Loading...
Note: File does not exist in v3.5.6.
1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/module.h>
3#include <linux/sort.h>
4#include <asm/ptrace.h>
5#include <asm/stacktrace.h>
6#include <asm/unwind.h>
7#include <asm/orc_types.h>
8#include <asm/orc_lookup.h>
9
10#define orc_warn(fmt, ...) \
11 printk_deferred_once(KERN_WARNING "WARNING: " fmt, ##__VA_ARGS__)
12
13#define orc_warn_current(args...) \
14({ \
15 if (state->task == current) \
16 orc_warn(args); \
17})
18
19extern int __start_orc_unwind_ip[];
20extern int __stop_orc_unwind_ip[];
21extern struct orc_entry __start_orc_unwind[];
22extern struct orc_entry __stop_orc_unwind[];
23
24static bool orc_init __ro_after_init;
25static unsigned int lookup_num_blocks __ro_after_init;
26
27static inline unsigned long orc_ip(const int *ip)
28{
29 return (unsigned long)ip + *ip;
30}
31
32static struct orc_entry *__orc_find(int *ip_table, struct orc_entry *u_table,
33 unsigned int num_entries, unsigned long ip)
34{
35 int *first = ip_table;
36 int *last = ip_table + num_entries - 1;
37 int *mid = first, *found = first;
38
39 if (!num_entries)
40 return NULL;
41
42 /*
43 * Do a binary range search to find the rightmost duplicate of a given
44 * starting address. Some entries are section terminators which are
45 * "weak" entries for ensuring there are no gaps. They should be
46 * ignored when they conflict with a real entry.
47 */
48 while (first <= last) {
49 mid = first + ((last - first) / 2);
50
51 if (orc_ip(mid) <= ip) {
52 found = mid;
53 first = mid + 1;
54 } else
55 last = mid - 1;
56 }
57
58 return u_table + (found - ip_table);
59}
60
61#ifdef CONFIG_MODULES
62static struct orc_entry *orc_module_find(unsigned long ip)
63{
64 struct module *mod;
65
66 mod = __module_address(ip);
67 if (!mod || !mod->arch.orc_unwind || !mod->arch.orc_unwind_ip)
68 return NULL;
69 return __orc_find(mod->arch.orc_unwind_ip, mod->arch.orc_unwind,
70 mod->arch.num_orcs, ip);
71}
72#else
73static struct orc_entry *orc_module_find(unsigned long ip)
74{
75 return NULL;
76}
77#endif
78
79#ifdef CONFIG_DYNAMIC_FTRACE
80static struct orc_entry *orc_find(unsigned long ip);
81
82/*
83 * Ftrace dynamic trampolines do not have orc entries of their own.
84 * But they are copies of the ftrace entries that are static and
85 * defined in ftrace_*.S, which do have orc entries.
86 *
87 * If the unwinder comes across a ftrace trampoline, then find the
88 * ftrace function that was used to create it, and use that ftrace
89 * function's orc entry, as the placement of the return code in
90 * the stack will be identical.
91 */
92static struct orc_entry *orc_ftrace_find(unsigned long ip)
93{
94 struct ftrace_ops *ops;
95 unsigned long caller;
96
97 ops = ftrace_ops_trampoline(ip);
98 if (!ops)
99 return NULL;
100
101 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
102 caller = (unsigned long)ftrace_regs_call;
103 else
104 caller = (unsigned long)ftrace_call;
105
106 /* Prevent unlikely recursion */
107 if (ip == caller)
108 return NULL;
109
110 return orc_find(caller);
111}
112#else
113static struct orc_entry *orc_ftrace_find(unsigned long ip)
114{
115 return NULL;
116}
117#endif
118
119/*
120 * If we crash with IP==0, the last successfully executed instruction
121 * was probably an indirect function call with a NULL function pointer,
122 * and we don't have unwind information for NULL.
123 * This hardcoded ORC entry for IP==0 allows us to unwind from a NULL function
124 * pointer into its parent and then continue normally from there.
