~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/x86/kernel/unwind_orc.c

Version: ~ [ linux-5.5-rc1 ] ~ [ linux-5.4.2 ] ~ [ linux-5.3.15 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.88 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.158 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.206 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.206 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.78 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 #include <linux/module.h>
  2 #include <linux/sort.h>
  3 #include <asm/ptrace.h>
  4 #include <asm/stacktrace.h>
  5 #include <asm/unwind.h>
  6 #include <asm/orc_types.h>
  7 #include <asm/orc_lookup.h>
  8 
  9 #define orc_warn(fmt, ...) \
 10         printk_deferred_once(KERN_WARNING pr_fmt("WARNING: " fmt), ##__VA_ARGS__)
 11 
 12 extern int __start_orc_unwind_ip[];
 13 extern int __stop_orc_unwind_ip[];
 14 extern struct orc_entry __start_orc_unwind[];
 15 extern struct orc_entry __stop_orc_unwind[];
 16 
 17 static DEFINE_MUTEX(sort_mutex);
 18 int *cur_orc_ip_table = __start_orc_unwind_ip;
 19 struct orc_entry *cur_orc_table = __start_orc_unwind;
 20 
 21 unsigned int lookup_num_blocks;
 22 bool orc_init;
 23 
 24 static inline unsigned long orc_ip(const int *ip)
 25 {
 26         return (unsigned long)ip + *ip;
 27 }
 28 
 29 static struct orc_entry *__orc_find(int *ip_table, struct orc_entry *u_table,
 30                                     unsigned int num_entries, unsigned long ip)
 31 {
 32         int *first = ip_table;
 33         int *last = ip_table + num_entries - 1;
 34         int *mid = first, *found = first;
 35 
 36         if (!num_entries)
 37                 return NULL;
 38 
 39         /*
 40          * Do a binary range search to find the rightmost duplicate of a given
 41          * starting address.  Some entries are section terminators which are
 42          * "weak" entries for ensuring there are no gaps.  They should be
 43          * ignored when they conflict with a real entry.
 44          */
 45         while (first <= last) {
 46                 mid = first + ((last - first) / 2);
 47 
 48                 if (orc_ip(mid) <= ip) {
 49                         found = mid;
 50                         first = mid + 1;
 51                 } else
 52                         last = mid - 1;
 53         }
 54 
 55         return u_table + (found - ip_table);
 56 }
 57 
 58 #ifdef CONFIG_MODULES
 59 static struct orc_entry *orc_module_find(unsigned long ip)
 60 {
 61         struct module *mod;
 62 
 63         mod = __module_address(ip);
 64         if (!mod || !mod->arch.orc_unwind || !mod->arch.orc_unwind_ip)
 65                 return NULL;
 66         return __orc_find(mod->arch.orc_unwind_ip, mod->arch.orc_unwind,
 67                           mod->arch.num_orcs, ip);
 68 }
 69 #else
 70 static struct orc_entry *orc_module_find(unsigned long ip)
 71 {
 72         return NULL;
 73 }
 74 #endif
 75 
 76 #ifdef CONFIG_DYNAMIC_FTRACE
 77 static struct orc_entry *orc_find(unsigned long ip);
 78 
 79 /*
 80  * Ftrace dynamic trampolines do not have orc entries of their own.
 81  * But they are copies of the ftrace entries that are static and
 82  * defined in ftrace_*.S, which do have orc entries.
 83  *
 84  * If the undwinder comes across a ftrace trampoline, then find the
 85  * ftrace function that was used to create it, and use that ftrace
 86  * function's orc entrie, as the placement of the return code in
 87  * the stack will be identical.
