return -EFAULT; 1253 return -EFAULT;
1597 1254
1598 if (resp.flags & ~SECCOMP_USER_NOTIF_ 1255 if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1599 return -EINVAL; 1256 return -EINVAL;
1600 1257
1601 if ((resp.flags & SECCOMP_USER_NOTIF_ 1258 if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1602 (resp.error || resp.val)) 1259 (resp.error || resp.val))
1603 return -EINVAL; 1260 return -EINVAL;
1604 1261
1605 ret = mutex_lock_interruptible(&filte 1262 ret = mutex_lock_interruptible(&filter->notify_lock);
1606 if (ret < 0) 1263 if (ret < 0)
1607 return ret; 1264 return ret;
1608 1265
1609 knotif = find_notification(filter, re 1266 knotif = find_notification(filter, resp.id);
1610 if (!knotif) { 1267 if (!knotif) {
1611 ret = -ENOENT; 1268 ret = -ENOENT;
1612 goto out; 1269 goto out;
1613 } 1270 }
1614 1271
1615 /* Allow exactly one reply. */ 1272 /* Allow exactly one reply. */
1616 if (knotif->state != SECCOMP_NOTIFY_S 1273 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1617 ret = -EINPROGRESS; 1274 ret = -EINPROGRESS;
1618 goto out; 1275 goto out;
1619 } 1276 }
1620 1277
1621 ret = 0; 1278 ret = 0;
1622 knotif->state = SECCOMP_NOTIFY_REPLIE 1279 knotif->state = SECCOMP_NOTIFY_REPLIED;
1623 knotif->error = resp.error; 1280 knotif->error = resp.error;
1624 knotif->val = resp.val; 1281 knotif->val = resp.val;
1625 knotif->flags = resp.flags; 1282 knotif->flags = resp.flags;
1626 if (filter->notif->flags & SECCOMP_US !! 1283 complete(&knotif->ready);
1627 complete_on_current_cpu(&knot <<
1628 else <<
1629 complete(&knotif->ready); <<
1630 out: 1284 out:
1631 mutex_unlock(&filter->notify_lock); 1285 mutex_unlock(&filter->notify_lock);
1632 return ret; 1286 return ret;
1633 } 1287 }
1634 1288
1635 static long seccomp_notify_id_valid(struct se 1289 static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1636 void __us 1290 void __user *buf)
1637 { 1291 {
1638 struct seccomp_knotif *knotif; 1292 struct seccomp_knotif *knotif;
1639 u64 id; 1293 u64 id;
1640 long ret; 1294 long ret;
1641 1295
1642 if (copy_from_user(&id, buf, sizeof(i 1296 if (copy_from_user(&id, buf, sizeof(id)))
1643 return -EFAULT; 1297 return -EFAULT;
1644 1298
1645 ret = mutex_lock_interruptible(&filte 1299 ret = mutex_lock_interruptible(&filter->notify_lock);
1646 if (ret < 0) 1300 if (ret < 0)
1647 return ret; 1301 return ret;
1648 1302
1649 knotif = find_notification(filter, id 1303 knotif = find_notification(filter, id);
1650 if (knotif && knotif->state == SECCOM 1304 if (knotif && knotif->state == SECCOMP_NOTIFY_SENT)
1651 ret = 0; 1305 ret = 0;
1652 else 1306 else
1653 ret = -ENOENT; 1307 ret = -ENOENT;
1654 1308
1655 mutex_unlock(&filter->notify_lock); 1309 mutex_unlock(&filter->notify_lock);
1656 return ret; 1310 return ret;
1657 } 1311 }
1658 1312
1659 static long seccomp_notify_set_flags(struct s <<
1660 unsigned <<
1661 { <<
1662 long ret; <<
1663 <<
1664 if (flags & ~SECCOMP_USER_NOTIF_FD_SY <<
1665 return -EINVAL; <<
1666 <<
1667 ret = mutex_lock_interruptible(&filte <<
1668 if (ret < 0) <<
1669 return ret; <<
1670 filter->notif->flags = flags; <<
1671 mutex_unlock(&filter->notify_lock); <<
1672 return 0; <<
1673 } <<
1674 <<
1675 static long seccomp_notify_addfd(struct secco 1313 static long seccomp_notify_addfd(struct seccomp_filter *filter,
1676 struct secco 1314 struct seccomp_notif_addfd __user *uaddfd,
1677 unsigned int 1315 unsigned int size)
1678 { 1316 {
1679 struct seccomp_notif_addfd addfd; 1317 struct seccomp_notif_addfd addfd;
1680 struct seccomp_knotif *knotif; 1318 struct seccomp_knotif *knotif;
1681 struct seccomp_kaddfd kaddfd; 1319 struct seccomp_kaddfd kaddfd;
1682 int ret; 1320 int ret;
1683 1321
1684 BUILD_BUG_ON(sizeof(addfd) < SECCOMP_ 1322 BUILD_BUG_ON(sizeof(addfd) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0);
1685 BUILD_BUG_ON(sizeof(addfd) != SECCOMP 1323 BUILD_BUG_ON(sizeof(addfd) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST);
1686 1324
1687 if (size < SECCOMP_NOTIFY_ADDFD_SIZE_ 1325 if (size < SECCOMP_NOTIFY_ADDFD_SIZE_VER0 || size >= PAGE_SIZE)
1688 return -EINVAL; 1326 return -EINVAL;
1689 1327
1690 ret = copy_struct_from_user(&addfd, s 1328 ret = copy_struct_from_user(&addfd, sizeof(addfd), uaddfd, size);
1691 if (ret) 1329 if (ret)
1692 return ret; 1330 return ret;
1693 1331
1694 if (addfd.newfd_flags & ~O_CLOEXEC) 1332 if (addfd.newfd_flags & ~O_CLOEXEC)
1695 return -EINVAL; 1333 return -EINVAL;
1696 1334
1697 if (addfd.flags & ~(SECCOMP_ADDFD_FLA !! 1335 if (addfd.flags & ~SECCOMP_ADDFD_FLAG_SETFD)
1698 return -EINVAL; 1336 return -EINVAL;
1699 1337
1700 if (addfd.newfd && !(addfd.flags & SE 1338 if (addfd.newfd && !(addfd.flags & SECCOMP_ADDFD_FLAG_SETFD))
1701 return -EINVAL; 1339 return -EINVAL;
1702 1340
1703 kaddfd.file = fget(addfd.srcfd); 1341 kaddfd.file = fget(addfd.srcfd);
1704 if (!kaddfd.file) 1342 if (!kaddfd.file)
1705 return -EBADF; 1343 return -EBADF;
1706 1344
1707 kaddfd.ioctl_flags = addfd.flags; <<
1708 kaddfd.flags = addfd.newfd_flags; 1345 kaddfd.flags = addfd.newfd_flags;
1709 kaddfd.setfd = addfd.flags & SECCOMP_ !! 1346 kaddfd.fd = (addfd.flags & SECCOMP_ADDFD_FLAG_SETFD) ?