125 */
126static struct orc_entry null_orc_entry = {
127 .sp_offset = sizeof(long),
128 .sp_reg = ORC_REG_SP,
129 .bp_reg = ORC_REG_UNDEFINED,
130 .type = ORC_TYPE_CALL
131};
132
133/* Fake frame pointer entry -- used as a fallback for generated code */
134static struct orc_entry orc_fp_entry = {
135 .type = ORC_TYPE_CALL,
136 .sp_reg = ORC_REG_BP,
137 .sp_offset = 16,
138 .bp_reg = ORC_REG_PREV_SP,
139 .bp_offset = -16,
140 .end = 0,
141};
142
143static struct orc_entry *orc_find(unsigned long ip)
144{
145 static struct orc_entry *orc;
146
147 if (ip == 0)
148 return &null_orc_entry;
149
150 /* For non-init vmlinux addresses, use the fast lookup table: */
151 if (ip >= LOOKUP_START_IP && ip < LOOKUP_STOP_IP) {
152 unsigned int idx, start, stop;
153
154 idx = (ip - LOOKUP_START_IP) / LOOKUP_BLOCK_SIZE;
155
156 if (unlikely((idx >= lookup_num_blocks-1))) {
157 orc_warn("WARNING: bad lookup idx: idx=%u num=%u ip=%pB\n",
158 idx, lookup_num_blocks, (void *)ip);
159 return NULL;
160 }
161
162 start = orc_lookup[idx];
163 stop = orc_lookup[idx + 1] + 1;
164
165 if (unlikely((__start_orc_unwind + start >= __stop_orc_unwind) ||
166 (__start_orc_unwind + stop > __stop_orc_unwind))) {
167 orc_warn("WARNING: bad lookup value: idx=%u num=%u start=%u stop=%u ip=%pB\n",
168 idx, lookup_num_blocks, start, stop, (void *)ip);
169 return NULL;
170 }
171
172 return __orc_find(__start_orc_unwind_ip + start,
173 __start_orc_unwind + start, stop - start, ip);
174 }
175
176 /* vmlinux .init slow lookup: */
177 if (init_kernel_text(ip))
178 return __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
179 __stop_orc_unwind_ip - __start_orc_unwind_ip, ip);
180
181 /* Module lookup: */
182 orc = orc_module_find(ip);
183 if (orc)
184 return orc;
185
186 return orc_ftrace_find(ip);
187}
188
189#ifdef CONFIG_MODULES
190
191static DEFINE_MUTEX(sort_mutex);
192static int *cur_orc_ip_table = __start_orc_unwind_ip;
193static struct orc_entry *cur_orc_table = __start_orc_unwind;
194
195static void orc_sort_swap(void *_a, void *_b, int size)
196{
197 struct orc_entry *orc_a, *orc_b;
198 struct orc_entry orc_tmp;
199 int *a = _a, *b = _b, tmp;
200 int delta = _b - _a;
201
202 /* Swap the .orc_unwind_ip entries: */
203 tmp = *a;
204 *a = *b + delta;
205 *b = tmp - delta;
206
207 /* Swap the corresponding .orc_unwind entries: */
208 orc_a = cur_orc_table + (a - cur_orc_ip_table);
209 orc_b = cur_orc_table + (b - cur_orc_ip_table);
210 orc_tmp = *orc_a;
211 *orc_a = *orc_b;
212 *orc_b = orc_tmp;
213}
214
215static int orc_sort_cmp(const void *_a, const void *_b)
216{
217 struct orc_entry *orc_a;
218 const int *a = _a, *b = _b;
219 unsigned long a_val = orc_ip(a);
220 unsigned long b_val = orc_ip(b);
221
222 if (a_val > b_val)
223 return 1;
224 if (a_val < b_val)
225 return -1;
226
227 /*
228 * The "weak" section terminator entries need to always be on the left
229 * to ensure the lookup code skips them in favor of real entries.
