common.c 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420
  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. #include <linux/kernel.h>
  10. #include <linux/sched.h>
  11. #include <linux/mm.h>
  12. #include <linux/smp.h>
  13. #include <linux/errno.h>
  14. #include <linux/ptrace.h>
  15. #include <linux/tracehook.h>
  16. #include <linux/audit.h>
  17. #include <linux/seccomp.h>
  18. #include <linux/signal.h>
  19. #include <linux/export.h>
  20. #include <linux/context_tracking.h>
  21. #include <linux/user-return-notifier.h>
  22. #include <linux/uprobes.h>
  23. #include <asm/desc.h>
  24. #include <asm/traps.h>
  25. #include <asm/vdso.h>
  26. #include <asm/uaccess.h>
  27. #include <asm/cpufeature.h>
  28. #define CREATE_TRACE_POINTS
  29. #include <trace/events/syscalls.h>
  30. static struct thread_info *pt_regs_to_thread_info(struct pt_regs *regs)
  31. {
  32. unsigned long top_of_stack =
  33. (unsigned long)(regs + 1) + TOP_OF_KERNEL_STACK_PADDING;
  34. return (struct thread_info *)(top_of_stack - THREAD_SIZE);
  35. }
  36. #ifdef CONFIG_CONTEXT_TRACKING
  37. /* Called on entry from user mode with IRQs off. */
  38. __visible inline void enter_from_user_mode(void)
  39. {
  40. CT_WARN_ON(ct_state() != CONTEXT_USER);
  41. user_exit_irqoff();
  42. }
  43. #else
  44. static inline void enter_from_user_mode(void) {}
  45. #endif
  46. static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
  47. {
  48. #ifdef CONFIG_X86_64
  49. if (arch == AUDIT_ARCH_X86_64) {
  50. audit_syscall_entry(regs->orig_ax, regs->di,
  51. regs->si, regs->dx, regs->r10);
  52. } else
  53. #endif
  54. {
  55. audit_syscall_entry(regs->orig_ax, regs->bx,
  56. regs->cx, regs->dx, regs->si);
  57. }
  58. }
  59. /*
  60. * Returns the syscall nr to run (which should match regs->orig_ax) or -1
  61. * to skip the syscall.
  62. */
  63. static long syscall_trace_enter(struct pt_regs *regs)
  64. {
  65. u32 arch = in_ia32_syscall() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
  66. struct thread_info *ti = pt_regs_to_thread_info(regs);
  67. unsigned long ret = 0;
  68. bool emulated = false;
  69. u32 work;
  70. if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
  71. BUG_ON(regs != task_pt_regs(current));
  72. work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
  73. if (unlikely(work & _TIF_SYSCALL_EMU))
  74. emulated = true;
  75. if ((emulated || (work & _TIF_SYSCALL_TRACE)) &&
  76. tracehook_report_syscall_entry(regs))
  77. return -1L;
  78. if (emulated)
  79. return -1L;
  80. #ifdef CONFIG_SECCOMP
  81. /*
  82. * Do seccomp after ptrace, to catch any tracer changes.
  83. */
  84. if (work & _TIF_SECCOMP) {
  85. struct seccomp_data sd;
  86. sd.arch = arch;
  87. sd.nr = regs->orig_ax;
  88. sd.instruction_pointer = regs->ip;
  89. #ifdef CONFIG_X86_64
  90. if (arch == AUDIT_ARCH_X86_64) {
  91. sd.args[0] = regs->di;
  92. sd.args[1] = regs->si;
  93. sd.args[2] = regs->dx;
  94. sd.args[3] = regs->r10;
  95. sd.args[4] = regs->r8;
  96. sd.args[5] = regs->r9;
  97. } else
  98. #endif
  99. {
  100. sd.args[0] = regs->bx;
  101. sd.args[1] = regs->cx;
  102. sd.args[2] = regs->dx;
  103. sd.args[3] = regs->si;
  104. sd.args[4] = regs->di;
  105. sd.args[5] = regs->bp;
  106. }
  107. ret = __secure_computing(&sd);
  108. if (ret == -1)
  109. return ret;
  110. }
  111. #endif
  112. if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
  113. trace_sys_enter(regs, regs->orig_ax);
  114. do_audit_syscall_entry(regs, arch);
  115. return ret ?: regs->orig_ax;
  116. }
  117. #define EXIT_TO_USERMODE_LOOP_FLAGS \
  118. (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
  119. _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY)
  120. static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
  121. {
  122. /*
  123. * In order to return to user mode, we need to have IRQs off with
  124. * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY,
  125. * _TIF_UPROBE, or _TIF_NEED_RESCHED set. Several of these flags
  126. * can be set at any time on preemptable kernels if we have IRQs on,
  127. * so we need to loop. Disabling preemption wouldn't help: doing the
  128. * work to clear some of the flags can sleep.
