1 // SPDX-License-Identifier: GPL-2.0+ 1 // SPDX-License-Identifier: GPL-2.0+
2 /* 2 /*
3 * Restartable sequences system call 3 * Restartable sequences system call
4 * 4 *
5 * Copyright (C) 2015, Google, Inc., 5 * Copyright (C) 2015, Google, Inc.,
6 * Paul Turner <pjt@google.com> and Andrew Hun 6 * Paul Turner <pjt@google.com> and Andrew Hunter <ahh@google.com>
7 * Copyright (C) 2015-2018, EfficiOS Inc., 7 * Copyright (C) 2015-2018, EfficiOS Inc.,
8 * Mathieu Desnoyers <mathieu.desnoyers@effici 8 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
9 */ 9 */
10 10
11 #include <linux/sched.h> 11 #include <linux/sched.h>
12 #include <linux/uaccess.h> 12 #include <linux/uaccess.h>
13 #include <linux/syscalls.h> 13 #include <linux/syscalls.h>
14 #include <linux/rseq.h> 14 #include <linux/rseq.h>
15 #include <linux/types.h> 15 #include <linux/types.h>
16 #include <asm/ptrace.h> 16 #include <asm/ptrace.h>
17 17
18 #define CREATE_TRACE_POINTS 18 #define CREATE_TRACE_POINTS
19 #include <trace/events/rseq.h> 19 #include <trace/events/rseq.h>
20 20
21 /* The original rseq structure size (including !! 21 #define RSEQ_CS_PREEMPT_MIGRATE_FLAGS (RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE | \
22 #define ORIG_RSEQ_SIZE 32 !! 22 RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT)
23 <<
24 #define RSEQ_CS_NO_RESTART_FLAGS (RSEQ_CS_FLAG <<
25 RSEQ_CS_FLAG <<
26 RSEQ_CS_FLAG <<
27 23
28 /* 24 /*
29 * 25 *
30 * Restartable sequences are a lightweight int 26 * Restartable sequences are a lightweight interface that allows
31 * user-level code to be executed atomically r 27 * user-level code to be executed atomically relative to scheduler
32 * preemption and signal delivery. Typically u 28 * preemption and signal delivery. Typically used for implementing
33 * per-cpu operations. 29 * per-cpu operations.
34 * 30 *
35 * It allows user-space to perform update oper 31 * It allows user-space to perform update operations on per-cpu data
36 * without requiring heavy-weight atomic opera 32 * without requiring heavy-weight atomic operations.
37 * 33 *
38 * Detailed algorithm of rseq user-space assem 34 * Detailed algorithm of rseq user-space assembly sequences:
39 * 35 *
40 * init(rseq_cs) 36 * init(rseq_cs)
41 * cpu = TLS->rseq::cpu_id 37 * cpu = TLS->rseq::cpu_id_start
42 * [1] TLS->rseq::rseq_cs = rs 38 * [1] TLS->rseq::rseq_cs = rseq_cs
43 * [start_ip] ----------------------- 39 * [start_ip] ----------------------------
44 * [2] if (cpu != TLS->rseq::c 40 * [2] if (cpu != TLS->rseq::cpu_id)
45 * goto abort_ip; 41 * goto abort_ip;
46 * [3] <last_instruction_in_cs 42 * [3] <last_instruction_in_cs>
47 * [post_commit_ip] ----------------------- 43 * [post_commit_ip] ----------------------------
48 * 44 *
49 * The address of jump target abort_ip must 45 * The address of jump target abort_ip must be outside the critical
50 * region, i.e.: 46 * region, i.e.:
51 * 47 *
52 * [abort_ip] < [start_ip] || [abort_ip] 48 * [abort_ip] < [start_ip] || [abort_ip] >= [post_commit_ip]
53 * 49 *
54 * Steps [2]-[3] (inclusive) need to be a se 50 * Steps [2]-[3] (inclusive) need to be a sequence of instructions in
55 * userspace that can handle being interrupt 51 * userspace that can handle being interrupted between any of those
56 * instructions, and then resumed to the abo 52 * instructions, and then resumed to the abort_ip.