 88  */
 89 static struct orc_entry *orc_ftrace_find(unsigned long ip)
 90 {
 91         struct ftrace_ops *ops;
 92         unsigned long caller;
 93 
 94         ops = ftrace_ops_trampoline(ip);
 95         if (!ops)
 96                 return NULL;
 97 
 98         if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
 99                 caller = (unsigned long)ftrace_regs_call;
100         else
101                 caller = (unsigned long)ftrace_call;
102 
103         /* Prevent unlikely recursion */
104         if (ip == caller)
105                 return NULL;
106 
107         return orc_find(caller);
108 }
109 #else
110 static struct orc_entry *orc_ftrace_find(unsigned long ip)
111 {
112         return NULL;
113 }
114 #endif
115 
116 static struct orc_entry *orc_find(unsigned long ip)
117 {
118         static struct orc_entry *orc;
119 
120         if (!orc_init)
121                 return NULL;
122 
123         /* For non-init vmlinux addresses, use the fast lookup table: */
124         if (ip >= LOOKUP_START_IP && ip < LOOKUP_STOP_IP) {
125                 unsigned int idx, start, stop;
126 
127                 idx = (ip - LOOKUP_START_IP) / LOOKUP_BLOCK_SIZE;
128 
129                 if (unlikely((idx >= lookup_num_blocks-1))) {
130                         orc_warn("WARNING: bad lookup idx: idx=%u num=%u ip=%pB\n",
131                                  idx, lookup_num_blocks, (void *)ip);
132                         return NULL;
133                 }
134 
135                 start = orc_lookup[idx];
136                 stop = orc_lookup[idx + 1] + 1;
137 
138                 if (unlikely((__start_orc_unwind + start >= __stop_orc_unwind) ||
139                              (__start_orc_unwind + stop > __stop_orc_unwind))) {
140                         orc_warn("WARNING: bad lookup value: idx=%u num=%u start=%u stop=%u ip=%pB\n",
141                                  idx, lookup_num_blocks, start, stop, (void *)ip);
142                         return NULL;
143                 }
144 
145                 return __orc_find(__start_orc_unwind_ip + start,
146                                   __start_orc_unwind + start, stop - start, ip);
147         }
148 
149         /* vmlinux .init slow lookup: */
150         if (init_kernel_text(ip))
151                 return __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
152                                   __stop_orc_unwind_ip - __start_orc_unwind_ip, ip);
153 
154         /* Module lookup: */
155         orc = orc_module_find(ip);
156         if (orc)
157                 return orc;
158 
159         return orc_ftrace_find(ip);
160 }
161 
162 static void orc_sort_swap(void *_a, void *_b, int size)
163 {
164         struct orc_entry *orc_a, *orc_b;
165         struct orc_entry orc_tmp;
166         int *a = _a, *b = _b, tmp;
167         int delta = _b - _a;
168 
169         /* Swap the .orc_unwind_ip entries: */
170         tmp = *a;
171         *a = *b + delta;
172         *b = tmp - delta;
173 
174         /* Swap the corresponding .orc_unwind entries: */
175         orc_a = cur_orc_table + (a - cur_orc_ip_table);
176         orc_b = cur_orc_table + (b - cur_orc_ip_table);
177         orc_tmp = *orc_a;
178         *orc_a = *orc_b;
179         *orc_b = orc_tmp;
180 }
181 
182 static int orc_sort_cmp(const void *_a, const void *_b)
183 {
184         struct orc_entry *orc_a;
185         const int *a = _a, *b = _b;
186         unsigned long a_val = orc_ip(a);
187         unsigned long b_val = orc_ip(b);
188 
189         if (a_val > b_val)
190                 return 1;
191         if (a_val < b_val)
192                 return -1;
193 
194         /*
195          * The "weak" section terminator entries need to always be on the left
196          * to ensure the lookup code skips them in favor of real entries.
197          * These terminator entries exist to handle any gaps created by
198          * whitelisted .o files which didn't get objtool generation.
199          */
200         orc_a = cur_orc_table + (a - cur_orc_ip_table);
201         return orc_a->sp_reg == ORC_REG_UNDEFINED && !orc_a->end ? -1 : 1;
202 }
203 
204 #ifdef CONFIG_MODULES
205 void unwind_module_init(struct module *mod, void *_orc_ip, size_t orc_ip_size,
206                         void *_orc, size_t orc_size)
207 {
208         int *orc_ip = _orc_ip;
209         struct orc_entry *orc = _orc;
210         unsigned int num_entries = orc_ip_size / sizeof(int);
211 
212         WARN_ON_ONCE(orc_ip_size % sizeof(int) != 0 ||
213                      orc_size % sizeof(*orc) != 0 ||
214                      num_entries != orc_size / sizeof(*orc));
215 
216         /*
217          * The 'cur_orc_*' globals allow the orc_sort_swap() callback to
218          * associate an .orc_unwind_ip table entry with its corresponding
219          * .orc_unwind entry so they can both be swapped.
