linux/arch/x86/kernel/unwind_orc.c
<<
>>
Prefs
   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#include <asm/sections.h>
   9
  10#define orc_warn(fmt, ...) \
  11        printk_deferred_once(KERN_WARNING pr_fmt("WARNING: " fmt), ##__VA_ARGS__)
  12
  13extern int __start_orc_unwind_ip[];
  14extern int __stop_orc_unwind_ip[];
  15extern struct orc_entry __start_orc_unwind[];
  16extern struct orc_entry __stop_orc_unwind[];
  17
  18static DEFINE_MUTEX(sort_mutex);
  19int *cur_orc_ip_table = __start_orc_unwind_ip;
  20struct orc_entry *cur_orc_table = __start_orc_unwind;
  21
  22unsigned int lookup_num_blocks;
  23bool orc_init;
  24
  25static inline unsigned long orc_ip(const int *ip)
  26{
  27        return (unsigned long)ip + *ip;
  28}
  29
  30static struct orc_entry *__orc_find(int *ip_table, struct orc_entry *u_table,
  31                                    unsigned int num_entries, unsigned long ip)
  32{
  33        int *first = ip_table;
  34        int *last = ip_table + num_entries - 1;
  35        int *mid = first, *found = first;
  36
  37        if (!num_entries)
  38                return NULL;
  39
  40        /*
  41         * Do a binary range search to find the rightmost duplicate of a given
  42         * starting address.  Some entries are section terminators which are
  43         * "weak" entries for ensuring there are no gaps.  They should be
  44         * ignored when they conflict with a real entry.
  45         */
  46        while (first <= last) {
  47                mid = first + ((last - first) / 2);
  48
  49                if (orc_ip(mid) <= ip) {
  50                        found = mid;
  51                        first = mid + 1;
  52                } else
  53                        last = mid - 1;
  54        }
  55
  56        return u_table + (found - ip_table);
  57}
  58
  59#ifdef CONFIG_MODULES
  60static struct orc_entry *orc_module_find(unsigned long ip)
  61{
  62        struct module *mod;
  63
  64        mod = __module_address(ip);
  65        if (!mod || !mod->arch.orc_unwind || !mod->arch.orc_unwind_ip)
  66                return NULL;
  67        return __orc_find(mod->arch.orc_unwind_ip, mod->arch.orc_unwind,
  68                          mod->arch.num_orcs, ip);
  69}
  70#else
  71static struct orc_entry *orc_module_find(unsigned long ip)
  72{
  73        return NULL;
  74}
  75#endif
  76
  77static struct orc_entry *orc_find(unsigned long ip)
  78{
  79        if (!orc_init)
  80                return NULL;
  81
  82        /* For non-init vmlinux addresses, use the fast lookup table: */
  83        if (ip >= LOOKUP_START_IP && ip < LOOKUP_STOP_IP) {
  84                unsigned int idx, start, stop;
  85
  86                idx = (ip - LOOKUP_START_IP) / LOOKUP_BLOCK_SIZE;
  87
  88                if (unlikely((idx >= lookup_num_blocks-1))) {
  89                        orc_warn("WARNING: bad lookup idx: idx=%u num=%u ip=%pB\n",
  90                                 idx, lookup_num_blocks, (void *)ip);
  91                        return NULL;
  92                }
  93
  94                start = orc_lookup[idx];
  95                stop = orc_lookup[idx + 1] + 1;
  96
  97                if (unlikely((__start_orc_unwind + start >= __stop_orc_unwind) ||
  98                             (__start_orc_unwind + stop > __stop_orc_unwind))) {
  99                        orc_warn("WARNING: bad lookup value: idx=%u num=%u start=%u stop=%u ip=%pB\n",
 100                                 idx, lookup_num_blocks, start, stop, (void *)ip);
 101                        return NULL;
 102                }
 103
 104                return __orc_find(__start_orc_unwind_ip + start,
 105                                  __start_orc_unwind + start, stop - start, ip);
 106        }
 107
 108        /* vmlinux .init slow lookup: */
 109        if (ip >= (unsigned long)_sinittext && ip < (unsigned long)_einittext)
 110                return __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
 111                                  __stop_orc_unwind_ip - __start_orc_unwind_ip, ip);
 112
 113        /* Module lookup: */
 114        return orc_module_find(ip);
 115}
 116
 117static void orc_sort_swap(void *_a, void *_b, int size)