57 * 53 *
58 * 1. Userspace stores the address of the s 54 * 1. Userspace stores the address of the struct rseq_cs assembly
59 * block descriptor into the rseq_cs fie 55 * block descriptor into the rseq_cs field of the registered
60 * struct rseq TLS area. This update is 56 * struct rseq TLS area. This update is performed through a single
61 * store within the inline assembly inst 57 * store within the inline assembly instruction sequence.
62 * [start_ip] 58 * [start_ip]
63 * 59 *
64 * 2. Userspace tests to check whether the 60 * 2. Userspace tests to check whether the current cpu_id field match
65 * the cpu number loaded before start_ip 61 * the cpu number loaded before start_ip, branching to abort_ip
66 * in case of a mismatch. 62 * in case of a mismatch.
67 * 63 *
68 * If the sequence is preempted or inter 64 * If the sequence is preempted or interrupted by a signal
69 * at or after start_ip and before post_ 65 * at or after start_ip and before post_commit_ip, then the kernel
70 * clears TLS->__rseq_abi::rseq_cs, and 66 * clears TLS->__rseq_abi::rseq_cs, and sets the user-space return
71 * ip to abort_ip before returning to us 67 * ip to abort_ip before returning to user-space, so the preempted
72 * execution resumes at abort_ip. 68 * execution resumes at abort_ip.
73 * 69 *
74 * 3. Userspace critical section final inst 70 * 3. Userspace critical section final instruction before
75 * post_commit_ip is the commit. The cri 71 * post_commit_ip is the commit. The critical section is
76 * self-terminating. 72 * self-terminating.
77 * [post_commit_ip] 73 * [post_commit_ip]
78 * 74 *
79 * 4. <success> 75 * 4. <success>
80 * 76 *
81 * On failure at [2], or if interrupted by p 77 * On failure at [2], or if interrupted by preempt or signal delivery
82 * between [1] and [3]: 78 * between [1] and [3]:
83 * 79 *
84 * [abort_ip] 80 * [abort_ip]
85 * F1. <failure> 81 * F1. <failure>
86 */ 82 */
87 83
88 static int rseq_update_cpu_node_id(struct task !! 84 static int rseq_update_cpu_id(struct task_struct *t)
89 { 85 {
90 struct rseq __user *rseq = t->rseq; <<
91 u32 cpu_id = raw_smp_processor_id(); 86 u32 cpu_id = raw_smp_processor_id();
92 u32 node_id = cpu_to_node(cpu_id); !! 87 struct rseq __user *rseq = t->rseq;
93 u32 mm_cid = task_mm_cid(t); <<
94 88
95 WARN_ON_ONCE((int) mm_cid < 0); !! 89 if (!user_write_access_begin(rseq, sizeof(*rseq)))
96 if (!user_write_access_begin(rseq, t-> <<
97 goto efault; 90 goto efault;
98 unsafe_put_user(cpu_id, &rseq->cpu_id_ 91 unsafe_put_user(cpu_id, &rseq->cpu_id_start, efault_end);
99 unsafe_put_user(cpu_id, &rseq->cpu_id, 92 unsafe_put_user(cpu_id, &rseq->cpu_id, efault_end);
100 unsafe_put_user(node_id, &rseq->node_i <<
101 unsafe_put_user(mm_cid, &rseq->mm_cid, <<
102 /* <<
103 * Additional feature fields added aft <<
104 * need to be conditionally updated on <<
105 * t->rseq_len != ORIG_RSEQ_SIZE. <<
106 */ <<
107 user_write_access_end(); 93 user_write_access_end();
108 trace_rseq_update(t); 94 trace_rseq_update(t);
109 return 0; 95 return 0;
110 96
111 efault_end: 97 efault_end:
112 user_write_access_end(); 98 user_write_access_end();
113 efault: 99 efault:
114 return -EFAULT; 100 return -EFAULT;
115 } 101 }
116 102
117 static int rseq_reset_rseq_cpu_node_id(struct !! 103 static int rseq_reset_rseq_cpu_id(struct task_struct *t)
118 { 104 {
119 u32 cpu_id_start = 0, cpu_id = RSEQ_CP !! 105 u32 cpu_id_start = 0, cpu_id = RSEQ_CPU_ID_UNINITIALIZED;
120 mm_cid = 0; <<
121 106
122 /* 107 /*
123 * Reset cpu_id_start to its initial s 108 * Reset cpu_id_start to its initial state (0).
