arch/x86/kernel/unwind_frame.c

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