 118{
 119        struct orc_entry *orc_a, *orc_b;
 120        struct orc_entry orc_tmp;
 121        int *a = _a, *b = _b, tmp;
 122        int delta = _b - _a;
 123
 124        /* Swap the .orc_unwind_ip entries: */
 125        tmp = *a;
 126        *a = *b + delta;
 127        *b = tmp - delta;
 128
 129        /* Swap the corresponding .orc_unwind entries: */
 130        orc_a = cur_orc_table + (a - cur_orc_ip_table);
 131        orc_b = cur_orc_table + (b - cur_orc_ip_table);
 132        orc_tmp = *orc_a;
 133        *orc_a = *orc_b;
 134        *orc_b = orc_tmp;
 135}
 136
 137static int orc_sort_cmp(const void *_a, const void *_b)
 138{
 139        struct orc_entry *orc_a;
 140        const int *a = _a, *b = _b;
 141        unsigned long a_val = orc_ip(a);
 142        unsigned long b_val = orc_ip(b);
 143
 144        if (a_val > b_val)
 145                return 1;
 146        if (a_val < b_val)
 147                return -1;
 148
 149        /*
 150         * The "weak" section terminator entries need to always be on the left
 151         * to ensure the lookup code skips them in favor of real entries.
 152         * These terminator entries exist to handle any gaps created by
 153         * whitelisted .o files which didn't get objtool generation.
 154         */
 155        orc_a = cur_orc_table + (a - cur_orc_ip_table);
 156        return orc_a->sp_reg == ORC_REG_UNDEFINED ? -1 : 1;
 157}
 158
 159#ifdef CONFIG_MODULES
 160void unwind_module_init(struct module *mod, void *_orc_ip, size_t orc_ip_size,
 161                        void *_orc, size_t orc_size)
 162{
 163        int *orc_ip = _orc_ip;
 164        struct orc_entry *orc = _orc;
 165        unsigned int num_entries = orc_ip_size / sizeof(int);
 166
 167        WARN_ON_ONCE(orc_ip_size % sizeof(int) != 0 ||
 168                     orc_size % sizeof(*orc) != 0 ||
 169                     num_entries != orc_size / sizeof(*orc));
 170
 171        /*
 172         * The 'cur_orc_*' globals allow the orc_sort_swap() callback to
 173         * associate an .orc_unwind_ip table entry with its corresponding
 174         * .orc_unwind entry so they can both be swapped.
 175         */
 176        mutex_lock(&sort_mutex);
 177        cur_orc_ip_table = orc_ip;
 178        cur_orc_table = orc;
 179        sort(orc_ip, num_entries, sizeof(int), orc_sort_cmp, orc_sort_swap);
 180        mutex_unlock(&sort_mutex);
 181
 182        mod->arch.orc_unwind_ip = orc_ip;
 183        mod->arch.orc_unwind = orc;
 184        mod->arch.num_orcs = num_entries;
 185}
 186#endif
 187
 188void __init unwind_init(void)
 189{
 190        size_t orc_ip_size = (void *)__stop_orc_unwind_ip - (void *)__start_orc_unwind_ip;
 191        size_t orc_size = (void *)__stop_orc_unwind - (void *)__start_orc_unwind;
 192        size_t num_entries = orc_ip_size / sizeof(int);
 193        struct orc_entry *orc;
 194        int i;
 195
 196        if (!num_entries || orc_ip_size % sizeof(int) != 0 ||
 197            orc_size % sizeof(struct orc_entry) != 0 ||
 198            num_entries != orc_size / sizeof(struct orc_entry)) {
 199                orc_warn("WARNING: Bad or missing .orc_unwind table.  Disabling unwinder.\n");
 200                return;
 201        }
 202
 203        /* Sort the .orc_unwind and .orc_unwind_ip tables: */
 204        sort(__start_orc_unwind_ip, num_entries, sizeof(int), orc_sort_cmp,
 205             orc_sort_swap);
 206
 207        /* Initialize the fast lookup table: */
 208        lookup_num_blocks = orc_lookup_end - orc_lookup;
 209        for (i = 0; i < lookup_num_blocks-1; i++) {
 210                orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
 211                                 num_entries,
 212                                 LOOKUP_START_IP + (LOOKUP_BLOCK_SIZE * i));
 213                if (!orc) {
 214                        orc_warn("WARNING: Corrupt .orc_unwind table.  Disabling unwinder.\n");
 215                        return;
 216                }
 217
 218                orc_lookup[i] = orc - __start_orc_unwind;
 219        }
 220
 221        /* Initialize the ending block: */
 222        orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind, num_entries,