124 */ 109 */
125 if (put_user(cpu_id_start, &t->rseq->c 110 if (put_user(cpu_id_start, &t->rseq->cpu_id_start))
126 return -EFAULT; 111 return -EFAULT;
127 /* 112 /*
128 * Reset cpu_id to RSEQ_CPU_ID_UNINITI 113 * Reset cpu_id to RSEQ_CPU_ID_UNINITIALIZED, so any user coming
129 * in after unregistration can figure 114 * in after unregistration can figure out that rseq needs to be
130 * registered again. 115 * registered again.
131 */ 116 */
132 if (put_user(cpu_id, &t->rseq->cpu_id) 117 if (put_user(cpu_id, &t->rseq->cpu_id))
133 return -EFAULT; 118 return -EFAULT;
134 /* <<
135 * Reset node_id to its initial state <<
136 */ <<
137 if (put_user(node_id, &t->rseq->node_i <<
138 return -EFAULT; <<
139 /* <<
140 * Reset mm_cid to its initial state ( <<
141 */ <<
142 if (put_user(mm_cid, &t->rseq->mm_cid) <<
143 return -EFAULT; <<
144 /* <<
145 * Additional feature fields added aft <<
146 * need to be conditionally reset only <<
147 * t->rseq_len != ORIG_RSEQ_SIZE. <<
148 */ <<
149 return 0; 119 return 0;
150 } 120 }
151 121
152 static int rseq_get_rseq_cs(struct task_struct 122 static int rseq_get_rseq_cs(struct task_struct *t, struct rseq_cs *rseq_cs)
153 { 123 {
154 struct rseq_cs __user *urseq_cs; 124 struct rseq_cs __user *urseq_cs;
155 u64 ptr; 125 u64 ptr;
156 u32 __user *usig; 126 u32 __user *usig;
157 u32 sig; 127 u32 sig;
158 int ret; 128 int ret;
159 129
160 #ifdef CONFIG_64BIT 130 #ifdef CONFIG_64BIT
161 if (get_user(ptr, &t->rseq->rseq_cs)) 131 if (get_user(ptr, &t->rseq->rseq_cs))
162 return -EFAULT; 132 return -EFAULT;
163 #else 133 #else
164 if (copy_from_user(&ptr, &t->rseq->rse 134 if (copy_from_user(&ptr, &t->rseq->rseq_cs, sizeof(ptr)))
165 return -EFAULT; 135 return -EFAULT;
166 #endif 136 #endif
167 if (!ptr) { 137 if (!ptr) {
168 memset(rseq_cs, 0, sizeof(*rse 138 memset(rseq_cs, 0, sizeof(*rseq_cs));
169 return 0; 139 return 0;
170 } 140 }
171 if (ptr >= TASK_SIZE) 141 if (ptr >= TASK_SIZE)
172 return -EINVAL; 142 return -EINVAL;
173 urseq_cs = (struct rseq_cs __user *)(u 143 urseq_cs = (struct rseq_cs __user *)(unsigned long)ptr;
174 if (copy_from_user(rseq_cs, urseq_cs, 144 if (copy_from_user(rseq_cs, urseq_cs, sizeof(*rseq_cs)))
175 return -EFAULT; 145 return -EFAULT;
176 146
177 if (rseq_cs->start_ip >= TASK_SIZE || 147 if (rseq_cs->start_ip >= TASK_SIZE ||
178 rseq_cs->start_ip + rseq_cs->post_ 148 rseq_cs->start_ip + rseq_cs->post_commit_offset >= TASK_SIZE ||
179 rseq_cs->abort_ip >= TASK_SIZE || 149 rseq_cs->abort_ip >= TASK_SIZE ||
180 rseq_cs->version > 0) 150 rseq_cs->version > 0)
181 return -EINVAL; 151 return -EINVAL;
182 /* Check for overflow. */ 152 /* Check for overflow. */
183 if (rseq_cs->start_ip + rseq_cs->post_ 153 if (rseq_cs->start_ip + rseq_cs->post_commit_offset < rseq_cs->start_ip)
