1 // SPDX-License-Identifier: GPL-2.0-only 2 #include <linux/sched.h> 3 #include <linux/sched/task.h> 4 #include <linux/sched/task_stack.h> 5 #include <linux/interrupt.h> 6 #include <asm/sections.h> 7 #include <asm/ptrace.h> 8 #include <asm/bitops.h> 9 #include <asm/stacktrace.h> 10 #include <asm/unwind.h> 11 12 #define FRAME_HEADER_SIZE (sizeof(long) * 2) 13 14 unsigned long unwind_get_return_address(struct unwind_state *state) 15 { 16 if (unwind_done(state)) 17 return 0; 18 19 return __kernel_text_address(state->ip) ? state->ip : 0; 20 } 21 EXPORT_SYMBOL_GPL(unwind_get_return_address); 22 23 unsigned long *unwind_get_return_address_ptr(struct unwind_state *state) 24 { 25 if (unwind_done(state)) 26 return NULL; 27 28 return state->regs ? &state->regs->ip : state->bp + 1; 29 } 30 31 static void unwind_dump(struct unwind_state *state) 32 { 33 static bool dumped_before = false; 34 bool prev_zero, zero = false; 35 unsigned long word, *sp; 36 struct stack_info stack_info = {0}; 37 unsigned long visit_mask = 0; 38 39 if (dumped_before) 40 return; 41 42 dumped_before = true; 43 44 printk_deferred("unwind stack type:%d next_sp:%p mask:0x%lx graph_idx:%d\n", 45 state->stack_info.type, state->stack_info.next_sp, 46 state->stack_mask, state->graph_idx); 47 48 for (sp = PTR_ALIGN(state->orig_sp, sizeof(long)); sp; 49 sp = PTR_ALIGN(stack_info.next_sp, sizeof(long))) { 50 if (get_stack_info(sp, state->task, &stack_info, &visit_mask)) 51 break; 52 53 for (; sp < stack_info.end; sp++) { 54 55 word = READ_ONCE_NOCHECK(*sp); 56 57 prev_zero = zero; 58 zero = word == 0; 59 60 if (zero) { 61 if (!prev_zero) 62 printk_deferred("%p: %0*x ...\n", 63 sp, BITS_PER_LONG/4, 0); 64 continue; 65 } 66 67 printk_deferred("%p: %0*lx (%pB)\n", 68 sp, BITS_PER_LONG/4, word, (void *)word); 69 } 70 } 71 } 72 73 static size_t regs_size(struct pt_regs *regs) 74 { 75 /* x86_32 regs from kernel mode are two words shorter: */ 76 if (IS_ENABLED(CONFIG_X86_32) && !user_mode(regs)) 77 return sizeof(*regs) - 2*sizeof(long); 78 79 return sizeof(*regs); 80 } 81 82 static bool in_entry_code(unsigned long ip) 83 { 84 char *addr = (char *)ip; 85 86 if (addr >= __entry_text_start && addr < __entry_text_end) 87 return true; 88 89 if (addr >= __irqentry_text_start && addr < __irqentry_text_end) 90 return true; 91 92 return false; 93 } 94 95 static inline unsigned long *last_frame(struct unwind_state *state) 96 { 97 return (unsigned long *)task_pt_regs(state->task) - 2; 98 } 99 100 static bool is_last_frame(struct unwind_state *state) 101 { 102 return state->bp == last_frame(state); 103 } 104 105 #ifdef CONFIG_X86_32 106 #define GCC_REALIGN_WORDS 3 107 #else 108 #define GCC_REALIGN_WORDS 1 109 #endif 110 111 static inline unsigned long *last_aligned_frame(struct unwind_state *state) 112 { 113 return last_frame(state) - GCC_REALIGN_WORDS; 114 } 115 116 static bool is_last_aligned_frame(struct unwind_state *state) 117 { 118 unsigned long *last_bp = last_frame(state); 119 unsigned long *aligned_bp = last_aligned_frame(state); 120 121 /* 122 * GCC can occasionally decide to realign the stack pointer and change 123 * the offset of the stack frame in the prologue of a function called 124 * by head/entry code. Examples: 125 * 126 * <start_secondary>: 127 * push %edi 128 * lea 0x8(%esp),%edi 129 * and $0xfffffff8,%esp 130 * pushl -0x4(%edi) 131 * push %ebp 132 * mov %esp,%ebp 133 * 134 * <x86_64_start_kernel>: 135 * lea 0x8(%rsp),%r10 136 * and $0xfffffffffffffff0,%rsp 137 * pushq -0x8(%r10) 138 * push %rbp 139 * mov %rsp,%rbp 140 * 141 * After aligning the stack, it pushes a duplicate copy of the return 142 * address before pushing the frame pointer. 143 */ 144 return (state->bp == aligned_bp && *(aligned_bp + 1) == *(last_bp + 1)); 145 } 146 147 static bool is_last_ftrace_frame(struct unwind_state *state) 148 { 149 unsigned long *last_bp = last_frame(state); 150 unsigned long *last_ftrace_bp = last_bp - 3; 151 152 /* 153 * When unwinding from an ftrace handler of a function called by entry 154 * code, the stack layout of the last frame is: 155 * 156 * bp 157 * parent ret addr 158 * bp 159 * function ret addr 160 * parent ret addr 161 * pt_regs 162 * ----------------- 163 */ 164 return (state->bp == last_ftrace_bp && 165 *state->bp == *(state->bp + 2) && 166 *(state->bp + 1) == *(state->bp + 4)); 167 } 168 169 static bool is_last_task_frame(struct unwind_state *state) 170 { 171 return is_last_frame(state) || is_last_aligned_frame(state) || 172 is_last_ftrace_frame(state); 173 } 174 175 /* 176 * This determines if the frame pointer actually contains an encoded pointer to 177 * pt_regs on the stack. See ENCODE_FRAME_POINTER. 