 223                         LOOKUP_STOP_IP);
 224        if (!orc) {
 225                orc_warn("WARNING: Corrupt .orc_unwind table.  Disabling unwinder.\n");
 226                return;
 227        }
 228        orc_lookup[lookup_num_blocks-1] = orc - __start_orc_unwind;
 229
 230        orc_init = true;
 231}
 232
 233unsigned long unwind_get_return_address(struct unwind_state *state)
 234{
 235        if (unwind_done(state))
 236                return 0;
 237
 238        return __kernel_text_address(state->ip) ? state->ip : 0;
 239}
 240EXPORT_SYMBOL_GPL(unwind_get_return_address);
 241
 242unsigned long *unwind_get_return_address_ptr(struct unwind_state *state)
 243{
 244        if (unwind_done(state))
 245                return NULL;
 246
 247        if (state->regs)
 248                return &state->regs->ip;
 249
 250        if (state->sp)
 251                return (unsigned long *)state->sp - 1;
 252
 253        return NULL;
 254}
 255
 256static bool stack_access_ok(struct unwind_state *state, unsigned long addr,
 257                            size_t len)
 258{
 259        struct stack_info *info = &state->stack_info;
 260
 261        /*
 262         * If the address isn't on the current stack, switch to the next one.
 263         *
 264         * We may have to traverse multiple stacks to deal with the possibility
 265         * that info->next_sp could point to an empty stack and the address
 266         * could be on a subsequent stack.
 267         */
 268        while (!on_stack(info, (void *)addr, len))
 269                if (get_stack_info(info->next_sp, state->task, info,
 270                                   &state->stack_mask))
 271                        return false;
 272
 273        return true;
 274}
 275
 276static bool deref_stack_reg(struct unwind_state *state, unsigned long addr,
 277                            unsigned long *val)
 278{
 279        if (!stack_access_ok(state, addr, sizeof(long)))
 280                return false;
 281
 282        *val = READ_ONCE_NOCHECK(*(unsigned long *)addr);
 283        return true;
 284}
 285
 286#define REGS_SIZE (sizeof(struct pt_regs))
 287#define SP_OFFSET (offsetof(struct pt_regs, sp))
 288#define IRET_REGS_SIZE (REGS_SIZE - offsetof(struct pt_regs, ip))
 289#define IRET_SP_OFFSET (SP_OFFSET - offsetof(struct pt_regs, ip))
 290
 291static bool deref_stack_regs(struct unwind_state *state, unsigned long addr,
 292                             unsigned long *ip, unsigned long *sp, bool full)
 293{
 294        size_t regs_size = full ? REGS_SIZE : IRET_REGS_SIZE;
 295        size_t sp_offset = full ? SP_OFFSET : IRET_SP_OFFSET;
 296        struct pt_regs *regs = (struct pt_regs *)(addr + regs_size - REGS_SIZE);
 297
 298        if (IS_ENABLED(CONFIG_X86_64)) {
 299                if (!stack_access_ok(state, addr, regs_size))
 300                        return false;
 301
 302                *ip = regs->ip;
 303                *sp = regs->sp;
 304
 305                return true;
 306        }
 307
 308        if (!stack_access_ok(state, addr, sp_offset))
 309                return false;
 310
 311        *ip = regs->ip;
 312
 313        if (user_mode(regs)) {
 314                if (!stack_access_ok(state, addr + sp_offset,
 315                                     REGS_SIZE - SP_OFFSET))
 316                        return false;
 317
 318                *sp = regs->sp;
 319        } else
 320                *sp = (unsigned long)&regs->sp;
 321
 322        return true;
 323}
 324
 325bool unwind_next_frame(struct unwind_state *state)
 326{
 327        unsigned long ip_p, sp, orig_ip, prev_sp = state->sp;
 328        enum stack_type prev_type = state->stack_info.type;
 329        struct orc_entry *orc;
 330        struct pt_regs *ptregs;
 331        bool indirect = false;
 332
 333        if (unwind_done(state))
 334                return false;
 335
 336        /* Don't let modules unload while we're reading their ORC data. */
 337        preempt_disable();
 338
 339        /* Have we reached the end? */
 340        if (state->regs && user_mode(state->regs))
 341                goto done;
 342
 343        /*
 344         * Find the orc_entry associated with the text address.