230 * These terminator entries exist to handle any gaps created by
231 * whitelisted .o files which didn't get objtool generation.
232 */
233 orc_a = cur_orc_table + (a - cur_orc_ip_table);
234 return orc_a->sp_reg == ORC_REG_UNDEFINED && !orc_a->end ? -1 : 1;
235}
236
237void unwind_module_init(struct module *mod, void *_orc_ip, size_t orc_ip_size,
238 void *_orc, size_t orc_size)
239{
240 int *orc_ip = _orc_ip;
241 struct orc_entry *orc = _orc;
242 unsigned int num_entries = orc_ip_size / sizeof(int);
243
244 WARN_ON_ONCE(orc_ip_size % sizeof(int) != 0 ||
245 orc_size % sizeof(*orc) != 0 ||
246 num_entries != orc_size / sizeof(*orc));
247
248 /*
249 * The 'cur_orc_*' globals allow the orc_sort_swap() callback to
250 * associate an .orc_unwind_ip table entry with its corresponding
251 * .orc_unwind entry so they can both be swapped.
252 */
253 mutex_lock(&sort_mutex);
254 cur_orc_ip_table = orc_ip;
255 cur_orc_table = orc;
256 sort(orc_ip, num_entries, sizeof(int), orc_sort_cmp, orc_sort_swap);
257 mutex_unlock(&sort_mutex);
258
259 mod->arch.orc_unwind_ip = orc_ip;
260 mod->arch.orc_unwind = orc;
261 mod->arch.num_orcs = num_entries;
262}
263#endif
264
265void __init unwind_init(void)
266{
267 size_t orc_ip_size = (void *)__stop_orc_unwind_ip - (void *)__start_orc_unwind_ip;
268 size_t orc_size = (void *)__stop_orc_unwind - (void *)__start_orc_unwind;
269 size_t num_entries = orc_ip_size / sizeof(int);
270 struct orc_entry *orc;
271 int i;
272
273 if (!num_entries || orc_ip_size % sizeof(int) != 0 ||
274 orc_size % sizeof(struct orc_entry) != 0 ||
275 num_entries != orc_size / sizeof(struct orc_entry)) {
276 orc_warn("WARNING: Bad or missing .orc_unwind table. Disabling unwinder.\n");
277 return;
278 }
279
280 /*
281 * Note, the orc_unwind and orc_unwind_ip tables were already
282 * sorted at build time via the 'sorttable' tool.
283 * It's ready for binary search straight away, no need to sort it.
284 */
285
286 /* Initialize the fast lookup table: */
287 lookup_num_blocks = orc_lookup_end - orc_lookup;
288 for (i = 0; i < lookup_num_blocks-1; i++) {
289 orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
290 num_entries,
291 LOOKUP_START_IP + (LOOKUP_BLOCK_SIZE * i));
292 if (!orc) {
293 orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
294 return;
295 }
296
297 orc_lookup[i] = orc - __start_orc_unwind;
298 }
299
300 /* Initialize the ending block: */
301 orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind, num_entries,
302 LOOKUP_STOP_IP);
303 if (!orc) {
304 orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
305 return;
306 }
307 orc_lookup[lookup_num_blocks-1] = orc - __start_orc_unwind;
308
309 orc_init = true;
310}
311
312unsigned long unwind_get_return_address(struct unwind_state *state)
313{
314 if (unwind_done(state))
315 return 0;
316
317 return __kernel_text_address(state->ip) ? state->ip : 0;
318}
319EXPORT_SYMBOL_GPL(unwind_get_return_address);
320
321unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
322{
323 struct task_struct *task = state->task;
324
325 if (unwind_done(state))
326 return NULL;
327
328 if (state->regs)
329 return &state->regs->ip;
330
331 if (task != current && state->sp == task->thread.sp) {
332 struct inactive_task_frame *frame = (void *)task->thread.sp;
333 return &frame->ret_addr;
334 }
335
336 if (state->sp)
337 return (unsigned long *)state->sp - 1;
338
339 return NULL;
340}
341