178 */ 179 #ifdef CONFIG_X86_64 180 static struct pt_regs *decode_frame_pointer(unsigned long *bp) 181 { 182 unsigned long regs = (unsigned long)bp; 183 184 if (!(regs & 0x1)) 185 return NULL; 186 187 return (struct pt_regs *)(regs & ~0x1); 188 } 189 #else 190 static struct pt_regs *decode_frame_pointer(unsigned long *bp) 191 { 192 unsigned long regs = (unsigned long)bp; 193 194 if (regs & 0x80000000) 195 return NULL; 196 197 return (struct pt_regs *)(regs | 0x80000000); 198 } 199 #endif 200 201 #ifdef CONFIG_X86_32 202 #define KERNEL_REGS_SIZE (sizeof(struct pt_regs) - 2*sizeof(long)) 203 #else 204 #define KERNEL_REGS_SIZE (sizeof(struct pt_regs)) 205 #endif 206 207 static bool update_stack_state(struct unwind_state *state, 208 unsigned long *next_bp) 209 { 210 struct stack_info *info = &state->stack_info; 211 enum stack_type prev_type = info->type; 212 struct pt_regs *regs; 213 unsigned long *frame, *prev_frame_end, *addr_p, addr; 214 size_t len; 215 216 if (state->regs) 217 prev_frame_end = (void *)state->regs + regs_size(state->regs); 218 else 219 prev_frame_end = (void *)state->bp + FRAME_HEADER_SIZE; 220 221 /* Is the next frame pointer an encoded pointer to pt_regs? */ 222 regs = decode_frame_pointer(next_bp); 223 if (regs) { 224 frame = (unsigned long *)regs; 225 len = KERNEL_REGS_SIZE; 226 state->got_irq = true; 227 } else { 228 frame = next_bp; 229 len = FRAME_HEADER_SIZE; 230 } 231 232 /* 233 * If the next bp isn't on the current stack, switch to the next one. 234 * 235 * We may have to traverse multiple stacks to deal with the possibility 236 * that info->next_sp could point to an empty stack and the next bp 237 * could be on a subsequent stack. 238 */ 239 while (!on_stack(info, frame, len)) 240 if (get_stack_info(info->next_sp, state->task, info, 241 &state->stack_mask)) 242 return false; 243 244 /* Make sure it only unwinds up and doesn't overlap the prev frame: */ 245 if (state->orig_sp && state->stack_info.type == prev_type && 246 frame < prev_frame_end) 247 return false; 248 249 /* 250 * On 32-bit with user mode regs, make sure the last two regs are safe 251 * to access: 252 */ 253 if (IS_ENABLED(CONFIG_X86_32) && regs && user_mode(regs) && 254 !on_stack(info, frame, len + 2*sizeof(long))) 255 return false; 256 257 /* Move state to the next frame: */ 258 if (regs) { 259 state->regs = regs; 260 state->bp = NULL; 261 } else { 262 state->bp = next_bp; 263 state->regs = NULL; 264 } 265 266 /* Save the return address: */ 267 if (state->regs && user_mode(state->regs)) 268 state->ip = 0; 269 else { 270 addr_p = unwind_get_return_address_ptr(state); 271 addr = READ_ONCE_TASK_STACK(state->task, *addr_p); 272 state->ip = ftrace_graph_ret_addr(state->task, &state->graph_idx, 273 addr, addr_p); 274 } 275 276 /* Save the original stack pointer for unwind_dump(): */ 277 if (!state->orig_sp) 278 state->orig_sp = frame; 279 280 return true; 281 } 282 283 bool unwind_next_frame(struct unwind_state *state) 284 { 285 struct pt_regs *regs; 286 unsigned long *next_bp; 287 288 if (unwind_done(state)) 289 return false; 290 291 /* Have we reached the end? */ 292 if (state->regs && user_mode(state->regs)) 293 goto the_end; 294 295 if (is_last_task_frame(state)) { 296 regs = task_pt_regs(state->task); 297 298 /* 299 * kthreads (other than the boot CPU's idle thread) have some 300 * partial regs at the end of their stack which were placed 301 * there by copy_thread_tls(). But the regs don't have any 302 * useful information, so we can skip them. 303 * 304 * This user_mode() check is slightly broader than a PF_KTHREAD 305 * check because it also catches the awkward situation where a 306 * newly forked kthread transitions into a user task by calling 307 * do_execve(), which eventually clears PF_KTHREAD. 