 345         *
 346         * Decrement call return addresses by one so they work for sibling
 347         * calls and calls to noreturn functions.
 348         */
 349        orc = orc_find(state->signal ? state->ip : state->ip - 1);
 350        if (!orc || orc->sp_reg == ORC_REG_UNDEFINED)
 351                goto done;
 352        orig_ip = state->ip;
 353
 354        /* Find the previous frame's stack: */
 355        switch (orc->sp_reg) {
 356        case ORC_REG_SP:
 357                sp = state->sp + orc->sp_offset;
 358                break;
 359
 360        case ORC_REG_BP:
 361                sp = state->bp + orc->sp_offset;
 362                break;
 363
 364        case ORC_REG_SP_INDIRECT:
 365                sp = state->sp + orc->sp_offset;
 366                indirect = true;
 367                break;
 368
 369        case ORC_REG_BP_INDIRECT:
 370                sp = state->bp + orc->sp_offset;
 371                indirect = true;
 372                break;
 373
 374        case ORC_REG_R10:
 375                if (!state->regs || !state->full_regs) {
 376                        orc_warn("missing regs for base reg R10 at ip %pB\n",
 377                                 (void *)state->ip);
 378                        goto done;
 379                }
 380                sp = state->regs->r10;
 381                break;
 382
 383        case ORC_REG_R13:
 384                if (!state->regs || !state->full_regs) {
 385                        orc_warn("missing regs for base reg R13 at ip %pB\n",
 386                                 (void *)state->ip);
 387                        goto done;
 388                }
 389                sp = state->regs->r13;
 390                break;
 391
 392        case ORC_REG_DI:
 393                if (!state->regs || !state->full_regs) {
 394                        orc_warn("missing regs for base reg DI at ip %pB\n",
 395                                 (void *)state->ip);
 396                        goto done;
 397                }
 398                sp = state->regs->di;
 399                break;
 400
 401        case ORC_REG_DX:
 402                if (!state->regs || !state->full_regs) {
 403                        orc_warn("missing regs for base reg DX at ip %pB\n",
 404                                 (void *)state->ip);
 405                        goto done;
 406                }
 407                sp = state->regs->dx;
 408                break;
 409
 410        default:
 411                orc_warn("unknown SP base reg %d for ip %pB\n",
 412                         orc->sp_reg, (void *)state->ip);
 413                goto done;
 414        }
 415
 416        if (indirect) {
 417                if (!deref_stack_reg(state, sp, &sp))
 418                        goto done;
 419        }
 420
 421        /* Find IP, SP and possibly regs: */
 422        switch (orc->type) {
 423        case ORC_TYPE_CALL:
 424                ip_p = sp - sizeof(long);
 425
 426                if (!deref_stack_reg(state, ip_p, &state->ip))
 427                        goto done;
 428
 429                state->ip = ftrace_graph_ret_addr(state->task, &state->graph_idx,
 430                                                  state->ip, (void *)ip_p);
 431
 432                state->sp = sp;
 433                state->regs = NULL;
 434                state->signal = false;
 435                break;
 436
 437        case ORC_TYPE_REGS:
 438                if (!deref_stack_regs(state, sp, &state->ip, &state->sp, true)) {
 439                        orc_warn("can't dereference registers at %p for ip %pB\n",
 440                                 (void *)sp, (void *)orig_ip);
 441                        goto done;
 442                }
 443
 444                state->regs = (struct pt_regs *)sp;
 445                state->full_regs = true;
 446                state->signal = true;
 447                break;
 448
 449        case ORC_TYPE_REGS_IRET:
 450                if (!deref_stack_regs(state, sp, &state->ip, &state->sp, false)) {
 451                        orc_warn("can't dereference iret registers at %p for ip %pB\n",
 452                                 (void *)sp, (void *)orig_ip);
 453                        goto done;
 454                }
 455
 456                ptregs = container_of((void *)sp, struct pt_regs, ip);
 457                if ((unsigned long)ptregs >= prev_sp &&
 458                    on_stack(&state->stack_info, ptregs, REGS_SIZE)) {
 459                        state->regs = ptregs;
 460                        state->full_regs = false;