1710 kaddfd.fd = addfd.newfd; !! 1347 addfd.newfd : -1;
1711 init_completion(&kaddfd.completion); 1348 init_completion(&kaddfd.completion);
1712 1349
1713 ret = mutex_lock_interruptible(&filte 1350 ret = mutex_lock_interruptible(&filter->notify_lock);
1714 if (ret < 0) 1351 if (ret < 0)
1715 goto out; 1352 goto out;
1716 1353
1717 knotif = find_notification(filter, ad 1354 knotif = find_notification(filter, addfd.id);
1718 if (!knotif) { 1355 if (!knotif) {
1719 ret = -ENOENT; 1356 ret = -ENOENT;
1720 goto out_unlock; 1357 goto out_unlock;
1721 } 1358 }
1722 1359
1723 /* 1360 /*
1724 * We do not want to allow for FD inj 1361 * We do not want to allow for FD injection to occur before the
1725 * notification has been picked up by 1362 * notification has been picked up by a userspace handler, or after
1726 * the notification has been replied 1363 * the notification has been replied to.
1727 */ 1364 */
1728 if (knotif->state != SECCOMP_NOTIFY_S 1365 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1729 ret = -EINPROGRESS; 1366 ret = -EINPROGRESS;
1730 goto out_unlock; 1367 goto out_unlock;
1731 } 1368 }
1732 1369
1733 if (addfd.flags & SECCOMP_ADDFD_FLAG_ <<
1734 /* <<
1735 * Disallow queuing an atomic <<
1736 * some addfd requests still <<
1737 * <<
1738 * There is no clear reason t <<
1739 * the loop on the other side <<
1740 */ <<
1741 if (!list_empty(&knotif->addf <<
1742 ret = -EBUSY; <<
1743 goto out_unlock; <<
1744 } <<
1745 <<
1746 /* Allow exactly only one rep <<
1747 knotif->state = SECCOMP_NOTIF <<
1748 } <<
1749 <<
1750 list_add(&kaddfd.list, &knotif->addfd 1370 list_add(&kaddfd.list, &knotif->addfd);
1751 complete(&knotif->ready); 1371 complete(&knotif->ready);
1752 mutex_unlock(&filter->notify_lock); 1372 mutex_unlock(&filter->notify_lock);
1753 1373
1754 /* Now we wait for it to be processed 1374 /* Now we wait for it to be processed or be interrupted */
1755 ret = wait_for_completion_interruptib 1375 ret = wait_for_completion_interruptible(&kaddfd.completion);
1756 if (ret == 0) { 1376 if (ret == 0) {
1757 /* 1377 /*
1758 * We had a successful comple 1378 * We had a successful completion. The other side has already
1759 * removed us from the addfd 1379 * removed us from the addfd queue, and
1760 * wait_for_completion_interr 1380 * wait_for_completion_interruptible has a memory barrier upon
1761 * success that lets us read 1381 * success that lets us read this value directly without
1762 * locking. 1382 * locking.
1763 */ 1383 */
1764 ret = kaddfd.ret; 1384 ret = kaddfd.ret;
1765 goto out; 1385 goto out;
1766 } 1386 }
1767 1387
1768 mutex_lock(&filter->notify_lock); 1388 mutex_lock(&filter->notify_lock);
1769 /* 1389 /*
1770 * Even though we were woken up by a 1390 * Even though we were woken up by a signal and not a successful
1771 * completion, a completion may have 1391 * completion, a completion may have happened in the mean time.
1772 * 1392 *
1773 * We need to check again if the addf 1393 * We need to check again if the addfd request has been handled,
1774 * and if not, we will remove it from 1394 * and if not, we will remove it from the queue.
1775 */ 1395 */
1776 if (list_empty(&kaddfd.list)) 1396 if (list_empty(&kaddfd.list))
1777 ret = kaddfd.ret; 1397 ret = kaddfd.ret;
1778 else 1398 else
1779 list_del(&kaddfd.list); 1399 list_del(&kaddfd.list);
1780 1400
1781 out_unlock: 1401 out_unlock:
1782 mutex_unlock(&filter->notify_lock); 1402 mutex_unlock(&filter->notify_lock);
1783 out: 1403 out:
1784 fput(kaddfd.file); 1404 fput(kaddfd.file);
1785 1405
1786 return ret; 1406 return ret;
1787 } 1407 }
1788 1408
1789 static long seccomp_notify_ioctl(struct file 1409 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1790 unsigned lon 1410 unsigned long arg)
1791 { 1411 {
1792 struct seccomp_filter *filter = file- 1412 struct seccomp_filter *filter = file->private_data;
1793 void __user *buf = (void __user *)arg 1413 void __user *buf = (void __user *)arg;
1794 1414
1795 /* Fixed-size ioctls */ 1415 /* Fixed-size ioctls */
1796 switch (cmd) { 1416 switch (cmd) {
1797 case SECCOMP_IOCTL_NOTIF_RECV: 1417 case SECCOMP_IOCTL_NOTIF_RECV:
1798 return seccomp_notify_recv(fi 1418 return seccomp_notify_recv(filter, buf);
1799 case SECCOMP_IOCTL_NOTIF_SEND: 1419 case SECCOMP_IOCTL_NOTIF_SEND:
1800 return seccomp_notify_send(fi 1420 return seccomp_notify_send(filter, buf);
1801 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRO 1421 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR:
1802 case SECCOMP_IOCTL_NOTIF_ID_VALID: 1422 case SECCOMP_IOCTL_NOTIF_ID_VALID:
1803 return seccomp_notify_id_vali 1423 return seccomp_notify_id_valid(filter, buf);
1804 case SECCOMP_IOCTL_NOTIF_SET_FLAGS: <<
1805 return seccomp_notify_set_fla <<
1806 } 1424 }
1807 1425
1808 /* Extensible Argument ioctls */ 1426 /* Extensible Argument ioctls */
1809 #define EA_IOCTL(cmd) ((cmd) & ~(IOC_INOUT 1427 #define EA_IOCTL(cmd) ((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1810 switch (EA_IOCTL(cmd)) { 1428 switch (EA_IOCTL(cmd)) {
1811 case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADD 1429 case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD):
1812 return seccomp_notify_addfd(f 1430 return seccomp_notify_addfd(filter, buf, _IOC_SIZE(cmd));
1813 default: 1431 default:
1814 return -EINVAL; 1432 return -EINVAL;
1815 } 1433 }
1816 } 1434 }
1817 1435
1818 static __poll_t seccomp_notify_poll(struct fi 1436 static __poll_t seccomp_notify_poll(struct file *file,
1819 struct po 1437 struct poll_table_struct *poll_tab)
1820 { 1438 {
1821 struct seccomp_filter *filter = file- 1439 struct seccomp_filter *filter = file->private_data;
1822 __poll_t ret = 0; 1440 __poll_t ret = 0;
1823 struct seccomp_knotif *cur; 1441 struct seccomp_knotif *cur;
1824 1442
1825 poll_wait(file, &filter->wqh, poll_ta 1443 poll_wait(file, &filter->wqh, poll_tab);
1826 1444
1827 if (mutex_lock_interruptible(&filter- 1445 if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1828 return EPOLLERR; 1446 return EPOLLERR;