  129. */
  130. while (true) {
  131. /* We have work to do. */
  132. local_irq_enable();
  133. if (cached_flags & _TIF_NEED_RESCHED)
  134. schedule();
  135. if (cached_flags & _TIF_UPROBE)
  136. uprobe_notify_resume(regs);
  137. /* deal with pending signal delivery */
  138. if (cached_flags & _TIF_SIGPENDING)
  139. do_signal(regs);
  140. if (cached_flags & _TIF_NOTIFY_RESUME) {
  141. clear_thread_flag(TIF_NOTIFY_RESUME);
  142. tracehook_notify_resume(regs);
  143. }
  144. if (cached_flags & _TIF_USER_RETURN_NOTIFY)
  145. fire_user_return_notifiers();
  146. /* Disable IRQs and retry */
  147. local_irq_disable();
  148. cached_flags = READ_ONCE(pt_regs_to_thread_info(regs)->flags);
  149. if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
  150. break;
  151. }
  152. }
  153. /* Called with IRQs disabled. */
  154. __visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
  155. {
  156. struct thread_info *ti = pt_regs_to_thread_info(regs);
  157. u32 cached_flags;
  158. if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
  159. local_irq_disable();
  160. lockdep_sys_exit();
  161. cached_flags = READ_ONCE(ti->flags);
  162. if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
  163. exit_to_usermode_loop(regs, cached_flags);
  164. #ifdef CONFIG_COMPAT
  165. /*
  166. * Compat syscalls set TS_COMPAT. Make sure we clear it before
  167. * returning to user mode. We need to clear it *after* signal
  168. * handling, because syscall restart has a fixup for compat
  169. * syscalls. The fixup is exercised by the ptrace_syscall_32
  170. * selftest.
  171. */
  172. ti->status &= ~TS_COMPAT;
  173. #endif
  174. user_enter_irqoff();
  175. }
  176. #define SYSCALL_EXIT_WORK_FLAGS \
  177. (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
  178. _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
  179. static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags)
  180. {
  181. bool step;
  182. audit_syscall_exit(regs);
  183. if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
  184. trace_sys_exit(regs, regs->ax);
  185. /*
  186. * If TIF_SYSCALL_EMU is set, we only get here because of
  187. * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
  188. * We already reported this syscall instruction in
  189. * syscall_trace_enter().
  190. */
  191. step = unlikely(
  192. (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
  193. == _TIF_SINGLESTEP);
  194. if (step || cached_flags & _TIF_SYSCALL_TRACE)
  195. tracehook_report_syscall_exit(regs, step);
  196. }
  197. /*
  198. * Called with IRQs on and fully valid regs. Returns with IRQs off in a
  199. * state such that we can immediately switch to user mode.
  200. */
  201. __visible inline void syscall_return_slowpath(struct pt_regs *regs)
  202. {
  203. struct thread_info *ti = pt_regs_to_thread_info(regs);
  204. u32 cached_flags = READ_ONCE(ti->flags);
  205. CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
  206. if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
  207. WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
  208. local_irq_enable();
  209. /*
  210. * First do one-time work. If these work items are enabled, we
  211. * want to run them exactly once per syscall exit with IRQs on.
  212. */
  213. if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS))
  214. syscall_slow_exit_work(regs, cached_flags);
  215. local_irq_disable();
  216. prepare_exit_to_usermode(regs);
  217. }
  218. #ifdef CONFIG_X86_64
  219. __visible void do_syscall_64(struct pt_regs *regs)
  220. {
  221. struct thread_info *ti = pt_regs_to_thread_info(regs);
  222. unsigned long nr = regs->orig_ax;
  223. enter_from_user_mode();
  224. local_irq_enable();
  225. if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY)
  226. nr = syscall_trace_enter(regs);
  227. /*
  228. * NB: Native and x32 syscalls are dispatched from the same
  229. * table. The only functional difference is the x32 bit in
  230. * regs->orig_ax, which changes the behavior of some syscalls.