184 return -EINVAL; 154 return -EINVAL;
185 /* Ensure that abort_ip is not in the 155 /* Ensure that abort_ip is not in the critical section. */
186 if (rseq_cs->abort_ip - rseq_cs->start 156 if (rseq_cs->abort_ip - rseq_cs->start_ip < rseq_cs->post_commit_offset)
187 return -EINVAL; 157 return -EINVAL;
188 158
189 usig = (u32 __user *)(unsigned long)(r 159 usig = (u32 __user *)(unsigned long)(rseq_cs->abort_ip - sizeof(u32));
190 ret = get_user(sig, usig); 160 ret = get_user(sig, usig);
191 if (ret) 161 if (ret)
192 return ret; 162 return ret;
193 163
194 if (current->rseq_sig != sig) { 164 if (current->rseq_sig != sig) {
195 printk_ratelimited(KERN_WARNIN 165 printk_ratelimited(KERN_WARNING
196 "Possible attack attem 166 "Possible attack attempt. Unexpected rseq signature 0x%x, expecting 0x%x (pid=%d, addr=%p).\n",
197 sig, current->rseq_sig 167 sig, current->rseq_sig, current->pid, usig);
198 return -EINVAL; 168 return -EINVAL;
199 } 169 }
200 return 0; 170 return 0;
201 } 171 }
202 172
203 static bool rseq_warn_flags(const char *str, u <<
204 { <<
205 u32 test_flags; <<
206 <<
207 if (!flags) <<
208 return false; <<
209 test_flags = flags & RSEQ_CS_NO_RESTAR <<
210 if (test_flags) <<
211 pr_warn_once("Deprecated flags <<
212 test_flags = flags & ~RSEQ_CS_NO_RESTA <<
213 if (test_flags) <<
214 pr_warn_once("Unknown flags (% <<
215 return true; <<
216 } <<
217 <<
218 static int rseq_need_restart(struct task_struc 173 static int rseq_need_restart(struct task_struct *t, u32 cs_flags)
219 { 174 {
220 u32 flags, event_mask; 175 u32 flags, event_mask;
221 int ret; 176 int ret;
222 177
223 if (rseq_warn_flags("rseq_cs", cs_flag <<
224 return -EINVAL; <<
225 <<
226 /* Get thread flags. */ 178 /* Get thread flags. */
227 ret = get_user(flags, &t->rseq->flags) 179 ret = get_user(flags, &t->rseq->flags);
228 if (ret) 180 if (ret)
229 return ret; 181 return ret;
230 182
231 if (rseq_warn_flags("rseq", flags)) !! 183 /* Take critical section flags into account. */
>> 184 flags |= cs_flags;
>> 185
>> 186 /*
>> 187 * Restart on signal can only be inhibited when restart on
>> 188 * preempt and restart on migrate are inhibited too. Otherwise,
>> 189 * a preempted signal handler could fail to restart the prior
>> 190 * execution context on sigreturn.
>> 191 */
>> 192 if (unlikely((flags & RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL) &&
>> 193 (flags & RSEQ_CS_PREEMPT_MIGRATE_FLAGS) !=
>> 194 RSEQ_CS_PREEMPT_MIGRATE_FLAGS))
232 return -EINVAL; 195 return -EINVAL;
233 196
234 /* 197 /*
235 * Load and clear event mask atomicall 198 * Load and clear event mask atomically with respect to
236 * scheduler preemption. 199 * scheduler preemption.