220          */
221         mutex_lock(&sort_mutex);
222         cur_orc_ip_table = orc_ip;
223         cur_orc_table = orc;
224         sort(orc_ip, num_entries, sizeof(int), orc_sort_cmp, orc_sort_swap);
225         mutex_unlock(&sort_mutex);
226 
227         mod->arch.orc_unwind_ip = orc_ip;
228         mod->arch.orc_unwind = orc;
229         mod->arch.num_orcs = num_entries;
230 }
231 #endif
232 
233 void __init unwind_init(void)
234 {
235         size_t orc_ip_size = (void *)__stop_orc_unwind_ip - (void *)__start_orc_unwind_ip;
236         size_t orc_size = (void *)__stop_orc_unwind - (void *)__start_orc_unwind;
237         size_t num_entries = orc_ip_size / sizeof(int);
238         struct orc_entry *orc;
239         int i;
240 
241         if (!num_entries || orc_ip_size % sizeof(int) != 0 ||
242             orc_size % sizeof(struct orc_entry) != 0 ||
243             num_entries != orc_size / sizeof(struct orc_entry)) {
244                 orc_warn("WARNING: Bad or missing .orc_unwind table.  Disabling unwinder.\n");
245                 return;
246         }
247 
248         /* Sort the .orc_unwind and .orc_unwind_ip tables: */
249         sort(__start_orc_unwind_ip, num_entries, sizeof(int), orc_sort_cmp,
250              orc_sort_swap);
251 
252         /* Initialize the fast lookup table: */
253         lookup_num_blocks = orc_lookup_end - orc_lookup;
254         for (i = 0; i < lookup_num_blocks-1; i++) {
255                 orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
256                                  num_entries,
257                                  LOOKUP_START_IP + (LOOKUP_BLOCK_SIZE * i));
258                 if (!orc) {
259                         orc_warn("WARNING: Corrupt .orc_unwind table.  Disabling unwinder.\n");
260                         return;
261                 }
262 
263                 orc_lookup[i] = orc - __start_orc_unwind;
264         }
265 
266         /* Initialize the ending block: */
267         orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind, num_entries,
268                          LOOKUP_STOP_IP);
269         if (!orc) {
270                 orc_warn("WARNING: Corrupt .orc_unwind table.  Disabling unwinder.\n");
271                 return;
272         }
273         orc_lookup[lookup_num_blocks-1] = orc - __start_orc_unwind;
274 
275         orc_init = true;
276 }
277 
278 unsigned long unwind_get_return_address(struct unwind_state *state)
279 {
280         if (unwind_done(state))
281                 return 0;
282 
283         return __kernel_text_address(state->ip) ? state->ip : 0;
284 }
285 EXPORT_SYMBOL_GPL(unwind_get_return_address);
286 
287 unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
288 {
289         if (unwind_done(state))
290                 return NULL;
291 
292         if (state->regs)
293                 return &state->regs->ip;
294 
295         if (state->sp)
296                 return (unsigned long *)state->sp - 1;
297 
298         return NULL;
299 }
300 
301 static bool stack_access_ok(struct unwind_state *state, unsigned long _addr,
302                             size_t len)
303 {
304         struct stack_info *info = &state->stack_info;
305         void *addr = (void *)_addr;
306 
307         if (!on_stack(info, addr, len) &&
308             (get_stack_info(addr, state->task, info, &state->stack_mask)))
309                 return false;
310 
311         return true;
312 }
313 
314 static bool deref_stack_reg(struct unwind_state *state, unsigned long addr,
315                             unsigned long *val)
316 {
317         if (!stack_access_ok(state, addr, sizeof(long)))
318                 return false;
319 
320         *val = READ_ONCE_NOCHECK(*(unsigned long *)addr);
321         return true;
322 }
323 
324 static bool deref_stack_regs(struct unwind_state *state, unsigned long addr,
325                              unsigned long *ip, unsigned long *sp)
326 {
327         struct pt_regs *regs = (struct pt_regs *)addr;
328 
329         /* x86-32 support will be more complicated due to the &regs->sp hack */
330         BUILD_BUG_ON(IS_ENABLED(CONFIG_X86_32));
331 
332         if (!stack_access_ok(state, addr, sizeof(struct pt_regs)))
333                 return false;
334 
335         *ip = regs->ip;
336         *sp = regs->sp;
337         return true;
338 }
339 
340 static bool deref_stack_iret_regs(struct unwind_state *state, unsigned long addr,
341                                   unsigned long *ip, unsigned long *sp)
342 {
343         struct pt_regs *regs = (void *)addr - IRET_FRAME_OFFSET;
344 
345         if (!stack_access_ok(state, addr, IRET_FRAME_SIZE))
346                 return false;
347 
348         *ip = regs->ip;
349         *sp = regs->sp;
350         return true;
351 }
352 
353 bool unwind_next_frame(struct unwind_state *state)
354 {
355         unsigned long ip_p, sp, orig_ip = state->ip, prev_sp = state->sp;
356         enum stack_type prev_type = state->stack_info.type;
357         struct orc_entry *orc;
358         bool indirect = false;
359 
360         if (unwind_done(state))
361                 return false;
362 
363         /* Don't let modules unload while we're reading their ORC data. */
364         preempt_disable();
365 
366         /* End-of-stack check for user tasks: */
367         if (state->regs && user_mode(state->regs))
368                 goto the_end;
369 
370         /*
371          * Find the orc_entry associated with the text address.