 461                } else
 462                        state->regs = NULL;
 463
 464                state->signal = true;
 465                break;
 466
 467        default:
 468                orc_warn("unknown .orc_unwind entry type %d for ip %pB\n",
 469                         orc->type, (void *)orig_ip);
 470                break;
 471        }
 472
 473        /* Find BP: */
 474        switch (orc->bp_reg) {
 475        case ORC_REG_UNDEFINED:
 476                if (state->regs && state->full_regs)
 477                        state->bp = state->regs->bp;
 478                break;
 479
 480        case ORC_REG_PREV_SP:
 481                if (!deref_stack_reg(state, sp + orc->bp_offset, &state->bp))
 482                        goto done;
 483                break;
 484
 485        case ORC_REG_BP:
 486                if (!deref_stack_reg(state, state->bp + orc->bp_offset, &state->bp))
 487                        goto done;
 488                break;
 489
 490        default:
 491                orc_warn("unknown BP base reg %d for ip %pB\n",
 492                         orc->bp_reg, (void *)orig_ip);
 493                goto done;
 494        }
 495
 496        /* Prevent a recursive loop due to bad ORC data: */
 497        if (state->stack_info.type == prev_type &&
 498            on_stack(&state->stack_info, (void *)state->sp, sizeof(long)) &&
 499            state->sp <= prev_sp) {
 500                orc_warn("stack going in the wrong direction? ip=%pB\n",
 501                         (void *)orig_ip);
 502                goto done;
 503        }
 504
 505        preempt_enable();
 506        return true;
 507
 508done:
 509        preempt_enable();
 510        state->stack_info.type = STACK_TYPE_UNKNOWN;
 511        return false;
 512}
 513EXPORT_SYMBOL_GPL(unwind_next_frame);
 514
 515void __unwind_start(struct unwind_state *state, struct task_struct *task,
 516                    struct pt_regs *regs, unsigned long *first_frame)
 517{
 518        memset(state, 0, sizeof(*state));
 519        state->task = task;
 520
 521        /*
 522         * Refuse to unwind the stack of a task while it's executing on another
 523         * CPU.  This check is racy, but that's ok: the unwinder has other
 524         * checks to prevent it from going off the rails.
 525         */
 526        if (task_on_another_cpu(task))
 527                goto done;
 528
 529        if (regs) {
 530                if (user_mode(regs))
 531                        goto done;
 532
 533                state->ip = regs->ip;
 534                state->sp = kernel_stack_pointer(regs);
 535                state->bp = regs->bp;
 536                state->regs = regs;
 537                state->full_regs = true;
 538                state->signal = true;
 539
 540        } else if (task == current) {
 541                asm volatile("lea (%%rip), %0\n\t"
 542                             "mov %%rsp, %1\n\t"
 543                             "mov %%rbp, %2\n\t"
 544                             : "=r" (state->ip), "=r" (state->sp),
 545                               "=r" (state->bp));
 546
 547        } else {
 548                struct inactive_task_frame *frame = (void *)task->thread.sp;
 549
 550                state->sp = task->thread.sp;
 551                state->bp = READ_ONCE_NOCHECK(frame->bp);
 552                state->ip = READ_ONCE_NOCHECK(frame->ret_addr);
 553        }
 554
 555        if (get_stack_info((unsigned long *)state->sp, state->task,
 556                           &state->stack_info, &state->stack_mask))
 557                return;
 558
 559        /*
 560         * The caller can provide the address of the first frame directly
 561         * (first_frame) or indirectly (regs->sp) to indicate which stack frame
 562         * to start unwinding at.  Skip ahead until we reach it.
 563         */
 564
 565        /* When starting from regs, skip the regs frame: */
 566        if (regs) {
 567                unwind_next_frame(state);
 568                return;
 569        }
 570
 571        /* Otherwise, skip ahead to the user-specified starting frame: */
 572        while (!unwind_done(state) &&
 573               (!on_stack(&state->stack_info, first_frame, sizeof(long)) ||
 574                        state->sp <= (unsigned long)first_frame))
 575                unwind_next_frame(state);
 576
 577        return;
 578
 579done:
 580        state->stack_info.type = STACK_TYPE_UNKNOWN;
 581        return;
 582}
 583EXPORT_SYMBOL_GPL(__unwind_start);
 584