342static bool stack_access_ok(struct unwind_state *state, unsigned long _addr,
343 size_t len)
344{
345 struct stack_info *info = &state->stack_info;
346 void *addr = (void *)_addr;
347
348 if (!on_stack(info, addr, len) &&
349 (get_stack_info(addr, state->task, info, &state->stack_mask)))
350 return false;
351
352 return true;
353}
354
355static bool deref_stack_reg(struct unwind_state *state, unsigned long addr,
356 unsigned long *val)
357{
358 if (!stack_access_ok(state, addr, sizeof(long)))
359 return false;
360
361 *val = READ_ONCE_NOCHECK(*(unsigned long *)addr);
362 return true;
363}
364
365static bool deref_stack_regs(struct unwind_state *state, unsigned long addr,
366 unsigned long *ip, unsigned long *sp)
367{
368 struct pt_regs *regs = (struct pt_regs *)addr;
369
370 /* x86-32 support will be more complicated due to the ®s->sp hack */
371 BUILD_BUG_ON(IS_ENABLED(CONFIG_X86_32));
372
373 if (!stack_access_ok(state, addr, sizeof(struct pt_regs)))
374 return false;
375
376 *ip = regs->ip;
377 *sp = regs->sp;
378 return true;
379}
380
381static bool deref_stack_iret_regs(struct unwind_state *state, unsigned long addr,
382 unsigned long *ip, unsigned long *sp)
383{
384 struct pt_regs *regs = (void *)addr - IRET_FRAME_OFFSET;
385
386 if (!stack_access_ok(state, addr, IRET_FRAME_SIZE))
387 return false;
388
389 *ip = regs->ip;
390 *sp = regs->sp;
391 return true;
392}
393
394/*
395 * If state->regs is non-NULL, and points to a full pt_regs, just get the reg
396 * value from state->regs.
397 *
398 * Otherwise, if state->regs just points to IRET regs, and the previous frame
399 * had full regs, it's safe to get the value from the previous regs. This can
400 * happen when early/late IRQ entry code gets interrupted by an NMI.
401 */
402static bool get_reg(struct unwind_state *state, unsigned int reg_off,
403 unsigned long *val)
404{
405 unsigned int reg = reg_off/8;
406
407 if (!state->regs)
408 return false;
409
410 if (state->full_regs) {
411 *val = ((unsigned long *)state->regs)[reg];
412 return true;
413 }
414
415 if (state->prev_regs) {
416 *val = ((unsigned long *)state->prev_regs)[reg];
417 return true;
418 }
419
420 return false;
421}
422
423bool unwind_next_frame(struct unwind_state *state)
424{
425 unsigned long ip_p, sp, tmp, orig_ip = state->ip, prev_sp = state->sp;
426 enum stack_type prev_type = state->stack_info.type;
427 struct orc_entry *orc;
428 bool indirect = false;
429
430 if (unwind_done(state))
431 return false;
432
433 /* Don't let modules unload while we're reading their ORC data. */
434 preempt_disable();
435
436 /* End-of-stack check for user tasks: */
437 if (state->regs && user_mode(state->regs))
438 goto the_end;
439
440 /*
441 * Find the orc_entry associated with the text address.
442 *
443 * For a call frame (as opposed to a signal frame), state->ip points to
444 * the instruction after the call. That instruction's stack layout
445 * could be different from the call instruction's layout, for example
446 * if the call was to a noreturn function. So get the ORC data for the
447 * call instruction itself.
448 */
449 orc = orc_find(state->signal ? state->ip : state->ip - 1);
450 if (!orc) {
451 /*
452 * As a fallback, try to assume this code uses a frame pointer.
453 * This is useful for generated code, like BPF, which ORC
454 * doesn't know about. This is just a guess, so the rest of
455 * the unwind is no longer considered reliable.