237 */ 200 */
238 preempt_disable(); 201 preempt_disable();
239 event_mask = t->rseq_event_mask; 202 event_mask = t->rseq_event_mask;
240 t->rseq_event_mask = 0; 203 t->rseq_event_mask = 0;
241 preempt_enable(); 204 preempt_enable();
242 205
243 return !!event_mask; !! 206 return !!(event_mask & ~flags);
244 } 207 }
245 208
246 static int clear_rseq_cs(struct task_struct *t 209 static int clear_rseq_cs(struct task_struct *t)
247 { 210 {
248 /* 211 /*
249 * The rseq_cs field is set to NULL on 212 * The rseq_cs field is set to NULL on preemption or signal
250 * delivery on top of rseq assembly bl 213 * delivery on top of rseq assembly block, as well as on top
251 * of code outside of the rseq assembl 214 * of code outside of the rseq assembly block. This performs
252 * a lazy clear of the rseq_cs field. 215 * a lazy clear of the rseq_cs field.
253 * 216 *
254 * Set rseq_cs to NULL. 217 * Set rseq_cs to NULL.
255 */ 218 */
256 #ifdef CONFIG_64BIT 219 #ifdef CONFIG_64BIT
257 return put_user(0UL, &t->rseq->rseq_cs 220 return put_user(0UL, &t->rseq->rseq_cs);
258 #else 221 #else
259 if (clear_user(&t->rseq->rseq_cs, size 222 if (clear_user(&t->rseq->rseq_cs, sizeof(t->rseq->rseq_cs)))
260 return -EFAULT; 223 return -EFAULT;
261 return 0; 224 return 0;
262 #endif 225 #endif
263 } 226 }
264 227
265 /* 228 /*
266 * Unsigned comparison will be true when ip >= 229 * Unsigned comparison will be true when ip >= start_ip, and when
267 * ip < start_ip + post_commit_offset. 230 * ip < start_ip + post_commit_offset.
268 */ 231 */
269 static bool in_rseq_cs(unsigned long ip, struc 232 static bool in_rseq_cs(unsigned long ip, struct rseq_cs *rseq_cs)
270 { 233 {
271 return ip - rseq_cs->start_ip < rseq_c 234 return ip - rseq_cs->start_ip < rseq_cs->post_commit_offset;
272 } 235 }
273 236
274 static int rseq_ip_fixup(struct pt_regs *regs) 237 static int rseq_ip_fixup(struct pt_regs *regs)
275 { 238 {
276 unsigned long ip = instruction_pointer 239 unsigned long ip = instruction_pointer(regs);
277 struct task_struct *t = current; 240 struct task_struct *t = current;
278 struct rseq_cs rseq_cs; 241 struct rseq_cs rseq_cs;
279 int ret; 242 int ret;
280 243
281 ret = rseq_get_rseq_cs(t, &rseq_cs); 244 ret = rseq_get_rseq_cs(t, &rseq_cs);
282 if (ret) 245 if (ret)
283 return ret; 246 return ret;
284 247
285 /* 248 /*
286 * Handle potentially not being within 249 * Handle potentially not being within a critical section.
287 * If not nested over a rseq critical 250 * If not nested over a rseq critical section, restart is useless.
288 * Clear the rseq_cs pointer and retur 251 * Clear the rseq_cs pointer and return.
289 */ 252 */
290 if (!in_rseq_cs(ip, &rseq_cs)) 253 if (!in_rseq_cs(ip, &rseq_cs))
291 return clear_rseq_cs(t); 254 return clear_rseq_cs(t);
292 ret = rseq_need_restart(t, rseq_cs.fla 255 ret = rseq_need_restart(t, rseq_cs.flags);
293 if (ret <= 0) 256 if (ret <= 0)
294 return ret; 257 return ret;
295 ret = clear_rseq_cs(t); 258 ret = clear_rseq_cs(t);
296 if (ret) 259 if (ret)
297 return ret; 260 return ret;
298 trace_rseq_ip_fixup(ip, rseq_cs.start_ 261 trace_rseq_ip_fixup(ip, rseq_cs.start_ip, rseq_cs.post_commit_offset,
299 rseq_cs.abort_ip); 262 rseq_cs.abort_ip);
300 instruction_pointer_set(regs, (unsigne 263 instruction_pointer_set(regs, (unsigned long)rseq_cs.abort_ip);
301 return 0; 264 return 0;
302 } 265 }
303 266
304 /* 267 /*
305 * This resume handler must always be executed 268 * This resume handler must always be executed between any of:
306 * - preemption, 269 * - preemption,
307 * - signal delivery, 270 * - signal delivery,
308 * and return to user-space. 271 * and return to user-space.