1829 1447
1830 list_for_each_entry(cur, &filter->not 1448 list_for_each_entry(cur, &filter->notif->notifications, list) {
1831 if (cur->state == SECCOMP_NOT 1449 if (cur->state == SECCOMP_NOTIFY_INIT)
1832 ret |= EPOLLIN | EPOL 1450 ret |= EPOLLIN | EPOLLRDNORM;
1833 if (cur->state == SECCOMP_NOT 1451 if (cur->state == SECCOMP_NOTIFY_SENT)
1834 ret |= EPOLLOUT | EPO 1452 ret |= EPOLLOUT | EPOLLWRNORM;
1835 if ((ret & EPOLLIN) && (ret & 1453 if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1836 break; 1454 break;
1837 } 1455 }
1838 1456
1839 mutex_unlock(&filter->notify_lock); 1457 mutex_unlock(&filter->notify_lock);
1840 1458
1841 if (refcount_read(&filter->users) == 1459 if (refcount_read(&filter->users) == 0)
1842 ret |= EPOLLHUP; 1460 ret |= EPOLLHUP;
1843 1461
1844 return ret; 1462 return ret;
1845 } 1463 }
1846 1464
1847 static const struct file_operations seccomp_n 1465 static const struct file_operations seccomp_notify_ops = {
1848 .poll = seccomp_notify_poll, 1466 .poll = seccomp_notify_poll,
1849 .release = seccomp_notify_release, 1467 .release = seccomp_notify_release,
1850 .unlocked_ioctl = seccomp_notify_ioct 1468 .unlocked_ioctl = seccomp_notify_ioctl,
1851 .compat_ioctl = seccomp_notify_ioctl, 1469 .compat_ioctl = seccomp_notify_ioctl,
1852 }; 1470 };
1853 1471
1854 static struct file *init_listener(struct secc 1472 static struct file *init_listener(struct seccomp_filter *filter)
1855 { 1473 {
1856 struct file *ret; 1474 struct file *ret;
1857 1475
1858 ret = ERR_PTR(-ENOMEM); 1476 ret = ERR_PTR(-ENOMEM);
1859 filter->notif = kzalloc(sizeof(*(filt 1477 filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1860 if (!filter->notif) 1478 if (!filter->notif)
1861 goto out; 1479 goto out;
1862 1480
>> 1481 sema_init(&filter->notif->request, 0);
1863 filter->notif->next_id = get_random_u 1482 filter->notif->next_id = get_random_u64();
1864 INIT_LIST_HEAD(&filter->notif->notifi 1483 INIT_LIST_HEAD(&filter->notif->notifications);
1865 1484
1866 ret = anon_inode_getfile("seccomp not 1485 ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1867 filter, O_RD 1486 filter, O_RDWR);
1868 if (IS_ERR(ret)) 1487 if (IS_ERR(ret))
1869 goto out_notif; 1488 goto out_notif;
1870 1489
1871 /* The file has a reference to it now 1490 /* The file has a reference to it now */
1872 __get_seccomp_filter(filter); 1491 __get_seccomp_filter(filter);
1873 1492
1874 out_notif: 1493 out_notif:
1875 if (IS_ERR(ret)) 1494 if (IS_ERR(ret))
1876 seccomp_notify_free(filter); 1495 seccomp_notify_free(filter);
1877 out: 1496 out:
1878 return ret; 1497 return ret;
1879 } 1498 }
1880 1499
1881 /* 1500 /*
1882 * Does @new_child have a listener while an a 1501 * Does @new_child have a listener while an ancestor also has a listener?
1883 * If so, we'll want to reject this filter. 1502 * If so, we'll want to reject this filter.
1884 * This only has to be tested for the current 1503 * This only has to be tested for the current process, even in the TSYNC case,
1885 * because TSYNC installs @child with the sam 1504 * because TSYNC installs @child with the same parent on all threads.
1886 * Note that @new_child is not hooked up to i 1505 * Note that @new_child is not hooked up to its parent at this point yet, so
1887 * we use current->seccomp.filter. 1506 * we use current->seccomp.filter.
1888 */ 1507 */
1889 static bool has_duplicate_listener(struct sec 1508 static bool has_duplicate_listener(struct seccomp_filter *new_child)
1890 { 1509 {
1891 struct seccomp_filter *cur; 1510 struct seccomp_filter *cur;
1892 1511
1893 /* must be protected against concurre 1512 /* must be protected against concurrent TSYNC */
1894 lockdep_assert_held(¤t->sighand 1513 lockdep_assert_held(¤t->sighand->siglock);
1895 1514
1896 if (!new_child->notif) 1515 if (!new_child->notif)
1897 return false; 1516 return false;
1898 for (cur = current->seccomp.filter; c 1517 for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1899 if (cur->notif) 1518 if (cur->notif)
1900 return true; 1519 return true;
1901 } 1520 }
1902 1521
1903 return false; 1522 return false;
1904 } 1523 }
1905 1524
1906 /** 1525 /**
1907 * seccomp_set_mode_filter: internal function 1526 * seccomp_set_mode_filter: internal function for setting seccomp filter
1908 * @flags: flags to change filter behavior 1527 * @flags: flags to change filter behavior
1909 * @filter: struct sock_fprog containing filt 1528 * @filter: struct sock_fprog containing filter
1910 * 1529 *
1911 * This function may be called repeatedly to 1530 * This function may be called repeatedly to install additional filters.
1912 * Every filter successfully installed will b 1531 * Every filter successfully installed will be evaluated (in reverse order)
1913 * for each system call the task makes. 1532 * for each system call the task makes.
1914 * 1533 *
1915 * Once current->seccomp.mode is non-zero, it 1534 * Once current->seccomp.mode is non-zero, it may not be changed.
1916 * 1535 *
1917 * Returns 0 on success or -EINVAL on failure 1536 * Returns 0 on success or -EINVAL on failure.
1918 */ 1537 */
1919 static long seccomp_set_mode_filter(unsigned 1538 static long seccomp_set_mode_filter(unsigned int flags,
1920 const cha 1539 const char __user *filter)
1921 { 1540 {
1922 const unsigned long seccomp_mode = SE 1541 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1923 struct seccomp_filter *prepared = NUL 1542 struct seccomp_filter *prepared = NULL;
1924 long ret = -EINVAL; 1543 long ret = -EINVAL;
1925 int listener = -1; 1544 int listener = -1;
1926 struct file *listener_f = NULL; 1545 struct file *listener_f = NULL;
1927 1546
1928 /* Validate flags. */ 1547 /* Validate flags. */
1929 if (flags & ~SECCOMP_FILTER_FLAG_MASK 1548 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1930 return -EINVAL; 1549 return -EINVAL;
1931 1550
1932 /* 1551 /*
1933 * In the successful case, NEW_LISTEN 1552 * In the successful case, NEW_LISTENER returns the new listener fd.