456 */
457 orc = &orc_fp_entry;
458 state->error = true;
459 }
460
461 /* End-of-stack check for kernel threads: */
462 if (orc->sp_reg == ORC_REG_UNDEFINED) {
463 if (!orc->end)
464 goto err;
465
466 goto the_end;
467 }
468
469 /* Find the previous frame's stack: */
470 switch (orc->sp_reg) {
471 case ORC_REG_SP:
472 sp = state->sp + orc->sp_offset;
473 break;
474
475 case ORC_REG_BP:
476 sp = state->bp + orc->sp_offset;
477 break;
478
479 case ORC_REG_SP_INDIRECT:
480 sp = state->sp + orc->sp_offset;
481 indirect = true;
482 break;
483
484 case ORC_REG_BP_INDIRECT:
485 sp = state->bp + orc->sp_offset;
486 indirect = true;
487 break;
488
489 case ORC_REG_R10:
490 if (!get_reg(state, offsetof(struct pt_regs, r10), &sp)) {
491 orc_warn_current("missing R10 value at %pB\n",
492 (void *)state->ip);
493 goto err;
494 }
495 break;
496
497 case ORC_REG_R13:
498 if (!get_reg(state, offsetof(struct pt_regs, r13), &sp)) {
499 orc_warn_current("missing R13 value at %pB\n",
500 (void *)state->ip);
501 goto err;
502 }
503 break;
504
505 case ORC_REG_DI:
506 if (!get_reg(state, offsetof(struct pt_regs, di), &sp)) {
507 orc_warn_current("missing RDI value at %pB\n",
508 (void *)state->ip);
509 goto err;
510 }
511 break;
512
513 case ORC_REG_DX:
514 if (!get_reg(state, offsetof(struct pt_regs, dx), &sp)) {
515 orc_warn_current("missing DX value at %pB\n",
516 (void *)state->ip);
517 goto err;
518 }
519 break;
520
521 default:
522 orc_warn("unknown SP base reg %d at %pB\n",
523 orc->sp_reg, (void *)state->ip);
524 goto err;
525 }
526
527 if (indirect) {
528 if (!deref_stack_reg(state, sp, &sp))
529 goto err;
530 }
531
532 /* Find IP, SP and possibly regs: */
533 switch (orc->type) {
534 case ORC_TYPE_CALL:
535 ip_p = sp - sizeof(long);
536
537 if (!deref_stack_reg(state, ip_p, &state->ip))
538 goto err;
539
540 state->ip = ftrace_graph_ret_addr(state->task, &state->graph_idx,
541 state->ip, (void *)ip_p);
542
543 state->sp = sp;
544 state->regs = NULL;
545 state->prev_regs = NULL;
546 state->signal = false;
547 break;
548
549 case ORC_TYPE_REGS:
550 if (!deref_stack_regs(state, sp, &state->ip, &state->sp)) {
551 orc_warn_current("can't access registers at %pB\n",
552 (void *)orig_ip);
553 goto err;
554 }
555
556 state->regs = (struct pt_regs *)sp;
557 state->prev_regs = NULL;
558 state->full_regs = true;
559 state->signal = true;
560 break;
561
562 case ORC_TYPE_REGS_IRET:
563 if (!deref_stack_iret_regs(state, sp, &state->ip, &state->sp)) {
564 orc_warn_current("can't access iret registers at %pB\n",
565 (void *)orig_ip);
566 goto err;
567 }
568
569 if (state->full_regs)
570 state->prev_regs = state->regs;
571 state->regs = (void *)sp - IRET_FRAME_OFFSET;
572 state->full_regs = false;
573 state->signal = true;
574 break;
575
576 default:
577 orc_warn("unknown .orc_unwind entry type %d at %pB\n",
578 orc->type, (void *)orig_ip);
579 goto err;
580 }
581
582 /* Find BP: */
583 switch (orc->bp_reg) {
584 case ORC_REG_UNDEFINED:
585 if (get_reg(state, offsetof(struct pt_regs, bp), &tmp))
586 state->bp = tmp;
587 break;
588
589 case ORC_REG_PREV_SP:
590 if (!deref_stack_reg(state, sp + orc->bp_offset, &state->bp))
591 goto err;
592 break;
593
594 case ORC_REG_BP:
595 if (!deref_stack_reg(state, state->bp + orc->bp_offset, &state->bp))
596 goto err;
597 break;
598
599 default:
600 orc_warn("unknown BP base reg %d for ip %pB\n",
601 orc->bp_reg, (void *)orig_ip);
602 goto err;
603 }
604
605 /* Prevent a recursive loop due to bad ORC data: */
606 if (state->stack_info.type == prev_type &&
607 on_stack(&state->stack_info, (void *)state->sp, sizeof(long)) &&
608 state->sp <= prev_sp) {
609 orc_warn_current("stack going in the wrong direction? at %pB\n",
610 (void *)orig_ip);
611 goto err;
612 }
613
614 preempt_enable();
615 return true;
616
617err:
618 state->error = true;
619
620the_end:
621 preempt_enable();
622 state->stack_info.type = STACK_TYPE_UNKNOWN;
623 return false;
624}
625EXPORT_SYMBOL_GPL(unwind_next_frame);
626
627void __unwind_start(struct unwind_state *state, struct task_struct *task,
628 struct pt_regs *regs, unsigned long *first_frame)
629{
630 memset(state, 0, sizeof(*state));
631 state->task = task;
632
633 if (!orc_init)
634 goto err;
635
636 /*
637 * Refuse to unwind the stack of a task while it's executing on another
638 * CPU. This check is racy, but that's ok: the unwinder has other
639 * checks to prevent it from going off the rails.
640 */
641 if (task_on_another_cpu(task))
642 goto err;
643
644 if (regs) {
645 if (user_mode(regs))
646 goto the_end;
647
648 state->ip = regs->ip;
649 state->sp = regs->sp;
650 state->bp = regs->bp;
651 state->regs = regs;
652 state->full_regs = true;
653 state->signal = true;
654
655 } else if (task == current) {
656 asm volatile("lea (%%rip), %0\n\t"
657 "mov %%rsp, %1\n\t"
658 "mov %%rbp, %2\n\t"
659 : "=r" (state->ip), "=r" (state->sp),
660 "=r" (state->bp));
661
662 } else {
663 struct inactive_task_frame *frame = (void *)task->thread.sp;
664
665 state->sp = task->thread.sp;
666 state->bp = READ_ONCE_NOCHECK(frame->bp);
667 state->ip = READ_ONCE_NOCHECK(frame->ret_addr);
668 state->signal = (void *)state->ip == ret_from_fork;
669 }
670
671 if (get_stack_info((unsigned long *)state->sp, state->task,
672 &state->stack_info, &state->stack_mask)) {
673 /*
674 * We weren't on a valid stack. It's possible that
675 * we overflowed a valid stack into a guard page.
676 * See if the next page up is valid so that we can
677 * generate some kind of backtrace if this happens.
678 */
679 void *next_page = (void *)PAGE_ALIGN((unsigned long)state->sp);
680 state->error = true;
681 if (get_stack_info(next_page, state->task, &state->stack_info,
682 &state->stack_mask))
683 return;
684 }
685
686 /*
687 * The caller can provide the address of the first frame directly
688 * (first_frame) or indirectly (regs->sp) to indicate which stack frame
689 * to start unwinding at. Skip ahead until we reach it.
690 */
691
692 /* When starting from regs, skip the regs frame: */
693 if (regs) {
694 unwind_next_frame(state);
695 return;
696 }
697
698 /* Otherwise, skip ahead to the user-specified starting frame: */
699 while (!unwind_done(state) &&
700 (!on_stack(&state->stack_info, first_frame, sizeof(long)) ||
701 state->sp < (unsigned long)first_frame))
702 unwind_next_frame(state);
703
704 return;
705
706err:
707 state->error = true;
708the_end:
709 state->stack_info.type = STACK_TYPE_UNKNOWN;
710}
711EXPORT_SYMBOL_GPL(__unwind_start);