Commit | Line | Data |
---|---|---|
1f484aa6 AL |
1 | /* |
2 | * common.c - C code for kernel entry and exit | |
3 | * Copyright (c) 2015 Andrew Lutomirski | |
4 | * GPL v2 | |
5 | * | |
6 | * Based on asm and ptrace code by many authors. The code here originated | |
7 | * in ptrace.c and signal.c. | |
8 | */ | |
9 | ||
10 | #include <linux/kernel.h> | |
11 | #include <linux/sched.h> | |
12 | #include <linux/mm.h> | |
13 | #include <linux/smp.h> | |
14 | #include <linux/errno.h> | |
15 | #include <linux/ptrace.h> | |
16 | #include <linux/tracehook.h> | |
17 | #include <linux/audit.h> | |
18 | #include <linux/seccomp.h> | |
19 | #include <linux/signal.h> | |
20 | #include <linux/export.h> | |
21 | #include <linux/context_tracking.h> | |
22 | #include <linux/user-return-notifier.h> | |
23 | #include <linux/uprobes.h> | |
24 | ||
25 | #include <asm/desc.h> | |
26 | #include <asm/traps.h> | |
710246df AL |
27 | #include <asm/vdso.h> |
28 | #include <asm/uaccess.h> | |
1f484aa6 AL |
29 | |
30 | #define CREATE_TRACE_POINTS | |
31 | #include <trace/events/syscalls.h> | |
32 | ||
dd636071 AL |
33 | static struct thread_info *pt_regs_to_thread_info(struct pt_regs *regs) |
34 | { | |
35 | unsigned long top_of_stack = | |
36 | (unsigned long)(regs + 1) + TOP_OF_KERNEL_STACK_PADDING; | |
37 | return (struct thread_info *)(top_of_stack - THREAD_SIZE); | |
38 | } | |
39 | ||
feed36cd AL |
40 | #ifdef CONFIG_CONTEXT_TRACKING |
41 | /* Called on entry from user mode with IRQs off. */ | |
42 | __visible void enter_from_user_mode(void) | |
43 | { | |
44 | CT_WARN_ON(ct_state() != CONTEXT_USER); | |
45 | user_exit(); | |
46 | } | |
47 | #endif | |
48 | ||
1f484aa6 AL |
49 | static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch) |
50 | { | |
51 | #ifdef CONFIG_X86_64 | |
52 | if (arch == AUDIT_ARCH_X86_64) { | |
53 | audit_syscall_entry(regs->orig_ax, regs->di, | |
54 | regs->si, regs->dx, regs->r10); | |
55 | } else | |
56 | #endif | |
57 | { | |
58 | audit_syscall_entry(regs->orig_ax, regs->bx, | |
59 | regs->cx, regs->dx, regs->si); | |
60 | } | |
61 | } | |
62 | ||
63 | /* | |
64 | * We can return 0 to resume the syscall or anything else to go to phase | |
65 | * 2. If we resume the syscall, we need to put something appropriate in | |
66 | * regs->orig_ax. | |
67 | * | |
68 | * NB: We don't have full pt_regs here, but regs->orig_ax and regs->ax | |
69 | * are fully functional. | |
70 | * | |
71 | * For phase 2's benefit, our return value is: | |
72 | * 0: resume the syscall | |
73 | * 1: go to phase 2; no seccomp phase 2 needed | |
74 | * anything else: go to phase 2; pass return value to seccomp | |
75 | */ | |
76 | unsigned long syscall_trace_enter_phase1(struct pt_regs *regs, u32 arch) | |
77 | { | |
dd636071 | 78 | struct thread_info *ti = pt_regs_to_thread_info(regs); |
1f484aa6 AL |
79 | unsigned long ret = 0; |
80 | u32 work; | |
81 | ||
4aabd140 AL |
82 | if (IS_ENABLED(CONFIG_DEBUG_ENTRY)) |
83 | BUG_ON(regs != task_pt_regs(current)); | |
1f484aa6 | 84 | |
dd636071 | 85 | work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY; |
1f484aa6 | 86 | |
feed36cd | 87 | #ifdef CONFIG_CONTEXT_TRACKING |
1f484aa6 AL |
88 | /* |
89 | * If TIF_NOHZ is set, we are required to call user_exit() before | |
90 | * doing anything that could touch RCU. | |
91 | */ | |
92 | if (work & _TIF_NOHZ) { | |
feed36cd | 93 | enter_from_user_mode(); |