309 * 272 *
310 * This is how we can ensure that the entire r 273 * This is how we can ensure that the entire rseq critical section
311 * will issue the commit instruction only if e 274 * will issue the commit instruction only if executed atomically with
312 * respect to other threads scheduled on the s 275 * respect to other threads scheduled on the same CPU, and with respect
313 * to signal handlers. 276 * to signal handlers.
314 */ 277 */
315 void __rseq_handle_notify_resume(struct ksigna 278 void __rseq_handle_notify_resume(struct ksignal *ksig, struct pt_regs *regs)
316 { 279 {
317 struct task_struct *t = current; 280 struct task_struct *t = current;
318 int ret, sig; 281 int ret, sig;
319 282
320 if (unlikely(t->flags & PF_EXITING)) 283 if (unlikely(t->flags & PF_EXITING))
321 return; 284 return;
322 285
323 /* 286 /*
324 * regs is NULL if and only if the cal 287 * regs is NULL if and only if the caller is in a syscall path. Skip
325 * fixup and leave rseq_cs as is so th 288 * fixup and leave rseq_cs as is so that rseq_sycall() will detect and
326 * kill a misbehaving userspace on deb 289 * kill a misbehaving userspace on debug kernels.
327 */ 290 */
328 if (regs) { 291 if (regs) {
329 ret = rseq_ip_fixup(regs); 292 ret = rseq_ip_fixup(regs);
330 if (unlikely(ret < 0)) 293 if (unlikely(ret < 0))
331 goto error; 294 goto error;
332 } 295 }
333 if (unlikely(rseq_update_cpu_node_id(t !! 296 if (unlikely(rseq_update_cpu_id(t)))
334 goto error; 297 goto error;
335 return; 298 return;
336 299
337 error: 300 error:
338 sig = ksig ? ksig->sig : 0; 301 sig = ksig ? ksig->sig : 0;
339 force_sigsegv(sig); 302 force_sigsegv(sig);
340 } 303 }
341 304
342 #ifdef CONFIG_DEBUG_RSEQ 305 #ifdef CONFIG_DEBUG_RSEQ
343 306
344 /* 307 /*
345 * Terminate the process if a syscall is issue 308 * Terminate the process if a syscall is issued within a restartable
346 * sequence. 309 * sequence.
347 */ 310 */
348 void rseq_syscall(struct pt_regs *regs) 311 void rseq_syscall(struct pt_regs *regs)
349 { 312 {
350 unsigned long ip = instruction_pointer 313 unsigned long ip = instruction_pointer(regs);
351 struct task_struct *t = current; 314 struct task_struct *t = current;
352 struct rseq_cs rseq_cs; 315 struct rseq_cs rseq_cs;
353 316
354 if (!t->rseq) 317 if (!t->rseq)
355 return; 318 return;
356 if (rseq_get_rseq_cs(t, &rseq_cs) || i 319 if (rseq_get_rseq_cs(t, &rseq_cs) || in_rseq_cs(ip, &rseq_cs))
357 force_sig(SIGSEGV); 320 force_sig(SIGSEGV);
358 } 321 }
359 322
360 #endif 323 #endif
361 324
362 /* 325 /*
363 * sys_rseq - setup restartable sequences for 326 * sys_rseq - setup restartable sequences for caller thread.