1934 * But in the failure case, TSYNC ret 1553 * But in the failure case, TSYNC returns the thread that died. If you
1935 * combine these two flags, there's n 1554 * combine these two flags, there's no way to tell whether something
1936 * succeeded or failed. So, let's dis 1555 * succeeded or failed. So, let's disallow this combination if the user
1937 * has not explicitly requested no er 1556 * has not explicitly requested no errors from TSYNC.
1938 */ 1557 */
1939 if ((flags & SECCOMP_FILTER_FLAG_TSYN 1558 if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
1940 (flags & SECCOMP_FILTER_FLAG_NEW_ 1559 (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1941 ((flags & SECCOMP_FILTER_FLAG_TSY 1560 ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
1942 return -EINVAL; 1561 return -EINVAL;
1943 1562
1944 /* <<
1945 * The SECCOMP_FILTER_FLAG_WAIT_KILLA <<
1946 * without the SECCOMP_FILTER_FLAG_NE <<
1947 */ <<
1948 if ((flags & SECCOMP_FILTER_FLAG_WAIT <<
1949 ((flags & SECCOMP_FILTER_FLAG_NEW <<
1950 return -EINVAL; <<
1951 <<
1952 /* Prepare the new filter before hold 1563 /* Prepare the new filter before holding any locks. */
1953 prepared = seccomp_prepare_user_filte 1564 prepared = seccomp_prepare_user_filter(filter);
1954 if (IS_ERR(prepared)) 1565 if (IS_ERR(prepared))
1955 return PTR_ERR(prepared); 1566 return PTR_ERR(prepared);
1956 1567
1957 if (flags & SECCOMP_FILTER_FLAG_NEW_L 1568 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1958 listener = get_unused_fd_flag 1569 listener = get_unused_fd_flags(O_CLOEXEC);
1959 if (listener < 0) { 1570 if (listener < 0) {
1960 ret = listener; 1571 ret = listener;
1961 goto out_free; 1572 goto out_free;
1962 } 1573 }
1963 1574
1964 listener_f = init_listener(pr 1575 listener_f = init_listener(prepared);
1965 if (IS_ERR(listener_f)) { 1576 if (IS_ERR(listener_f)) {
1966 put_unused_fd(listene 1577 put_unused_fd(listener);
1967 ret = PTR_ERR(listene 1578 ret = PTR_ERR(listener_f);
1968 goto out_free; 1579 goto out_free;
1969 } 1580 }
1970 } 1581 }
1971 1582
1972 /* 1583 /*
1973 * Make sure we cannot change seccomp 1584 * Make sure we cannot change seccomp or nnp state via TSYNC
1974 * while another thread is in the mid 1585 * while another thread is in the middle of calling exec.
1975 */ 1586 */
1976 if (flags & SECCOMP_FILTER_FLAG_TSYNC 1587 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1977 mutex_lock_killable(¤t->sig 1588 mutex_lock_killable(¤t->signal->cred_guard_mutex))
1978 goto out_put_fd; 1589 goto out_put_fd;
1979 1590
1980 spin_lock_irq(¤t->sighand->sigl 1591 spin_lock_irq(¤t->sighand->siglock);
1981 1592
1982 if (!seccomp_may_assign_mode(seccomp_ 1593 if (!seccomp_may_assign_mode(seccomp_mode))
1983 goto out; 1594 goto out;
1984 1595
1985 if (has_duplicate_listener(prepared)) 1596 if (has_duplicate_listener(prepared)) {
1986 ret = -EBUSY; 1597 ret = -EBUSY;
1987 goto out; 1598 goto out;
1988 } 1599 }
1989 1600
1990 ret = seccomp_attach_filter(flags, pr 1601 ret = seccomp_attach_filter(flags, prepared);
1991 if (ret) 1602 if (ret)
1992 goto out; 1603 goto out;
1993 /* Do not free the successfully attac 1604 /* Do not free the successfully attached filter. */
1994 prepared = NULL; 1605 prepared = NULL;
1995 1606
1996 seccomp_assign_mode(current, seccomp_ 1607 seccomp_assign_mode(current, seccomp_mode, flags);
1997 out: 1608 out:
1998 spin_unlock_irq(¤t->sighand->si 1609 spin_unlock_irq(¤t->sighand->siglock);
1999 if (flags & SECCOMP_FILTER_FLAG_TSYNC 1610 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
2000 mutex_unlock(¤t->signal 1611 mutex_unlock(¤t->signal->cred_guard_mutex);
2001 out_put_fd: 1612 out_put_fd:
2002 if (flags & SECCOMP_FILTER_FLAG_NEW_L 1613 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
2003 if (ret) { 1614 if (ret) {
2004 listener_f->private_d 1615 listener_f->private_data = NULL;
2005 fput(listener_f); 1616 fput(listener_f);
2006 put_unused_fd(listene 1617 put_unused_fd(listener);
2007 seccomp_notify_detach 1618 seccomp_notify_detach(prepared);
2008 } else { 1619 } else {
2009 fd_install(listener, 1620 fd_install(listener, listener_f);
2010 ret = listener; 1621 ret = listener;
2011 } 1622 }
2012 } 1623 }
2013 out_free: 1624 out_free:
2014 seccomp_filter_free(prepared); 1625 seccomp_filter_free(prepared);
2015 return ret; 1626 return ret;
2016 } 1627 }
2017 #else 1628 #else
2018 static inline long seccomp_set_mode_filter(un 1629 static inline long seccomp_set_mode_filter(unsigned int flags,
2019 co 1630 const char __user *filter)
2020 { 1631 {
2021 return -EINVAL; 1632 return -EINVAL;
2022 } 1633 }
2023 #endif 1634 #endif
2024 1635
2025 static long seccomp_get_action_avail(const ch 1636 static long seccomp_get_action_avail(const char __user *uaction)
2026 { 1637 {
2027 u32 action; 1638 u32 action;
2028 1639
2029 if (copy_from_user(&action, uaction, 1640 if (copy_from_user(&action, uaction, sizeof(action)))
2030 return -EFAULT; 1641 return -EFAULT;
2031 1642
2032 switch (action) { 1643 switch (action) {
2033 case SECCOMP_RET_KILL_PROCESS: 1644 case SECCOMP_RET_KILL_PROCESS:
2034 case SECCOMP_RET_KILL_THREAD: 1645 case SECCOMP_RET_KILL_THREAD:
2035 case SECCOMP_RET_TRAP: 1646 case SECCOMP_RET_TRAP:
2036 case SECCOMP_RET_ERRNO: 1647 case SECCOMP_RET_ERRNO:
2037 case SECCOMP_RET_USER_NOTIF: 1648 case SECCOMP_RET_USER_NOTIF:
2038 case SECCOMP_RET_TRACE: 1649 case SECCOMP_RET_TRACE:
2039 case SECCOMP_RET_LOG: 1650 case SECCOMP_RET_LOG:
2040 case SECCOMP_RET_ALLOW: 1651 case SECCOMP_RET_ALLOW:
2041 break; 1652 break;
2042 default: 1653 default:
2043 return -EOPNOTSUPP; 1654 return -EOPNOTSUPP;
2044 } 1655 }
2045 1656
2046 return 0; 1657 return 0;
2047 } 1658 }
2048 1659
2049 static long seccomp_get_notif_sizes(void __us 1660 static long seccomp_get_notif_sizes(void __user *usizes)
2050 { 1661 {
2051 struct seccomp_notif_sizes sizes = { 1662 struct seccomp_notif_sizes sizes = {
2052 .seccomp_notif = sizeof(struc 1663 .seccomp_notif = sizeof(struct seccomp_notif),
2053 .seccomp_notif_resp = sizeof( 1664 .seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
2054 .seccomp_data = sizeof(struct 1665 .seccomp_data = sizeof(struct seccomp_data),
2055 }; 1666 };
2056 1667
2057 if (copy_to_user(usizes, &sizes, size 1668 if (copy_to_user(usizes, &sizes, sizeof(sizes)))
2058 return -EFAULT; 1669 return -EFAULT;
2059 1670
2060 return 0; 1671 return 0;
2061 } 1672 }
2062 1673
2063 /* Common entry point for both prctl and sysc 1674 /* Common entry point for both prctl and syscall. */
2064 static long do_seccomp(unsigned int op, unsig 1675 static long do_seccomp(unsigned int op, unsigned int flags,
2065 void __user *uargs) 1676 void __user *uargs)
2066 { 1677 {
2067 switch (op) { 1678 switch (op) {
2068 case SECCOMP_SET_MODE_STRICT: 1679 case SECCOMP_SET_MODE_STRICT:
2069 if (flags != 0 || uargs != NU 1680 if (flags != 0 || uargs != NULL)
2070 return -EINVAL; 1681 return -EINVAL;
2071 return seccomp_set_mode_stric 1682 return seccomp_set_mode_strict();
2072 case SECCOMP_SET_MODE_FILTER: 1683 case SECCOMP_SET_MODE_FILTER:
2073 return seccomp_set_mode_filte 1684 return seccomp_set_mode_filter(flags, uargs);
2074 case SECCOMP_GET_ACTION_AVAIL: 1685 case SECCOMP_GET_ACTION_AVAIL:
2075 if (flags != 0) 1686 if (flags != 0)
2076 return -EINVAL; 1687 return -EINVAL;
2077 1688
2078 return seccomp_get_action_ava 1689 return seccomp_get_action_avail(uargs);
2079 case SECCOMP_GET_NOTIF_SIZES: 1690 case SECCOMP_GET_NOTIF_SIZES:
2080 if (flags != 0) 1691 if (flags != 0)
2081 return -EINVAL; 1692 return -EINVAL;
2082 1693
2083 return seccomp_get_notif_size 1694 return seccomp_get_notif_sizes(uargs);
2084 default: 1695 default:
2085 return -EINVAL; 1696 return -EINVAL;
2086 } 1697 }
2087 } 1698 }
2088 1699
2089 SYSCALL_DEFINE3(seccomp, unsigned int, op, un 1700 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
2090 void __user *, uargs 1701 void __user *, uargs)
2091 { 1702 {
2092 return do_seccomp(op, flags, uargs); 1703 return do_seccomp(op, flags, uargs);
2093 } 1704 }
2094 1705
2095 /** 1706 /**
2096 * prctl_set_seccomp: configures current->sec 1707 * prctl_set_seccomp: configures current->seccomp.mode
2097 * @seccomp_mode: requested mode to use 1708 * @seccomp_mode: requested mode to use
2098 * @filter: optional struct sock_fprog for us 1709 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
2099 * 1710 *
2100 * Returns 0 on success or -EINVAL on failure 1711 * Returns 0 on success or -EINVAL on failure.
2101 */ 1712 */
2102 long prctl_set_seccomp(unsigned long seccomp_ 1713 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
2103 { 1714 {
2104 unsigned int op; 1715 unsigned int op;
2105 void __user *uargs; 1716 void __user *uargs;
2106 1717
2107 switch (seccomp_mode) { 1718 switch (seccomp_mode) {
2108 case SECCOMP_MODE_STRICT: 1719 case SECCOMP_MODE_STRICT:
2109 op = SECCOMP_SET_MODE_STRICT; 1720 op = SECCOMP_SET_MODE_STRICT;
2110 /* 1721 /*
2111 * Setting strict mode throug 1722 * Setting strict mode through prctl always ignored filter,
2112 * so make sure it is always 1723 * so make sure it is always NULL here to pass the internal
2113 * check in do_seccomp(). 1724 * check in do_seccomp().
2114 */ 1725 */
2115 uargs = NULL; 1726 uargs = NULL;
2116 break; 1727 break;
2117 case SECCOMP_MODE_FILTER: 1728 case SECCOMP_MODE_FILTER:
2118 op = SECCOMP_SET_MODE_FILTER; 1729 op = SECCOMP_SET_MODE_FILTER;
2119 uargs = filter; 1730 uargs = filter;
2120 break; 1731 break;
2121 default: 1732 default:
2122 return -EINVAL; 1733 return -EINVAL;
2123 } 1734 }
2124 1735
2125 /* prctl interface doesn't have flags 1736 /* prctl interface doesn't have flags, so they are always zero. */
2126 return do_seccomp(op, 0, uargs); 1737 return do_seccomp(op, 0, uargs);
2127 } 1738 }
2128 1739
2129 #if defined(CONFIG_SECCOMP_FILTER) && defined 1740 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
2130 static struct seccomp_filter *get_nth_filter( 1741 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
2131 1742 unsigned long filter_off)
2132 { 1743 {
2133 struct seccomp_filter *orig, *filter; 1744 struct seccomp_filter *orig, *filter;
2134 unsigned long count; 1745 unsigned long count;
2135 1746
2136 /* 1747 /*
2137 * Note: this is only correct because 1748 * Note: this is only correct because the caller should be the (ptrace)
2138 * tracer of the task, otherwise lock 1749 * tracer of the task, otherwise lock_task_sighand is needed.