372          *
373          * Decrement call return addresses by one so they work for sibling
374          * calls and calls to noreturn functions.
375          */
376         orc = orc_find(state->signal ? state->ip : state->ip - 1);
377         if (!orc)
378                 goto err;
379 
380         /* End-of-stack check for kernel threads: */
381         if (orc->sp_reg == ORC_REG_UNDEFINED) {
382                 if (!orc->end)
383                         goto err;
384 
385                 goto the_end;
386         }
387 
388         /* Find the previous frame's stack: */
389         switch (orc->sp_reg) {
390         case ORC_REG_SP:
391                 sp = state->sp + orc->sp_offset;
392                 break;
393 
394         case ORC_REG_BP:
395                 sp = state->bp + orc->sp_offset;
396                 break;
397 
398         case ORC_REG_SP_INDIRECT:
399                 sp = state->sp + orc->sp_offset;
400                 indirect = true;
401                 break;
402 
403         case ORC_REG_BP_INDIRECT:
404                 sp = state->bp + orc->sp_offset;
405                 indirect = true;
406                 break;
407 
408         case ORC_REG_R10:
409                 if (!state->regs || !state->full_regs) {
410                         orc_warn("missing regs for base reg R10 at ip %pB\n",
411                                  (void *)state->ip);
412                         goto err;
413                 }
414                 sp = state->regs->r10;
415                 break;
416 
417         case ORC_REG_R13:
418                 if (!state->regs || !state->full_regs) {
419                         orc_warn("missing regs for base reg R13 at ip %pB\n",
420                                  (void *)state->ip);
421                         goto err;
422                 }
423                 sp = state->regs->r13;
424                 break;
425 
426         case ORC_REG_DI:
427                 if (!state->regs || !state->full_regs) {
428                         orc_warn("missing regs for base reg DI at ip %pB\n",
429                                  (void *)state->ip);
430                         goto err;
431                 }
432                 sp = state->regs->di;
433                 break;
434 
435         case ORC_REG_DX:
436                 if (!state->regs || !state->full_regs) {
437                         orc_warn("missing regs for base reg DX at ip %pB\n",
438                                  (void *)state->ip);
439                         goto err;
440                 }
441                 sp = state->regs->dx;
442                 break;
443 
444         default:
445                 orc_warn("unknown SP base reg %d for ip %pB\n",
446                          orc->sp_reg, (void *)state->ip);
447                 goto err;
448         }
449 
450         if (indirect) {
451                 if (!deref_stack_reg(state, sp, &sp))
452                         goto err;
453         }
454 
455         /* Find IP, SP and possibly regs: */
456         switch (orc->type) {
457         case ORC_TYPE_CALL:
458                 ip_p = sp - sizeof(long);
459 
460                 if (!deref_stack_reg(state, ip_p, &state->ip))
461                         goto err;
462 
463                 state->ip = ftrace_graph_ret_addr(state->task, &state->graph_idx,
464                                                   state->ip, (void *)ip_p);
465 
466                 state->sp = sp;
467                 state->regs = NULL;
468                 state->signal = false;
469                 break;
470 
471         case ORC_TYPE_REGS:
472                 if (!deref_stack_regs(state, sp, &state->ip, &state->sp)) {
473                         orc_warn("can't dereference registers at %p for ip %pB\n",
474                                  (void *)sp, (void *)orig_ip);
475                         goto err;
476                 }
477 
478                 state->regs = (struct pt_regs *)sp;
479                 state->full_regs = true;
480                 state->signal = true;
481                 break;
482 
483         case ORC_TYPE_REGS_IRET:
484                 if (!deref_stack_iret_regs(state, sp, &state->ip, &state->sp)) {
485                         orc_warn("can't dereference iret registers at %p for ip %pB\n",
486                                  (void *)sp, (void *)orig_ip);
487                         goto err;
488                 }
489 
490                 state->regs = (void *)sp - IRET_FRAME_OFFSET;
491                 state->full_regs = false;
492                 state->signal = true;
493                 break;
494 
495         default:
496                 orc_warn("unknown .orc_unwind entry type %d for ip %pB\n",
497                          orc->type, (void *)orig_ip);
498                 break;
499         }
500 
501         /* Find BP: */