1f484aa6 AL |
94 | work &= ~_TIF_NOHZ; |
95 | } | |
feed36cd | 96 | #endif |
1f484aa6 AL |
97 | |
98 | #ifdef CONFIG_SECCOMP | |
99 | /* | |
100 | * Do seccomp first -- it should minimize exposure of other | |
101 | * code, and keeping seccomp fast is probably more valuable | |
102 | * than the rest of this. | |
103 | */ | |
104 | if (work & _TIF_SECCOMP) { | |
105 | struct seccomp_data sd; | |
106 | ||
107 | sd.arch = arch; | |
108 | sd.nr = regs->orig_ax; | |
109 | sd.instruction_pointer = regs->ip; | |
110 | #ifdef CONFIG_X86_64 | |
111 | if (arch == AUDIT_ARCH_X86_64) { | |
112 | sd.args[0] = regs->di; | |
113 | sd.args[1] = regs->si; | |
114 | sd.args[2] = regs->dx; | |
115 | sd.args[3] = regs->r10; | |
116 | sd.args[4] = regs->r8; | |
117 | sd.args[5] = regs->r9; | |
118 | } else | |
119 | #endif | |
120 | { | |
121 | sd.args[0] = regs->bx; | |
122 | sd.args[1] = regs->cx; | |
123 | sd.args[2] = regs->dx; | |
124 | sd.args[3] = regs->si; | |
125 | sd.args[4] = regs->di; | |
126 | sd.args[5] = regs->bp; | |
127 | } | |
128 | ||
129 | BUILD_BUG_ON(SECCOMP_PHASE1_OK != 0); | |
130 | BUILD_BUG_ON(SECCOMP_PHASE1_SKIP != 1); | |
131 | ||
132 | ret = seccomp_phase1(&sd); | |
133 | if (ret == SECCOMP_PHASE1_SKIP) { | |
134 | regs->orig_ax = -1; | |
135 | ret = 0; | |
136 | } else if (ret != SECCOMP_PHASE1_OK) { | |
137 | return ret; /* Go directly to phase 2 */ | |
138 | } | |
139 | ||
140 | work &= ~_TIF_SECCOMP; | |
141 | } | |
142 | #endif | |
143 | ||
144 | /* Do our best to finish without phase 2. */ | |
145 | if (work == 0) | |
146 | return ret; /* seccomp and/or nohz only (ret == 0 here) */ | |
147 | ||
148 | #ifdef CONFIG_AUDITSYSCALL | |
149 | if (work == _TIF_SYSCALL_AUDIT) { | |
150 | /* | |
151 | * If there is no more work to be done except auditing, | |
152 | * then audit in phase 1. Phase 2 always audits, so, if | |
153 | * we audit here, then we can't go on to phase 2. | |
154 | */ | |
155 | do_audit_syscall_entry(regs, arch); | |
156 | return 0; | |
157 | } | |
158 | #endif | |
159 | ||
160 | return 1; /* Something is enabled that we can't handle in phase 1 */ | |
161 | } | |
162 | ||
163 | /* Returns the syscall nr to run (which should match regs->orig_ax). */ | |
164 | long syscall_trace_enter_phase2(struct pt_regs *regs, u32 arch, | |
165 | unsigned long phase1_result) | |
166 | { | |
dd636071 | 167 | struct thread_info *ti = pt_regs_to_thread_info(regs); |
1f484aa6 | 168 | long ret = 0; |
dd636071 | 169 | u32 work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY; |
1f484aa6 | 170 | |
4aabd140 AL |
171 | if (IS_ENABLED(CONFIG_DEBUG_ENTRY)) |
172 | BUG_ON(regs != task_pt_regs(current)); | |
1f484aa6 AL |
173 | |
174 | /* | |
175 | * If we stepped into a sysenter/syscall insn, it trapped in | |
176 | * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP. | |
177 | * If user-mode had set TF itself, then it's still clear from | |
178 | * do_debug() and we need to set it again to restore the user | |
179 | * state. If we entered on the slow path, TF was already set. | |
180 | */ | |
181 | if (work & _TIF_SINGLESTEP) | |
182 | regs->flags |= X86_EFLAGS_TF; | |
183 | ||
184 | #ifdef CONFIG_SECCOMP | |
185 | /* | |
186 | * Call seccomp_phase2 before running the other hooks so that | |
187 | * they can see any changes made by a seccomp tracer. | |
188 | */ | |
189 | if (phase1_result > 1 && seccomp_phase2(phase1_result)) { | |
190 | /* seccomp failures shouldn't expose any additional code. */ | |
191 | return -1; | |
192 | } | |
193 | #endif | |
194 | ||
195 | if (unlikely(work & _TIF_SYSCALL_EMU)) | |
196 | ret = -1L; | |
197 | ||
198 | if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) && | |
199 | tracehook_report_syscall_entry(regs)) | |
200 | ret = -1L; | |
201 | ||
202 | if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT))) | |
203 | trace_sys_enter(regs, regs->orig_ax); | |
204 | ||
205 | do_audit_syscall_entry(regs, arch); | |
206 | ||
207 | return ret ?: regs->orig_ax; | |
208 | } | |
209 | ||
210 | long syscall_trace_enter(struct pt_regs *regs) | |
211 | { | |
212 | u32 arch = is_ia32_task() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64; | |
213 | unsigned long phase1_result = syscall_trace_enter_phase1(regs, arch); | |
214 | ||
215 | if (phase1_result == 0) | |
216 | return regs->orig_ax; | |
217 | else | |
218 | return syscall_trace_enter_phase2(regs, arch, phase1_result); | |
219 | } | |
220 | ||
c5c46f59 AL |
221 | /* Called with IRQs disabled. */ |
222 | __visible void prepare_exit_to_usermode(struct pt_regs *regs) | |
223 | { | |
460d1245 | 224 | if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled())) |
c5c46f59 AL |
225 | local_irq_disable(); |
226 | ||
72f92478 AL |
227 | lockdep_sys_exit(); |
228 | ||
c5c46f59 AL |
229 | /* |
230 | * In order to return to user mode, we need to have IRQs off with | |
231 | * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY, | |
232 | * _TIF_UPROBE, or _TIF_NEED_RESCHED set. Several of these flags | |
233 | * can be set at any time on preemptable kernels if we have IRQs on, | |
234 | * so we need to loop. Disabling preemption wouldn't help: doing the | |
235 | * work to clear some of the flags can sleep. | |
236 | */ | |
237 | while (true) { | |
238 | u32 cached_flags = | |
239 | READ_ONCE(pt_regs_to_thread_info(regs)->flags); | |
240 | ||
241 | if (!(cached_flags & (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | | |
d132803e AL |
242 | _TIF_UPROBE | _TIF_NEED_RESCHED | |
243 | _TIF_USER_RETURN_NOTIFY))) | |
c5c46f59 AL |
244 | break; |
245 | ||
246 | /* We have work to do. */ | |
247 | local_irq_enable(); | |
248 | ||
249 | if (cached_flags & _TIF_NEED_RESCHED) | |
250 | schedule(); | |
251 | ||
252 | if (cached_flags & _TIF_UPROBE) | |
253 | uprobe_notify_resume(regs); | |
254 | ||
255 | /* deal with pending signal delivery */ | |
256 | if (cached_flags & _TIF_SIGPENDING) | |
257 | do_signal(regs); | |
258 | ||
259 | if (cached_flags & _TIF_NOTIFY_RESUME) { | |
260 | clear_thread_flag(TIF_NOTIFY_RESUME); | |
261 | tracehook_notify_resume(regs); | |
262 | } | |
263 | ||
264 | if (cached_flags & _TIF_USER_RETURN_NOTIFY) | |
265 | fire_user_return_notifiers(); | |
266 | ||
267 | /* Disable IRQs and retry */ | |
268 | local_irq_disable(); | |
269 | } | |
270 | ||
271 | user_enter(); | |
272 | } | |
273 | ||
274 | /* | |
275 | * Called with IRQs on and fully valid regs. Returns with IRQs off in a | |
276 | * state such that we can immediately switch to user mode. | |
277 | */ | |
278 | __visible void syscall_return_slowpath(struct pt_regs *regs) | |
279 | { | |
280 | struct thread_info *ti = pt_regs_to_thread_info(regs); | |
281 | u32 cached_flags = READ_ONCE(ti->flags); | |
282 | bool step; | |
283 | ||
