arch/x86/kernel/unwind_orc.c

   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 ? -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, 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         /* Have we reached the end? */
 367         if (state->regs && user_mode(state->regs))
 368                 goto done;
 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 || orc->sp_reg == ORC_REG_UNDEFINED)
 378                 goto done;
 379         orig_ip = state->ip;
 380
 381         /* Find the previous frame's stack: */
 382         switch (orc->sp_reg) {
 383         case ORC_REG_SP:
 384                 sp = state->sp + orc->sp_offset;
 385                 break;
 386
 387         case ORC_REG_BP:
 388                 sp = state->bp + orc->sp_offset;
 389                 break;
 390
 391         case ORC_REG_SP_INDIRECT:
 392                 sp = state->sp + orc->sp_offset;
 393                 indirect = true;
 394                 break;
 395
 396         case ORC_REG_BP_INDIRECT:
 397                 sp = state->bp + orc->sp_offset;
 398                 indirect = true;
 399                 break;
 400
 401         case ORC_REG_R10:
 402                 if (!state->regs || !state->full_regs) {
 403                         orc_warn("missing regs for base reg R10 at ip %pB\n",
 404                                  (void *)state->ip);
 405                         goto done;
 406                 }
 407                 sp = state->regs->r10;
 408                 break;
 409
 410         case ORC_REG_R13:
 411                 if (!state->regs || !state->full_regs) {
 412                         orc_warn("missing regs for base reg R13 at ip %pB\n",
 413                                  (void *)state->ip);
 414                         goto done;
 415                 }
 416                 sp = state->regs->r13;
 417                 break;
 418
 419         case ORC_REG_DI:
 420                 if (!state->regs || !state->full_regs) {
 421                         orc_warn("missing regs for base reg DI at ip %pB\n",
 422                                  (void *)state->ip);
 423                         goto done;
 424                 }
 425                 sp = state->regs->di;
 426                 break;
 427
 428         case ORC_REG_DX:
 429                 if (!state->regs || !state->full_regs) {
 430                         orc_warn("missing regs for base reg DX at ip %pB\n",
 431                                  (void *)state->ip);
 432                         goto done;
 433                 }
 434                 sp = state->regs->dx;
 435                 break;
 436
 437         default:
 438                 orc_warn("unknown SP base reg %d for ip %pB\n",
 439                          orc->sp_reg, (void *)state->ip);
 440                 goto done;
 441         }
 442
 443         if (indirect) {
 444                 if (!deref_stack_reg(state, sp, &sp))
 445                         goto done;
 446         }
 447
 448         /* Find IP, SP and possibly regs: */
 449         switch (orc->type) {
 450         case ORC_TYPE_CALL:
 451                 ip_p = sp - sizeof(long);
 452
 453                 if (!deref_stack_reg(state, ip_p, &state->ip))
 454                         goto done;
 455
 456                 state->ip = ftrace_graph_ret_addr(state->task, &state->graph_idx,
 457                                                   state->ip, (void *)ip_p);
 458
 459                 state->sp = sp;
 460                 state->regs = NULL;
 461                 state->signal = false;
 462                 break;
 463
 464         case ORC_TYPE_REGS:
 465                 if (!deref_stack_regs(state, sp, &state->ip, &state->sp)) {
 466                         orc_warn("can't dereference registers at %p for ip %pB\n",
 467                                  (void *)sp, (void *)orig_ip);
 468                         goto done;
 469                 }
 470
 471                 state->regs = (struct pt_regs *)sp;
 472                 state->full_regs = true;
 473                 state->signal = true;
 474                 break;
 475
 476         case ORC_TYPE_REGS_IRET:
 477                 if (!deref_stack_iret_regs(state, sp, &state->ip, &state->sp)) {
 478                         orc_warn("can't dereference iret registers at %p for ip %pB\n",
 479                                  (void *)sp, (void *)orig_ip);
 480                         goto done;
 481                 }
 482
 483                 state->regs = (void *)sp - IRET_FRAME_OFFSET;
 484                 state->full_regs = false;
 485                 state->signal = true;
 486                 break;
 487
 488         default:
 489                 orc_warn("unknown .orc_unwind entry type %d for ip %pB\n",
 490                          orc->type, (void *)orig_ip);
 491                 break;
 492         }