502         switch (orc->bp_reg) {
503         case ORC_REG_UNDEFINED:
504                 if (state->regs && state->full_regs)
505                         state->bp = state->regs->bp;
506                 break;
507 
508         case ORC_REG_PREV_SP:
509                 if (!deref_stack_reg(state, sp + orc->bp_offset, &state->bp))
510                         goto err;
511                 break;
512 
513         case ORC_REG_BP:
514                 if (!deref_stack_reg(state, state->bp + orc->bp_offset, &state->bp))
515                         goto err;
516                 break;
517 
518         default:
519                 orc_warn("unknown BP base reg %d for ip %pB\n",
520                          orc->bp_reg, (void *)orig_ip);
521                 goto err;
522         }
523 
524         /* Prevent a recursive loop due to bad ORC data: */
525         if (state->stack_info.type == prev_type &&
526             on_stack(&state->stack_info, (void *)state->sp, sizeof(long)) &&
527             state->sp <= prev_sp) {
528                 orc_warn("stack going in the wrong direction? ip=%pB\n",
529                          (void *)orig_ip);
530                 goto err;
531         }
532 
533         preempt_enable();
534         return true;
535 
536 err:
537         state->error = true;
538 
539 the_end:
540         preempt_enable();
541         state->stack_info.type = STACK_TYPE_UNKNOWN;
542         return false;
543 }
544 EXPORT_SYMBOL_GPL(unwind_next_frame);
545 
546 void __unwind_start(struct unwind_state *state, struct task_struct *task,
547                     struct pt_regs *regs, unsigned long *first_frame)
548 {
549         memset(state, 0, sizeof(*state));
550         state->task = task;
551 
552         /*
553          * Refuse to unwind the stack of a task while it's executing on another
554          * CPU.  This check is racy, but that's ok: the unwinder has other
555          * checks to prevent it from going off the rails.
556          */
557         if (task_on_another_cpu(task))
558                 goto done;
559 
560         if (regs) {
561                 if (user_mode(regs))
562                         goto done;
563 
564                 state->ip = regs->ip;
565                 state->sp = kernel_stack_pointer(regs);
566                 state->bp = regs->bp;
567                 state->regs = regs;
568                 state->full_regs = true;
569                 state->signal = true;
570 
571         } else if (task == current) {
572                 asm volatile("lea (%%rip), %0\n\t"
573                              "mov %%rsp, %1\n\t"
574                              "mov %%rbp, %2\n\t"
575                              : "=r" (state->ip), "=r" (state->sp),
576                                "=r" (state->bp));
577 
578         } else {
579                 struct inactive_task_frame *frame = (void *)task->thread.sp;
580 
581                 state->sp = task->thread.sp;
582                 state->bp = READ_ONCE_NOCHECK(frame->bp);
583                 state->ip = READ_ONCE_NOCHECK(frame->ret_addr);
584         }
585 
586         if (get_stack_info((unsigned long *)state->sp, state->task,
587                            &state->stack_info, &state->stack_mask)) {
588                 /*
589                  * We weren't on a valid stack.  It's possible that
590                  * we overflowed a valid stack into a guard page.
591                  * See if the next page up is valid so that we can
592                  * generate some kind of backtrace if this happens.
593                  */
594                 void *next_page = (void *)PAGE_ALIGN((unsigned long)state->sp);
595                 if (get_stack_info(next_page, state->task, &state->stack_info,
596                                    &state->stack_mask))
597                         return;
598         }
599 
600         /*
601          * The caller can provide the address of the first frame directly
602          * (first_frame) or indirectly (regs->sp) to indicate which stack frame
603          * to start unwinding at.  Skip ahead until we reach it.
604          */
605 
606         /* When starting from regs, skip the regs frame: */
607         if (regs) {
608                 unwind_next_frame(state);
609                 return;
610         }
611 
612         /* Otherwise, skip ahead to the user-specified starting frame: */
613         while (!unwind_done(state) &&
614                (!on_stack(&state->stack_info, first_frame, sizeof(long)) ||
615                         state->sp <= (unsigned long)first_frame))
616                 unwind_next_frame(state);
617 
618         return;
619 
620 done:
621         state->stack_info.type = STACK_TYPE_UNKNOWN;
622         return;
623 }
624 EXPORT_SYMBOL_GPL(__unwind_start);
625 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp