blob: d5eee851071cd6d8e08dac1369d47032fb46ea50 [file] [log] [blame]
Andy Lutomirski1f484aa2015-07-03 12:44:23 -07001/*
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>
Andy Lutomirski710246d2015-10-05 17:48:10 -070027#include <asm/vdso.h>
28#include <asm/uaccess.h>
Andy Lutomirski1f484aa2015-07-03 12:44:23 -070029
30#define CREATE_TRACE_POINTS
31#include <trace/events/syscalls.h>
32
Andy Lutomirskifeed36c2015-07-03 12:44:25 -070033#ifdef CONFIG_CONTEXT_TRACKING
34/* Called on entry from user mode with IRQs off. */
35__visible void enter_from_user_mode(void)
36{
37 CT_WARN_ON(ct_state() != CONTEXT_USER);
38 user_exit();
39}
40#endif
41
Andy Lutomirski1f484aa2015-07-03 12:44:23 -070042static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
43{
44#ifdef CONFIG_X86_64
45 if (arch == AUDIT_ARCH_X86_64) {
46 audit_syscall_entry(regs->orig_ax, regs->di,
47 regs->si, regs->dx, regs->r10);
48 } else
49#endif
50 {
51 audit_syscall_entry(regs->orig_ax, regs->bx,
52 regs->cx, regs->dx, regs->si);
53 }
54}
55
56/*
57 * We can return 0 to resume the syscall or anything else to go to phase
58 * 2. If we resume the syscall, we need to put something appropriate in
59 * regs->orig_ax.
60 *
61 * NB: We don't have full pt_regs here, but regs->orig_ax and regs->ax
62 * are fully functional.
63 *
64 * For phase 2's benefit, our return value is:
65 * 0: resume the syscall
66 * 1: go to phase 2; no seccomp phase 2 needed
67 * anything else: go to phase 2; pass return value to seccomp
68 */
69unsigned long syscall_trace_enter_phase1(struct pt_regs *regs, u32 arch)
70{
71 unsigned long ret = 0;
72 u32 work;
73
74 BUG_ON(regs != task_pt_regs(current));
75
76 work = ACCESS_ONCE(current_thread_info()->flags) &
77 _TIF_WORK_SYSCALL_ENTRY;
78
Andy Lutomirskifeed36c2015-07-03 12:44:25 -070079#ifdef CONFIG_CONTEXT_TRACKING
Andy Lutomirski1f484aa2015-07-03 12:44:23 -070080 /*
81 * If TIF_NOHZ is set, we are required to call user_exit() before
82 * doing anything that could touch RCU.
83 */
84 if (work & _TIF_NOHZ) {
Andy Lutomirskifeed36c2015-07-03 12:44:25 -070085 enter_from_user_mode();
Andy Lutomirski1f484aa2015-07-03 12:44:23 -070086 work &= ~_TIF_NOHZ;
87 }
Andy Lutomirskifeed36c2015-07-03 12:44:25 -070088#endif
Andy Lutomirski1f484aa2015-07-03 12:44:23 -070089
90#ifdef CONFIG_SECCOMP
91 /*
92 * Do seccomp first -- it should minimize exposure of other
93 * code, and keeping seccomp fast is probably more valuable
94 * than the rest of this.
95 */
96 if (work & _TIF_SECCOMP) {
97 struct seccomp_data sd;
98
99 sd.arch = arch;
100 sd.nr = regs->orig_ax;
101 sd.instruction_pointer = regs->ip;
102#ifdef CONFIG_X86_64
103 if (arch == AUDIT_ARCH_X86_64) {
104 sd.args[0] = regs->di;
105 sd.args[1] = regs->si;
106 sd.args[2] = regs->dx;
107 sd.args[3] = regs->r10;
108 sd.args[4] = regs->r8;
109 sd.args[5] = regs->r9;
110 } else
111#endif
112 {
113 sd.args[0] = regs->bx;
114 sd.args[1] = regs->cx;
115 sd.args[2] = regs->dx;
116 sd.args[3] = regs->si;
117 sd.args[4] = regs->di;
118 sd.args[5] = regs->bp;
119 }
120
121 BUILD_BUG_ON(SECCOMP_PHASE1_OK != 0);
122 BUILD_BUG_ON(SECCOMP_PHASE1_SKIP != 1);
123
124 ret = seccomp_phase1(&sd);
125 if (ret == SECCOMP_PHASE1_SKIP) {
126 regs->orig_ax = -1;
127 ret = 0;
128 } else if (ret != SECCOMP_PHASE1_OK) {
129 return ret; /* Go directly to phase 2 */
130 }
131
132 work &= ~_TIF_SECCOMP;
133 }
134#endif
135
136 /* Do our best to finish without phase 2. */
137 if (work == 0)
138 return ret; /* seccomp and/or nohz only (ret == 0 here) */
139
140#ifdef CONFIG_AUDITSYSCALL
141 if (work == _TIF_SYSCALL_AUDIT) {
142 /*
143 * If there is no more work to be done except auditing,
144 * then audit in phase 1. Phase 2 always audits, so, if
145 * we audit here, then we can't go on to phase 2.
146 */
147 do_audit_syscall_entry(regs, arch);
148 return 0;
149 }
150#endif
151
152 return 1; /* Something is enabled that we can't handle in phase 1 */
153}
154
155/* Returns the syscall nr to run (which should match regs->orig_ax). */
156long syscall_trace_enter_phase2(struct pt_regs *regs, u32 arch,
157 unsigned long phase1_result)
158{
159 long ret = 0;
160 u32 work = ACCESS_ONCE(current_thread_info()->flags) &
161 _TIF_WORK_SYSCALL_ENTRY;
162
163 BUG_ON(regs != task_pt_regs(current));
164
165 /*
166 * If we stepped into a sysenter/syscall insn, it trapped in
167 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
168 * If user-mode had set TF itself, then it's still clear from
169 * do_debug() and we need to set it again to restore the user
170 * state. If we entered on the slow path, TF was already set.
171 */
172 if (work & _TIF_SINGLESTEP)
173 regs->flags |= X86_EFLAGS_TF;
174
175#ifdef CONFIG_SECCOMP
176 /*
177 * Call seccomp_phase2 before running the other hooks so that
178 * they can see any changes made by a seccomp tracer.
179 */
180 if (phase1_result > 1 && seccomp_phase2(phase1_result)) {
181 /* seccomp failures shouldn't expose any additional code. */
182 return -1;
183 }
184#endif
185
186 if (unlikely(work & _TIF_SYSCALL_EMU))
187 ret = -1L;
188
189 if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
190 tracehook_report_syscall_entry(regs))
191 ret = -1L;
192
193 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
194 trace_sys_enter(regs, regs->orig_ax);
195
196 do_audit_syscall_entry(regs, arch);
197
198 return ret ?: regs->orig_ax;
199}
200
201long syscall_trace_enter(struct pt_regs *regs)
202{
203 u32 arch = is_ia32_task() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
204 unsigned long phase1_result = syscall_trace_enter_phase1(regs, arch);
205
206 if (phase1_result == 0)
207 return regs->orig_ax;
208 else
209 return syscall_trace_enter_phase2(regs, arch, phase1_result);
210}
211
Andy Lutomirskic5c46f52015-07-03 12:44:26 -0700212static struct thread_info *pt_regs_to_thread_info(struct pt_regs *regs)
213{
214 unsigned long top_of_stack =
215 (unsigned long)(regs + 1) + TOP_OF_KERNEL_STACK_PADDING;
216 return (struct thread_info *)(top_of_stack - THREAD_SIZE);
217}
218
219/* Called with IRQs disabled. */
220__visible void prepare_exit_to_usermode(struct pt_regs *regs)
221{
Andy Lutomirski460d1242015-10-05 17:48:18 -0700222 if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
Andy Lutomirskic5c46f52015-07-03 12:44:26 -0700223 local_irq_disable();
224
Andy Lutomirski72f92472015-10-05 17:47:54 -0700225 lockdep_sys_exit();
226
Andy Lutomirskic5c46f52015-07-03 12:44:26 -0700227 /*
228 * In order to return to user mode, we need to have IRQs off with
229 * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY,
230 * _TIF_UPROBE, or _TIF_NEED_RESCHED set. Several of these flags
231 * can be set at any time on preemptable kernels if we have IRQs on,
232 * so we need to loop. Disabling preemption wouldn't help: doing the
233 * work to clear some of the flags can sleep.
234 */
235 while (true) {
236 u32 cached_flags =
237 READ_ONCE(pt_regs_to_thread_info(regs)->flags);
238
239 if (!(cached_flags & (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME |
Andy Lutomirskid1328032015-07-15 14:25:16 -0700240 _TIF_UPROBE | _TIF_NEED_RESCHED |
241 _TIF_USER_RETURN_NOTIFY)))
Andy Lutomirskic5c46f52015-07-03 12:44:26 -0700242 break;
243
244 /* We have work to do. */
245 local_irq_enable();
246
247 if (cached_flags & _TIF_NEED_RESCHED)
248 schedule();
249
250 if (cached_flags & _TIF_UPROBE)
251 uprobe_notify_resume(regs);
252
253 /* deal with pending signal delivery */
254 if (cached_flags & _TIF_SIGPENDING)
255 do_signal(regs);
256
257 if (cached_flags & _TIF_NOTIFY_RESUME) {
258 clear_thread_flag(TIF_NOTIFY_RESUME);
259 tracehook_notify_resume(regs);
260 }
261
262 if (cached_flags & _TIF_USER_RETURN_NOTIFY)
263 fire_user_return_notifiers();
264
265 /* Disable IRQs and retry */
266 local_irq_disable();
267 }
268
269 user_enter();
270}
271
Andy Lutomirski1f484aa2015-07-03 12:44:23 -0700272/*
Andy Lutomirskic5c46f52015-07-03 12:44:26 -0700273 * Called with IRQs on and fully valid regs. Returns with IRQs off in a
274 * state such that we can immediately switch to user mode.
275 */
276__visible void syscall_return_slowpath(struct pt_regs *regs)
277{
278 struct thread_info *ti = pt_regs_to_thread_info(regs);
279 u32 cached_flags = READ_ONCE(ti->flags);
280 bool step;
281
282 CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
283
Andy Lutomirski460d1242015-10-05 17:48:18 -0700284 if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
285 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
Andy Lutomirskic5c46f52015-07-03 12:44:26 -0700286 local_irq_enable();
287
288 /*
289 * First do one-time work. If these work items are enabled, we
290 * want to run them exactly once per syscall exit with IRQs on.
291 */
292 if (cached_flags & (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT |
293 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)) {
294 audit_syscall_exit(regs);
295
296 if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
297 trace_sys_exit(regs, regs->ax);
298
299 /*
300 * If TIF_SYSCALL_EMU is set, we only get here because of
301 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
302 * We already reported this syscall instruction in
303 * syscall_trace_enter().
304 */
305 step = unlikely(
306 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
307 == _TIF_SINGLESTEP);
308 if (step || cached_flags & _TIF_SYSCALL_TRACE)
309 tracehook_report_syscall_exit(regs, step);
310 }
311
312#ifdef CONFIG_COMPAT
313 /*
314 * Compat syscalls set TS_COMPAT. Make sure we clear it before
315 * returning to user mode.
316 */
317 ti->status &= ~TS_COMPAT;
318#endif
319
320 local_irq_disable();
321 prepare_exit_to_usermode(regs);
322}
Andy Lutomirskibd2d3a32015-10-05 17:48:08 -0700323
324#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
325/*
Andy Lutomirski8b13c252015-10-05 17:48:17 -0700326 * Does a 32-bit syscall. Called with IRQs on and does all entry and
Andy Lutomirski33c521292015-10-05 17:48:19 -0700327 * exit work and returns with IRQs off. This function is extremely hot
328 * in workloads that use it, and it's usually called from
329 * do_fast_syscall_32, so forcibly inline it to improve performance.
Andy Lutomirskibd2d3a32015-10-05 17:48:08 -0700330 */
Andy Lutomirski33c521292015-10-05 17:48:19 -0700331static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
Andy Lutomirskibd2d3a32015-10-05 17:48:08 -0700332{
333 struct thread_info *ti = pt_regs_to_thread_info(regs);
334 unsigned int nr = (unsigned int)regs->orig_ax;
335
336#ifdef CONFIG_IA32_EMULATION
337 ti->status |= TS_COMPAT;
338#endif
339
Andy Lutomirskibd2d3a32015-10-05 17:48:08 -0700340 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
341 /*
342 * Subtlety here: if ptrace pokes something larger than
343 * 2^32-1 into orig_ax, this truncates it. This may or
344 * may not be necessary, but it matches the old asm
345 * behavior.
346 */
347 nr = syscall_trace_enter(regs);
348 }
349
Andy Lutomirski33c521292015-10-05 17:48:19 -0700350 if (likely(nr < IA32_NR_syscalls)) {
Andy Lutomirskibd2d3a32015-10-05 17:48:08 -0700351 /*
352 * It's possible that a 32-bit syscall implementation
353 * takes a 64-bit parameter but nonetheless assumes that
354 * the high bits are zero. Make sure we zero-extend all
355 * of the args.
356 */
357 regs->ax = ia32_sys_call_table[nr](
358 (unsigned int)regs->bx, (unsigned int)regs->cx,
359 (unsigned int)regs->dx, (unsigned int)regs->si,
360 (unsigned int)regs->di, (unsigned int)regs->bp);
361 }
362
363 syscall_return_slowpath(regs);
364}
Andy Lutomirski710246d2015-10-05 17:48:10 -0700365
Andy Lutomirski8b13c252015-10-05 17:48:17 -0700366/* Handles int $0x80 */
367__visible void do_int80_syscall_32(struct pt_regs *regs)
368{
369 local_irq_enable();
370 do_syscall_32_irqs_on(regs);
371}
372
Andy Lutomirski5f310f72015-10-05 17:48:15 -0700373/* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
Andy Lutomirski7841b402015-10-05 17:48:12 -0700374__visible long do_fast_syscall_32(struct pt_regs *regs)
Andy Lutomirski710246d2015-10-05 17:48:10 -0700375{
376 /*
377 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
378 * convention. Adjust regs so it looks like we entered using int80.
379 */
380
381 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
382 vdso_image_32.sym_int80_landing_pad;
383
384 /*
385 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
386 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
387 * Fix it up.
388 */
389 regs->ip = landing_pad;
390
391 /*
392 * Fetch ECX from where the vDSO stashed it.
393 *
394 * WARNING: We are in CONTEXT_USER and RCU isn't paying attention!
395 */
396 local_irq_enable();
397 if (get_user(*(u32 *)&regs->cx,
398 (u32 __user __force *)(unsigned long)(u32)regs->sp)) {
399 /* User code screwed up. */
400 local_irq_disable();
401 regs->ax = -EFAULT;
402#ifdef CONFIG_CONTEXT_TRACKING
403 enter_from_user_mode();
404#endif
405 prepare_exit_to_usermode(regs);
Andy Lutomirski7841b402015-10-05 17:48:12 -0700406 return 0; /* Keep it simple: use IRET. */
Andy Lutomirski710246d2015-10-05 17:48:10 -0700407 }
Andy Lutomirski710246d2015-10-05 17:48:10 -0700408
409 /* Now this is just like a normal syscall. */
Andy Lutomirski8b13c252015-10-05 17:48:17 -0700410 do_syscall_32_irqs_on(regs);
Andy Lutomirski7841b402015-10-05 17:48:12 -0700411
412#ifdef CONFIG_X86_64
413 /*
414 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
415 * SYSRETL is available on all 64-bit CPUs, so we don't need to
416 * bother with SYSEXIT.
417 *
418 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
419 * because the ECX fixup above will ensure that this is essentially
420 * never the case.
421 */
422 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
423 regs->ip == landing_pad &&
424 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
425#else
Andy Lutomirski5f310f72015-10-05 17:48:15 -0700426 /*
427 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
428 *
429 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
430 * because the ECX fixup above will ensure that this is essentially
431 * never the case.
432 *
433 * We don't allow syscalls at all from VM86 mode, but we still
434 * need to check VM, because we might be returning from sys_vm86.
435 */
436 return static_cpu_has(X86_FEATURE_SEP) &&
437 regs->cs == __USER_CS && regs->ss == __USER_DS &&
438 regs->ip == landing_pad &&
439 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;
Andy Lutomirski7841b402015-10-05 17:48:12 -0700440#endif
Andy Lutomirski710246d2015-10-05 17:48:10 -0700441}
Andy Lutomirskibd2d3a32015-10-05 17:48:08 -0700442#endif