2139 */ 1750 */
2140 spin_lock_irq(&task->sighand->siglock 1751 spin_lock_irq(&task->sighand->siglock);
2141 1752
2142 if (task->seccomp.mode != SECCOMP_MOD 1753 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
2143 spin_unlock_irq(&task->sighan 1754 spin_unlock_irq(&task->sighand->siglock);
2144 return ERR_PTR(-EINVAL); 1755 return ERR_PTR(-EINVAL);
2145 } 1756 }
2146 1757
2147 orig = task->seccomp.filter; 1758 orig = task->seccomp.filter;
2148 __get_seccomp_filter(orig); 1759 __get_seccomp_filter(orig);
2149 spin_unlock_irq(&task->sighand->siglo 1760 spin_unlock_irq(&task->sighand->siglock);
2150 1761
2151 count = 0; 1762 count = 0;
2152 for (filter = orig; filter; filter = 1763 for (filter = orig; filter; filter = filter->prev)
2153 count++; 1764 count++;
2154 1765
2155 if (filter_off >= count) { 1766 if (filter_off >= count) {
2156 filter = ERR_PTR(-ENOENT); 1767 filter = ERR_PTR(-ENOENT);
2157 goto out; 1768 goto out;
2158 } 1769 }
2159 1770
2160 count -= filter_off; 1771 count -= filter_off;
2161 for (filter = orig; filter && count > 1772 for (filter = orig; filter && count > 1; filter = filter->prev)
2162 count--; 1773 count--;
2163 1774
2164 if (WARN_ON(count != 1 || !filter)) { 1775 if (WARN_ON(count != 1 || !filter)) {
2165 filter = ERR_PTR(-ENOENT); 1776 filter = ERR_PTR(-ENOENT);
2166 goto out; 1777 goto out;
2167 } 1778 }
2168 1779
2169 __get_seccomp_filter(filter); 1780 __get_seccomp_filter(filter);
2170 1781
2171 out: 1782 out:
2172 __put_seccomp_filter(orig); 1783 __put_seccomp_filter(orig);
2173 return filter; 1784 return filter;
2174 } 1785 }
2175 1786
2176 long seccomp_get_filter(struct task_struct *t 1787 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
2177 void __user *data) 1788 void __user *data)
2178 { 1789 {
2179 struct seccomp_filter *filter; 1790 struct seccomp_filter *filter;
2180 struct sock_fprog_kern *fprog; 1791 struct sock_fprog_kern *fprog;
2181 long ret; 1792 long ret;
2182 1793
2183 if (!capable(CAP_SYS_ADMIN) || 1794 if (!capable(CAP_SYS_ADMIN) ||
2184 current->seccomp.mode != SECCOMP_ 1795 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2185 return -EACCES; 1796 return -EACCES;
2186 } 1797 }
2187 1798
2188 filter = get_nth_filter(task, filter_ 1799 filter = get_nth_filter(task, filter_off);
2189 if (IS_ERR(filter)) 1800 if (IS_ERR(filter))
2190 return PTR_ERR(filter); 1801 return PTR_ERR(filter);
2191 1802
2192 fprog = filter->prog->orig_prog; 1803 fprog = filter->prog->orig_prog;
2193 if (!fprog) { 1804 if (!fprog) {
2194 /* This must be a new non-cBP 1805 /* This must be a new non-cBPF filter, since we save
2195 * every cBPF filter's orig_p 1806 * every cBPF filter's orig_prog above when
2196 * CONFIG_CHECKPOINT_RESTORE 1807 * CONFIG_CHECKPOINT_RESTORE is enabled.
2197 */ 1808 */
2198 ret = -EMEDIUMTYPE; 1809 ret = -EMEDIUMTYPE;
2199 goto out; 1810 goto out;
2200 } 1811 }
2201 1812
2202 ret = fprog->len; 1813 ret = fprog->len;
2203 if (!data) 1814 if (!data)
2204 goto out; 1815 goto out;
2205 1816
2206 if (copy_to_user(data, fprog->filter, 1817 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
2207 ret = -EFAULT; 1818 ret = -EFAULT;
2208 1819
2209 out: 1820 out:
2210 __put_seccomp_filter(filter); 1821 __put_seccomp_filter(filter);
2211 return ret; 1822 return ret;
2212 } 1823 }
2213 1824
2214 long seccomp_get_metadata(struct task_struct 1825 long seccomp_get_metadata(struct task_struct *task,
2215 unsigned long size, 1826 unsigned long size, void __user *data)
2216 { 1827 {
2217 long ret; 1828 long ret;
2218 struct seccomp_filter *filter; 1829 struct seccomp_filter *filter;
2219 struct seccomp_metadata kmd = {}; 1830 struct seccomp_metadata kmd = {};
2220 1831
2221 if (!capable(CAP_SYS_ADMIN) || 1832 if (!capable(CAP_SYS_ADMIN) ||
2222 current->seccomp.mode != SECCOMP_ 1833 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2223 return -EACCES; 1834 return -EACCES;
2224 } 1835 }
2225 1836
2226 size = min_t(unsigned long, size, siz 1837 size = min_t(unsigned long, size, sizeof(kmd));
2227 1838
2228 if (size < sizeof(kmd.filter_off)) 1839 if (size < sizeof(kmd.filter_off))
2229 return -EINVAL; 1840 return -EINVAL;
2230 1841
2231 if (copy_from_user(&kmd.filter_off, d 1842 if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
2232 return -EFAULT; 1843 return -EFAULT;
2233 1844
2234 filter = get_nth_filter(task, kmd.fil 1845 filter = get_nth_filter(task, kmd.filter_off);
2235 if (IS_ERR(filter)) 1846 if (IS_ERR(filter))
2236 return PTR_ERR(filter); 1847 return PTR_ERR(filter);
2237 1848
2238 if (filter->log) 1849 if (filter->log)
2239 kmd.flags |= SECCOMP_FILTER_F 1850 kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
2240 1851
2241 ret = size; 1852 ret = size;
2242 if (copy_to_user(data, &kmd, size)) 1853 if (copy_to_user(data, &kmd, size))
2243 ret = -EFAULT; 1854 ret = -EFAULT;
2244 1855
2245 __put_seccomp_filter(filter); 1856 __put_seccomp_filter(filter);
2246 return ret; 1857 return ret;
2247 } 1858 }
2248 #endif 1859 #endif
2249 1860
2250 #ifdef CONFIG_SYSCTL 1861 #ifdef CONFIG_SYSCTL
2251 1862
2252 /* Human readable action names for friendly s 1863 /* Human readable action names for friendly sysctl interaction */
2253 #define SECCOMP_RET_KILL_PROCESS_NAME "kill 1864 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
2254 #define SECCOMP_RET_KILL_THREAD_NAME "kill 1865 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
2255 #define SECCOMP_RET_TRAP_NAME "trap 1866 #define SECCOMP_RET_TRAP_NAME "trap"
2256 #define SECCOMP_RET_ERRNO_NAME "errn 1867 #define SECCOMP_RET_ERRNO_NAME "errno"
2257 #define SECCOMP_RET_USER_NOTIF_NAME "user 1868 #define SECCOMP_RET_USER_NOTIF_NAME "user_notif"
2258 #define SECCOMP_RET_TRACE_NAME "trac 1869 #define SECCOMP_RET_TRACE_NAME "trace"
2259 #define SECCOMP_RET_LOG_NAME "log" 1870 #define SECCOMP_RET_LOG_NAME "log"
2260 #define SECCOMP_RET_ALLOW_NAME "allo 1871 #define SECCOMP_RET_ALLOW_NAME "allow"
2261 1872
2262 static const char seccomp_actions_avail[] = 1873 static const char seccomp_actions_avail[] =