  231. */
  232. if (likely((nr & __SYSCALL_MASK) < NR_syscalls)) {
  233. regs->ax = sys_call_table[nr & __SYSCALL_MASK](
  234. regs->di, regs->si, regs->dx,
  235. regs->r10, regs->r8, regs->r9);
  236. }
  237. syscall_return_slowpath(regs);
  238. }
  239. #endif
  240. #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
  241. /*
  242. * Does a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. Does
  243. * all entry and exit work and returns with IRQs off. This function is
  244. * extremely hot in workloads that use it, and it's usually called from
  245. * do_fast_syscall_32, so forcibly inline it to improve performance.
  246. */
  247. static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
  248. {
  249. struct thread_info *ti = pt_regs_to_thread_info(regs);
  250. unsigned int nr = (unsigned int)regs->orig_ax;
  251. #ifdef CONFIG_IA32_EMULATION
  252. ti->status |= TS_COMPAT;
  253. #endif
  254. if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
  255. /*
  256. * Subtlety here: if ptrace pokes something larger than
  257. * 2^32-1 into orig_ax, this truncates it. This may or
  258. * may not be necessary, but it matches the old asm
  259. * behavior.
  260. */
  261. nr = syscall_trace_enter(regs);
  262. }
  263. if (likely(nr < IA32_NR_syscalls)) {
  264. /*
  265. * It's possible that a 32-bit syscall implementation
  266. * takes a 64-bit parameter but nonetheless assumes that
  267. * the high bits are zero. Make sure we zero-extend all
  268. * of the args.
  269. */
  270. regs->ax = ia32_sys_call_table[nr](
  271. (unsigned int)regs->bx, (unsigned int)regs->cx,
  272. (unsigned int)regs->dx, (unsigned int)regs->si,
  273. (unsigned int)regs->di, (unsigned int)regs->bp);
  274. }
  275. syscall_return_slowpath(regs);
  276. }
  277. /* Handles int $0x80 */
  278. __visible void do_int80_syscall_32(struct pt_regs *regs)
  279. {
  280. enter_from_user_mode();
  281. local_irq_enable();
  282. do_syscall_32_irqs_on(regs);
  283. }
  284. /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
  285. __visible long do_fast_syscall_32(struct pt_regs *regs)
  286. {
  287. /*
  288. * Called using the internal vDSO SYSENTER/SYSCALL32 calling
  289. * convention. Adjust regs so it looks like we entered using int80.
  290. */
  291. unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
  292. vdso_image_32.sym_int80_landing_pad;
  293. /*
  294. * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
  295. * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
  296. * Fix it up.
  297. */
  298. regs->ip = landing_pad;
  299. enter_from_user_mode();
  300. local_irq_enable();
  301. /* Fetch EBP from where the vDSO stashed it. */
  302. if (
  303. #ifdef CONFIG_X86_64
  304. /*
  305. * Micro-optimization: the pointer we're following is explicitly
  306. * 32 bits, so it can't be out of range.
  307. */
  308. __get_user(*(u32 *)&regs->bp,
  309. (u32 __user __force *)(unsigned long)(u32)regs->sp)
  310. #else
  311. get_user(*(u32 *)&regs->bp,
  312. (u32 __user __force *)(unsigned long)(u32)regs->sp)
  313. #endif
  314. ) {
  315. /* User code screwed up. */
  316. local_irq_disable();
  317. regs->ax = -EFAULT;
  318. prepare_exit_to_usermode(regs);
  319. return 0; /* Keep it simple: use IRET. */
  320. }
  321. /* Now this is just like a normal syscall. */
  322. do_syscall_32_irqs_on(regs);
  323. #ifdef CONFIG_X86_64
  324. /*
  325. * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
  326. * SYSRETL is available on all 64-bit CPUs, so we don't need to
  327. * bother with SYSEXIT.
  328. *
  329. * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
  330. * because the ECX fixup above will ensure that this is essentially
  331. * never the case.
  332. */
  333. return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
  334. regs->ip == landing_pad &&
  335. (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
  336. #else
  337. /*
  338. * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
  339. *
  340. * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
  341. * because the ECX fixup above will ensure that this is essentially
  342. * never the case.
  343. *
  344. * We don't allow syscalls at all from VM86 mode, but we still
  345. * need to check VM, because we might be returning from sys_vm86.
  346. */
  347. return static_cpu_has(X86_FEATURE_SEP) &&
  348. regs->cs == __USER_CS && regs->ss == __USER_DS &&
  349. regs->ip == landing_pad &&
  350. (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;
  351. #endif
  352. }
  353. #endif