308 */ 309 if (!user_mode(regs)) 310 goto the_end; 311 312 /* 313 * We're almost at the end, but not quite: there's still the 314 * syscall regs frame. Entry code doesn't encode the regs 315 * pointer for syscalls, so we have to set it manually. 316 */ 317 state->regs = regs; 318 state->bp = NULL; 319 state->ip = 0; 320 return true; 321 } 322 323 /* Get the next frame pointer: */ 324 if (state->next_bp) { 325 next_bp = state->next_bp; 326 state->next_bp = NULL; 327 } else if (state->regs) { 328 next_bp = (unsigned long *)state->regs->bp; 329 } else { 330 next_bp = (unsigned long *)READ_ONCE_TASK_STACK(state->task, *state->bp); 331 } 332 333 /* Move to the next frame if it's safe: */ 334 if (!update_stack_state(state, next_bp)) 335 goto bad_address; 336 337 return true; 338 339 bad_address: 340 state->error = true; 341 342 /* 343 * When unwinding a non-current task, the task might actually be 344 * running on another CPU, in which case it could be modifying its 345 * stack while we're reading it. This is generally not a problem and 346 * can be ignored as long as the caller understands that unwinding 347 * another task will not always succeed. 348 */ 349 if (state->task != current) 350 goto the_end; 351 352 /* 353 * Don't warn if the unwinder got lost due to an interrupt in entry 354 * code or in the C handler before the first frame pointer got set up: 355 */ 356 if (state->got_irq && in_entry_code(state->ip)) 357 goto the_end; 358 if (state->regs && 359 state->regs->sp >= (unsigned long)last_aligned_frame(state) && 360 state->regs->sp < (unsigned long)task_pt_regs(state->task)) 361 goto the_end; 362 363 /* 364 * There are some known frame pointer issues on 32-bit. Disable 365 * unwinder warnings on 32-bit until it gets objtool support. 366 */ 367 if (IS_ENABLED(CONFIG_X86_32)) 368 goto the_end; 369 370 if (state->regs) { 371 printk_deferred_once(KERN_WARNING 372 "WARNING: kernel stack regs at %p in %s:%d has bad 'bp' value %p\n", 373 state->regs, state->task->comm, 374 state->task->pid, next_bp); 375 unwind_dump(state); 376 } else { 377 printk_deferred_once(KERN_WARNING 378 "WARNING: kernel stack frame pointer at %p in %s:%d has bad value %p\n", 379 state->bp, state->task->comm, 380 state->task->pid, next_bp); 381 unwind_dump(state); 382 } 383 the_end: 384 state->stack_info.type = STACK_TYPE_UNKNOWN; 385 return false; 386 } 387 EXPORT_SYMBOL_GPL(unwind_next_frame); 388 389 void __unwind_start(struct unwind_state *state, struct task_struct *task, 390 struct pt_regs *regs, unsigned long *first_frame) 391 { 392 unsigned long *bp; 393 394 memset(state, 0, sizeof(*state)); 395 state->task = task; 396 state->got_irq = (regs); 397 398 /* Don't even attempt to start from user mode regs: */ 399 if (regs && user_mode(regs)) { 400 state->stack_info.type = STACK_TYPE_UNKNOWN; 401 return; 402 } 403 404 bp = get_frame_pointer(task, regs); 405 406 /* 407 * If we crash with IP==0, the last successfully executed instruction 408 * was probably an indirect function call with a NULL function pointer. 409 * That means that SP points into the middle of an incomplete frame: 410 * *SP is a return pointer, and *(SP-sizeof(unsigned long)) is where we 411 * would have written a frame pointer if we hadn't crashed. 412 * Pretend that the frame is complete and that BP points to it, but save 413 * the real BP so that we can use it when looking for the next frame. 414 */ 415 if (regs && regs->ip == 0 && 416 (unsigned long *)kernel_stack_pointer(regs) >= first_frame) { 417 state->next_bp = bp; 418 bp = ((unsigned long *)kernel_stack_pointer(regs)) - 1; 419 } 420 421 /* Initialize stack info and make sure the frame data is accessible: */ 422 get_stack_info(bp, state->task, &state->stack_info, 423 &state->stack_mask); 424 update_stack_state(state, bp); 425 426 /* 427 * The caller can provide the address of the first frame directly 428 * (first_frame) or indirectly (regs->sp) to indicate which stack frame 429 * to start unwinding at. Skip ahead until we reach it. 430 */ 431 while (!unwind_done(state) && 432 (!on_stack(&state->stack_info, first_frame, sizeof(long)) || 433 (state->next_bp == NULL && state->bp < first_frame))) 434 unwind_next_frame(state); 435 } 436 EXPORT_SYMBOL_GPL(__unwind_start); 437