2263 SECCOMP_RET_K 1874 SECCOMP_RET_KILL_PROCESS_NAME " "
2264 SECCOMP_RET_K 1875 SECCOMP_RET_KILL_THREAD_NAME " "
2265 SECCOMP_RET_T 1876 SECCOMP_RET_TRAP_NAME " "
2266 SECCOMP_RET_E 1877 SECCOMP_RET_ERRNO_NAME " "
2267 SECCOMP_RET_U 1878 SECCOMP_RET_USER_NOTIF_NAME " "
2268 SECCOMP_RET_T 1879 SECCOMP_RET_TRACE_NAME " "
2269 SECCOMP_RET_L 1880 SECCOMP_RET_LOG_NAME " "
2270 SECCOMP_RET_A 1881 SECCOMP_RET_ALLOW_NAME;
2271 1882
2272 struct seccomp_log_name { 1883 struct seccomp_log_name {
2273 u32 log; 1884 u32 log;
2274 const char *name; 1885 const char *name;
2275 }; 1886 };
2276 1887
2277 static const struct seccomp_log_name seccomp_ 1888 static const struct seccomp_log_name seccomp_log_names[] = {
2278 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_R 1889 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
2279 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RE 1890 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
2280 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_ 1891 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
2281 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRN 1892 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
2282 { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET 1893 { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
2283 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRAC 1894 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
2284 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NA 1895 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
2285 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLO 1896 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
2286 { } 1897 { }
2287 }; 1898 };
2288 1899
2289 static bool seccomp_names_from_actions_logged 1900 static bool seccomp_names_from_actions_logged(char *names, size_t size,
2290 1901 u32 actions_logged,
2291 1902 const char *sep)
2292 { 1903 {
2293 const struct seccomp_log_name *cur; 1904 const struct seccomp_log_name *cur;
2294 bool append_sep = false; 1905 bool append_sep = false;
2295 1906
2296 for (cur = seccomp_log_names; cur->na 1907 for (cur = seccomp_log_names; cur->name && size; cur++) {
2297 ssize_t ret; 1908 ssize_t ret;
2298 1909
2299 if (!(actions_logged & cur->l 1910 if (!(actions_logged & cur->log))
2300 continue; 1911 continue;
2301 1912
2302 if (append_sep) { 1913 if (append_sep) {
2303 ret = strscpy(names, 1914 ret = strscpy(names, sep, size);
2304 if (ret < 0) 1915 if (ret < 0)
2305 return false; 1916 return false;
2306 1917
2307 names += ret; 1918 names += ret;
2308 size -= ret; 1919 size -= ret;
2309 } else 1920 } else
2310 append_sep = true; 1921 append_sep = true;
2311 1922
2312 ret = strscpy(names, cur->nam 1923 ret = strscpy(names, cur->name, size);
2313 if (ret < 0) 1924 if (ret < 0)
2314 return false; 1925 return false;
2315 1926
2316 names += ret; 1927 names += ret;
2317 size -= ret; 1928 size -= ret;
2318 } 1929 }
2319 1930
2320 return true; 1931 return true;
2321 } 1932 }
2322 1933
2323 static bool seccomp_action_logged_from_name(u 1934 static bool seccomp_action_logged_from_name(u32 *action_logged,
2324 c 1935 const char *name)
2325 { 1936 {
2326 const struct seccomp_log_name *cur; 1937 const struct seccomp_log_name *cur;
2327 1938
2328 for (cur = seccomp_log_names; cur->na 1939 for (cur = seccomp_log_names; cur->name; cur++) {
2329 if (!strcmp(cur->name, name)) 1940 if (!strcmp(cur->name, name)) {
2330 *action_logged = cur- 1941 *action_logged = cur->log;
2331 return true; 1942 return true;
2332 } 1943 }
2333 } 1944 }
2334 1945
2335 return false; 1946 return false;
2336 } 1947 }
2337 1948
2338 static bool seccomp_actions_logged_from_names 1949 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
2339 { 1950 {
2340 char *name; 1951 char *name;
2341 1952
2342 *actions_logged = 0; 1953 *actions_logged = 0;
2343 while ((name = strsep(&names, " ")) & 1954 while ((name = strsep(&names, " ")) && *name) {
2344 u32 action_logged = 0; 1955 u32 action_logged = 0;
2345 1956
2346 if (!seccomp_action_logged_fr 1957 if (!seccomp_action_logged_from_name(&action_logged, name))
2347 return false; 1958 return false;
2348 1959
2349 *actions_logged |= action_log 1960 *actions_logged |= action_logged;
2350 } 1961 }
2351 1962
2352 return true; 1963 return true;
2353 } 1964 }
2354 1965
2355 static int read_actions_logged(const struct c !! 1966 static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer,
2356 size_t *lenp, 1967 size_t *lenp, loff_t *ppos)
2357 { 1968 {
2358 char names[sizeof(seccomp_actions_ava 1969 char names[sizeof(seccomp_actions_avail)];
2359 struct ctl_table table; 1970 struct ctl_table table;
2360 1971
2361 memset(names, 0, sizeof(names)); 1972 memset(names, 0, sizeof(names));
2362 1973
2363 if (!seccomp_names_from_actions_logge 1974 if (!seccomp_names_from_actions_logged(names, sizeof(names),
2364 1975 seccomp_actions_logged, " "))
2365 return -EINVAL; 1976 return -EINVAL;
2366 1977
2367 table = *ro_table; 1978 table = *ro_table;
2368 table.data = names; 1979 table.data = names;
2369 table.maxlen = sizeof(names); 1980 table.maxlen = sizeof(names);
2370 return proc_dostring(&table, 0, buffe 1981 return proc_dostring(&table, 0, buffer, lenp, ppos);
2371 } 1982 }
2372 1983
2373 static int write_actions_logged(const struct !! 1984 static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer,
2374 size_t *lenp, 1985 size_t *lenp, loff_t *ppos, u32 *actions_logged)
2375 { 1986 {
2376 char names[sizeof(seccomp_actions_ava 1987 char names[sizeof(seccomp_actions_avail)];
2377 struct ctl_table table; 1988 struct ctl_table table;
2378 int ret; 1989 int ret;
2379 1990
2380 if (!capable(CAP_SYS_ADMIN)) 1991 if (!capable(CAP_SYS_ADMIN))
2381 return -EPERM; 1992 return -EPERM;
2382 1993
2383 memset(names, 0, sizeof(names)); 1994 memset(names, 0, sizeof(names));
2384 1995
2385 table = *ro_table; 1996 table = *ro_table;
2386 table.data = names; 1997 table.data = names;
2387 table.maxlen = sizeof(names); 1998 table.maxlen = sizeof(names);