 493
 494         /* Find BP: */
 495         switch (orc->bp_reg) {
 496         case ORC_REG_UNDEFINED:
 497                 if (state->regs && state->full_regs)
 498                         state->bp = state->regs->bp;
 499                 break;
 500
 501         case ORC_REG_PREV_SP:
 502                 if (!deref_stack_reg(state, sp + orc->bp_offset, &state->bp))
 503                         goto done;
 504                 break;
 505
 506         case ORC_REG_BP:
 507                 if (!deref_stack_reg(state, state->bp + orc->bp_offset, &state->bp))
 508                         goto done;
 509                 break;
 510
 511         default:
 512                 orc_warn("unknown BP base reg %d for ip %pB\n",
 513                          orc->bp_reg, (void *)orig_ip);
 514                 goto done;
 515         }
 516
 517         /* Prevent a recursive loop due to bad ORC data: */
 518         if (state->stack_info.type == prev_type &&
 519             on_stack(&state->stack_info, (void *)state->sp, sizeof(long)) &&
 520             state->sp <= prev_sp) {
 521                 orc_warn("stack going in the wrong direction? ip=%pB\n",
 522                          (void *)orig_ip);
 523                 goto done;
 524         }
 525
 526         preempt_enable();
 527         return true;
 528
 529 done:
 530         preempt_enable();
 531         state->stack_info.type = STACK_TYPE_UNKNOWN;
 532         return false;
 533 }
 534 EXPORT_SYMBOL_GPL(unwind_next_frame);
 535
 536 void __unwind_start(struct unwind_state *state, struct task_struct *task,
 537                     struct pt_regs *regs, unsigned long *first_frame)
 538 {
 539         memset(state, 0, sizeof(*state));
 540         state->task = task;
 541
 542         /*
 543          * Refuse to unwind the stack of a task while it's executing on another
 544          * CPU.  This check is racy, but that's ok: the unwinder has other
 545          * checks to prevent it from going off the rails.
 546          */
 547         if (task_on_another_cpu(task))
 548                 goto done;
 549
 550         if (regs) {
 551                 if (user_mode(regs))
 552                         goto done;
 553
 554                 state->ip = regs->ip;
 555                 state->sp = kernel_stack_pointer(regs);
 556                 state->bp = regs->bp;
 557                 state->regs = regs;
 558                 state->full_regs = true;
 559                 state->signal = true;
 560
 561         } else if (task == current) {
 562                 asm volatile("lea (%%rip), %0\n\t"
 563                              "mov %%rsp, %1\n\t"
 564                              "mov %%rbp, %2\n\t"
 565                              : "=r" (state->ip), "=r" (state->sp),
 566                                "=r" (state->bp));
 567
 568         } else {
 569                 struct inactive_task_frame *frame = (void *)task->thread.sp;
 570
 571                 state->sp = task->thread.sp;
 572                 state->bp = READ_ONCE_NOCHECK(frame->bp);
 573                 state->ip = READ_ONCE_NOCHECK(frame->ret_addr);
 574         }
 575
 576         if (get_stack_info((unsigned long *)state->sp, state->task,
 577                            &state->stack_info, &state->stack_mask)) {
 578                 /*
 579                  * We weren't on a valid stack.  It's possible that
 580                  * we overflowed a valid stack into a guard page.
 581                  * See if the next page up is valid so that we can
 582                  * generate some kind of backtrace if this happens.
 583                  */
 584                 void *next_page = (void *)PAGE_ALIGN((unsigned long)state->sp);
 585                 if (get_stack_info(next_page, state->task, &state->stack_info,
 586                                    &state->stack_mask))
 587                         return;
 588         }
 589
 590         /*
 591          * The caller can provide the address of the first frame directly
 592          * (first_frame) or indirectly (regs->sp) to indicate which stack frame
 593          * to start unwinding at.  Skip ahead until we reach it.
 594          */
 595
 596         /* When starting from regs, skip the regs frame: */
 597         if (regs) {
 598                 unwind_next_frame(state);
 599                 return;
 600         }
 601
 602         /* Otherwise, skip ahead to the user-specified starting frame: */
 603         while (!unwind_done(state) &&
 604                (!on_stack(&state->stack_info, first_frame, sizeof(long)) ||
 605                         state->sp <= (unsigned long)first_frame))
 606                 unwind_next_frame(state);
 607
 608         return;
 609
 610 done:
 611         state->stack_info.type = STACK_TYPE_UNKNOWN;
 612         return;
 613 }
 614 EXPORT_SYMBOL_GPL(__unwind_start);