284 | CT_WARN_ON(ct_state() != CONTEXT_KERNEL); | |
285 | ||
460d1245 AL |
286 | if (IS_ENABLED(CONFIG_PROVE_LOCKING) && |
287 | WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax)) | |
c5c46f59 AL |
288 | local_irq_enable(); |
289 | ||
290 | /* | |
291 | * First do one-time work. If these work items are enabled, we | |
292 | * want to run them exactly once per syscall exit with IRQs on. | |
293 | */ | |
294 | if (cached_flags & (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | | |
295 | _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)) { | |
296 | audit_syscall_exit(regs); | |
297 | ||
298 | if (cached_flags & _TIF_SYSCALL_TRACEPOINT) | |
299 | trace_sys_exit(regs, regs->ax); | |
300 | ||
301 | /* | |
302 | * If TIF_SYSCALL_EMU is set, we only get here because of | |
303 | * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP). | |
304 | * We already reported this syscall instruction in | |
305 | * syscall_trace_enter(). | |
306 | */ | |
307 | step = unlikely( | |
308 | (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU)) | |
309 | == _TIF_SINGLESTEP); | |
310 | if (step || cached_flags & _TIF_SYSCALL_TRACE) | |
311 | tracehook_report_syscall_exit(regs, step); | |
312 | } | |
313 | ||
314 | #ifdef CONFIG_COMPAT | |
315 | /* | |
316 | * Compat syscalls set TS_COMPAT. Make sure we clear it before | |
317 | * returning to user mode. | |
318 | */ | |
319 | ti->status &= ~TS_COMPAT; | |
320 | #endif | |
321 | ||
322 | local_irq_disable(); | |
323 | prepare_exit_to_usermode(regs); | |
324 | } | |
bd2d3a3b AL |
325 | |
326 | #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION) | |
327 | /* | |
8b13c255 | 328 | * Does a 32-bit syscall. Called with IRQs on and does all entry and |
33c52129 AL |
329 | * exit work and returns with IRQs off. This function is extremely hot |
330 | * in workloads that use it, and it's usually called from | |
331 | * do_fast_syscall_32, so forcibly inline it to improve performance. | |
bd2d3a3b | 332 | */ |
33c52129 | 333 | static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs) |
bd2d3a3b AL |
334 | { |
335 | struct thread_info *ti = pt_regs_to_thread_info(regs); | |
336 | unsigned int nr = (unsigned int)regs->orig_ax; | |
337 | ||
338 | #ifdef CONFIG_IA32_EMULATION | |
339 | ti->status |= TS_COMPAT; | |
340 | #endif | |
341 | ||
bd2d3a3b AL |
342 | if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) { |
343 | /* | |
344 | * Subtlety here: if ptrace pokes something larger than | |
345 | * 2^32-1 into orig_ax, this truncates it. This may or | |
346 | * may not be necessary, but it matches the old asm | |
347 | * behavior. | |
348 | */ | |
349 | nr = syscall_trace_enter(regs); | |
350 | } | |
351 | ||
33c52129 | 352 | if (likely(nr < IA32_NR_syscalls)) { |
bd2d3a3b AL |
353 | /* |
354 | * It's possible that a 32-bit syscall implementation | |
355 | * takes a 64-bit parameter but nonetheless assumes that | |
356 | * the high bits are zero. Make sure we zero-extend all | |
357 | * of the args. | |
358 | */ | |
359 | regs->ax = ia32_sys_call_table[nr]( | |
360 | (unsigned int)regs->bx, (unsigned int)regs->cx, | |
361 | (unsigned int)regs->dx, (unsigned int)regs->si, | |
362 | (unsigned int)regs->di, (unsigned int)regs->bp); | |
363 | } | |
364 | ||
365 | syscall_return_slowpath(regs); | |
366 | } | |
710246df | 367 | |
8b13c255 AL |
368 | /* Handles int $0x80 */ |