2388 ret = proc_dostring(&table, 1, buffer 1999 ret = proc_dostring(&table, 1, buffer, lenp, ppos);
2389 if (ret) 2000 if (ret)
2390 return ret; 2001 return ret;
2391 2002
2392 if (!seccomp_actions_logged_from_name 2003 if (!seccomp_actions_logged_from_names(actions_logged, table.data))
2393 return -EINVAL; 2004 return -EINVAL;
2394 2005
2395 if (*actions_logged & SECCOMP_LOG_ALL 2006 if (*actions_logged & SECCOMP_LOG_ALLOW)
2396 return -EINVAL; 2007 return -EINVAL;
2397 2008
2398 seccomp_actions_logged = *actions_log 2009 seccomp_actions_logged = *actions_logged;
2399 return 0; 2010 return 0;
2400 } 2011 }
2401 2012
2402 static void audit_actions_logged(u32 actions_ 2013 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
2403 int ret) 2014 int ret)
2404 { 2015 {
2405 char names[sizeof(seccomp_actions_ava 2016 char names[sizeof(seccomp_actions_avail)];
2406 char old_names[sizeof(seccomp_actions 2017 char old_names[sizeof(seccomp_actions_avail)];
2407 const char *new = names; 2018 const char *new = names;
2408 const char *old = old_names; 2019 const char *old = old_names;
2409 2020
2410 if (!audit_enabled) 2021 if (!audit_enabled)
2411 return; 2022 return;
2412 2023
2413 memset(names, 0, sizeof(names)); 2024 memset(names, 0, sizeof(names));
2414 memset(old_names, 0, sizeof(old_names 2025 memset(old_names, 0, sizeof(old_names));
2415 2026
2416 if (ret) 2027 if (ret)
2417 new = "?"; 2028 new = "?";
2418 else if (!actions_logged) 2029 else if (!actions_logged)
2419 new = "(none)"; 2030 new = "(none)";
2420 else if (!seccomp_names_from_actions_ 2031 else if (!seccomp_names_from_actions_logged(names, sizeof(names),
2421 2032 actions_logged, ","))
2422 new = "?"; 2033 new = "?";
2423 2034
2424 if (!old_actions_logged) 2035 if (!old_actions_logged)
2425 old = "(none)"; 2036 old = "(none)";
2426 else if (!seccomp_names_from_actions_ 2037 else if (!seccomp_names_from_actions_logged(old_names,
2427 2038 sizeof(old_names),
2428 2039 old_actions_logged, ","))
2429 old = "?"; 2040 old = "?";
2430 2041
2431 return audit_seccomp_actions_logged(n 2042 return audit_seccomp_actions_logged(new, old, !ret);
2432 } 2043 }
2433 2044
2434 static int seccomp_actions_logged_handler(con !! 2045 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
2435 voi 2046 void *buffer, size_t *lenp,
2436 lof 2047 loff_t *ppos)
2437 { 2048 {
2438 int ret; 2049 int ret;
2439 2050
2440 if (write) { 2051 if (write) {
2441 u32 actions_logged = 0; 2052 u32 actions_logged = 0;
2442 u32 old_actions_logged = secc 2053 u32 old_actions_logged = seccomp_actions_logged;
2443 2054
2444 ret = write_actions_logged(ro 2055 ret = write_actions_logged(ro_table, buffer, lenp, ppos,
2445 &a 2056 &actions_logged);
2446 audit_actions_logged(actions_ 2057 audit_actions_logged(actions_logged, old_actions_logged, ret);
2447 } else 2058 } else
2448 ret = read_actions_logged(ro_ 2059 ret = read_actions_logged(ro_table, buffer, lenp, ppos);
2449 2060
2450 return ret; 2061 return ret;
2451 } 2062 }
2452 2063
>> 2064 static struct ctl_path seccomp_sysctl_path[] = {
>> 2065 { .procname = "kernel", },
>> 2066 { .procname = "seccomp", },
>> 2067 { }
>> 2068 };
>> 2069
2453 static struct ctl_table seccomp_sysctl_table[ 2070 static struct ctl_table seccomp_sysctl_table[] = {
2454 { 2071 {
2455 .procname = "actions_av 2072 .procname = "actions_avail",
2456 .data = (void *) &s 2073 .data = (void *) &seccomp_actions_avail,
2457 .maxlen = sizeof(secc 2074 .maxlen = sizeof(seccomp_actions_avail),
2458 .mode = 0444, 2075 .mode = 0444,
2459 .proc_handler = proc_dostri 2076 .proc_handler = proc_dostring,
2460 }, 2077 },
2461 { 2078 {
2462 .procname = "actions_lo 2079 .procname = "actions_logged",
2463 .mode = 0644, 2080 .mode = 0644,
2464 .proc_handler = seccomp_act 2081 .proc_handler = seccomp_actions_logged_handler,
2465 }, 2082 },
>> 2083 { }
2466 }; 2084 };
2467 2085
2468 static int __init seccomp_sysctl_init(void) 2086 static int __init seccomp_sysctl_init(void)
2469 { 2087 {
2470 register_sysctl_init("kernel/seccomp" !! 2088 struct ctl_table_header *hdr;
>> 2089
>> 2090 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
>> 2091 if (!hdr)
>> 2092 pr_warn("sysctl registration failed\n");
>> 2093 else
>> 2094 kmemleak_not_leak(hdr);
>> 2095
2471 return 0; 2096 return 0;
2472 } 2097 }
2473 2098
2474 device_initcall(seccomp_sysctl_init) 2099 device_initcall(seccomp_sysctl_init)
2475 2100
2476 #endif /* CONFIG_SYSCTL */ 2101 #endif /* CONFIG_SYSCTL */
2477 <<
2478 #ifdef CONFIG_SECCOMP_CACHE_DEBUG <<
2479 /* Currently CONFIG_SECCOMP_CACHE_DEBUG impli <<
2480 static void proc_pid_seccomp_cache_arch(struc <<
2481 const <<
2482 { <<
2483 int nr; <<
2484 <<
2485 for (nr = 0; nr < bitmap_size; nr++) <<
2486 bool cached = test_bit(nr, bi <<
2487 char *status = cached ? "ALLO <<
2488 <<
2489 seq_printf(m, "%s %d %s\n", n <<
2490 } <<
2491 } <<
2492 <<
2493 int proc_pid_seccomp_cache(struct seq_file *m <<
2494 struct pid *pid, s <<
2495 { <<
2496 struct seccomp_filter *f; <<
2497 unsigned long flags; <<
2498 <<
2499 /* <<
2500 * We don't want some sandboxed proce <<
2501 * filters consist of. <<
2502 */ <<
2503 if (!file_ns_capable(m->file, &init_u <<
2504 return -EACCES; <<
2505 <<
2506 if (!lock_task_sighand(task, &flags)) <<
2507 return -ESRCH; <<
2508 <<
2509 f = READ_ONCE(task->seccomp.filter); <<
2510 if (!f) { <<
2511 unlock_task_sighand(task, &fl <<
2512 return 0; <<
2513 } <<
2514 <<
2515 /* prevent filter from being freed wh <<
2516 __get_seccomp_filter(f); <<
2517 unlock_task_sighand(task, &flags); <<
2518 <<
2519 proc_pid_seccomp_cache_arch(m, SECCOM <<
2520 f->cache. <<
2521 SECCOMP_A <<
2522 <<
2523 #ifdef SECCOMP_ARCH_COMPAT <<
2524 proc_pid_seccomp_cache_arch(m, SECCOM <<
2525 f->cache. <<
2526 SECCOMP_A <<
2527 #endif /* SECCOMP_ARCH_COMPAT */ <<
2528 <<
2529 __put_seccomp_filter(f); <<
2530 return 0; <<
2531 } <<
2532 #endif /* CONFIG_SECCOMP_CACHE_DEBUG */ <<
2533 2102
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.