364 */ 327 */
365 SYSCALL_DEFINE4(rseq, struct rseq __user *, rs 328 SYSCALL_DEFINE4(rseq, struct rseq __user *, rseq, u32, rseq_len,
366 int, flags, u32, sig) 329 int, flags, u32, sig)
367 { 330 {
368 int ret; 331 int ret;
369 332
370 if (flags & RSEQ_FLAG_UNREGISTER) { 333 if (flags & RSEQ_FLAG_UNREGISTER) {
371 if (flags & ~RSEQ_FLAG_UNREGIS 334 if (flags & ~RSEQ_FLAG_UNREGISTER)
372 return -EINVAL; 335 return -EINVAL;
373 /* Unregister rseq for current 336 /* Unregister rseq for current thread. */
374 if (current->rseq != rseq || ! 337 if (current->rseq != rseq || !current->rseq)
375 return -EINVAL; 338 return -EINVAL;
376 if (rseq_len != current->rseq_ !! 339 if (rseq_len != sizeof(*rseq))
377 return -EINVAL; 340 return -EINVAL;
378 if (current->rseq_sig != sig) 341 if (current->rseq_sig != sig)
379 return -EPERM; 342 return -EPERM;
380 ret = rseq_reset_rseq_cpu_node !! 343 ret = rseq_reset_rseq_cpu_id(current);
381 if (ret) 344 if (ret)
382 return ret; 345 return ret;
383 current->rseq = NULL; 346 current->rseq = NULL;
384 current->rseq_sig = 0; 347 current->rseq_sig = 0;
385 current->rseq_len = 0; <<
386 return 0; 348 return 0;
387 } 349 }
388 350
389 if (unlikely(flags)) 351 if (unlikely(flags))
390 return -EINVAL; 352 return -EINVAL;
391 353
392 if (current->rseq) { 354 if (current->rseq) {
393 /* 355 /*
394 * If rseq is already register 356 * If rseq is already registered, check whether
395 * the provided address differ 357 * the provided address differs from the prior
396 * one. 358 * one.
397 */ 359 */
398 if (current->rseq != rseq || r !! 360 if (current->rseq != rseq || rseq_len != sizeof(*rseq))
399 return -EINVAL; 361 return -EINVAL;
400 if (current->rseq_sig != sig) 362 if (current->rseq_sig != sig)
401 return -EPERM; 363 return -EPERM;
402 /* Already registered. */ 364 /* Already registered. */
403 return -EBUSY; 365 return -EBUSY;
404 } 366 }
405 367
406 /* 368 /*
407 * If there was no rseq previously reg !! 369 * If there was no rseq previously registered,
408 * is properly aligned, as communcated !! 370 * ensure the provided rseq is properly aligned and valid.
409 * auxiliary vector AT_RSEQ_ALIGN. If <<
410 * size, the required alignment is the <<
411 * <<
412 * In order to be valid, rseq_len is e <<
413 * large enough to contain all support <<
414 * user-space through the ELF auxiliar <<
415 */ 371 */
416 if (rseq_len < ORIG_RSEQ_SIZE || !! 372 if (!IS_ALIGNED((unsigned long)rseq, __alignof__(*rseq)) ||
417 (rseq_len == ORIG_RSEQ_SIZE && !IS !! 373 rseq_len != sizeof(*rseq))
418 (rseq_len != ORIG_RSEQ_SIZE && (!I <<
419 rs <<
420 return -EINVAL; 374 return -EINVAL;
421 if (!access_ok(rseq, rseq_len)) 375 if (!access_ok(rseq, rseq_len))
422 return -EFAULT; 376 return -EFAULT;
423 current->rseq = rseq; 377 current->rseq = rseq;
424 current->rseq_len = rseq_len; <<
425 current->rseq_sig = sig; 378 current->rseq_sig = sig;
426 /* 379 /*
427 * If rseq was previously inactive, an 380 * If rseq was previously inactive, and has just been
428 * registered, ensure the cpu_id_start 381 * registered, ensure the cpu_id_start and cpu_id fields
429 * are updated before returning to use 382 * are updated before returning to user-space.
430 */ 383 */
431 rseq_set_notify_resume(current); 384 rseq_set_notify_resume(current);
432 385
433 return 0; 386 return 0;
434 } 387 }
435 388
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.