369 | __visible void do_int80_syscall_32(struct pt_regs *regs) | |
370 | { | |
371 | local_irq_enable(); | |
372 | do_syscall_32_irqs_on(regs); | |
373 | } | |
374 | ||
5f310f73 | 375 | /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */ |
7841b408 | 376 | __visible long do_fast_syscall_32(struct pt_regs *regs) |
710246df AL |
377 | { |
378 | /* | |
379 | * Called using the internal vDSO SYSENTER/SYSCALL32 calling | |
380 | * convention. Adjust regs so it looks like we entered using int80. | |
381 | */ | |
382 | ||
383 | unsigned long landing_pad = (unsigned long)current->mm->context.vdso + | |
384 | vdso_image_32.sym_int80_landing_pad; | |
385 | ||
386 | /* | |
387 | * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward | |
388 | * so that 'regs->ip -= 2' lands back on an int $0x80 instruction. | |
389 | * Fix it up. | |
390 | */ | |
391 | regs->ip = landing_pad; | |
392 | ||
393 | /* | |
394 | * Fetch ECX from where the vDSO stashed it. | |
395 | * | |
396 | * WARNING: We are in CONTEXT_USER and RCU isn't paying attention! | |
397 | */ | |
398 | local_irq_enable(); | |
c68ca678 AL |
399 | if ( |
400 | #ifdef CONFIG_X86_64 | |
401 | /* | |
402 | * Micro-optimization: the pointer we're following is explicitly | |
403 | * 32 bits, so it can't be out of range. | |
404 | */ | |
405 | __get_user(*(u32 *)®s->cx, | |
406 | (u32 __user __force *)(unsigned long)(u32)regs->sp) | |
407 | #else | |
408 | get_user(*(u32 *)®s->cx, | |
409 | (u32 __user __force *)(unsigned long)(u32)regs->sp) | |
410 | #endif | |
411 | ) { | |
412 | ||
710246df AL |
413 | /* User code screwed up. */ |
414 | local_irq_disable(); | |
415 | regs->ax = -EFAULT; | |
416 | #ifdef CONFIG_CONTEXT_TRACKING | |
417 | enter_from_user_mode(); | |
418 | #endif | |
419 | prepare_exit_to_usermode(regs); | |
7841b408 | 420 | return 0; /* Keep it simple: use IRET. */ |
710246df | 421 | } |
710246df AL |
422 | |
423 | /* Now this is just like a normal syscall. */ | |
8b13c255 | 424 | do_syscall_32_irqs_on(regs); |
7841b408 AL |
425 | |
426 | #ifdef CONFIG_X86_64 | |
427 | /* | |
428 | * Opportunistic SYSRETL: if possible, try to return using SYSRETL. | |
429 | * SYSRETL is available on all 64-bit CPUs, so we don't need to | |
430 | * bother with SYSEXIT. | |
431 | * | |
432 | * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP, | |
433 | * because the ECX fixup above will ensure that this is essentially | |
434 | * never the case. | |
435 | */ | |
436 | return regs->cs == __USER32_CS && regs->ss == __USER_DS && | |
437 | regs->ip == landing_pad && | |
438 | (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0; | |
439 | #else | |
5f310f73 AL |
440 | /* |
441 | * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT. | |
442 | * | |
443 | * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP, | |
444 | * because the ECX fixup above will ensure that this is essentially | |
445 | * never the case. | |
446 | * | |
447 | * We don't allow syscalls at all from VM86 mode, but we still | |
448 | * need to check VM, because we might be returning from sys_vm86. | |
449 | */ | |
450 | return static_cpu_has(X86_FEATURE_SEP) && | |
451 | regs->cs == __USER_CS && regs->ss == __USER_DS && | |
452 | regs->ip == landing_pad && | |
453 | (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0; | |
7841b408 | 454 | #endif |
710246df | 455 | } |
bd2d3a3b | 456 | #endif |