1 #include <linux/sched.h>
2 #include <linux/sched/task.h>
3 #include <linux/sched/task_stack.h>
4 #include <asm/ptrace.h>
5 #include <asm/bitops.h>
6 #include <asm/stacktrace.h>
7 #include <asm/unwind.h>
9 #define FRAME_HEADER_SIZE (sizeof(long) * 2)
12 * This disables KASAN checking when reading a value from another task's stack,
13 * since the other task could be running on another CPU and could have poisoned
14 * the stack in the meantime.
16 #define READ_ONCE_TASK_STACK(task, x) \
19 if (task == current) \
22 val = READ_ONCE_NOCHECK(x); \
26 static void unwind_dump(struct unwind_state *state, unsigned long *sp)
28 static bool dumped_before = false;
29 bool prev_zero, zero = false;
37 printk_deferred("unwind stack type:%d next_sp:%p mask:%lx graph_idx:%d\n",
38 state->stack_info.type, state->stack_info.next_sp,
39 state->stack_mask, state->graph_idx);
41 for (sp = state->orig_sp; sp < state->stack_info.end; sp++) {
42 word = READ_ONCE_NOCHECK(*sp);
49 printk_deferred("%p: %016x ...\n", sp, 0);
53 printk_deferred("%p: %016lx (%pB)\n", sp, word, (void *)word);
57 unsigned long unwind_get_return_address(struct unwind_state *state)
60 unsigned long *addr_p = unwind_get_return_address_ptr(state);
62 if (unwind_done(state))
65 if (state->regs && user_mode(state->regs))
68 addr = READ_ONCE_TASK_STACK(state->task, *addr_p);
69 addr = ftrace_graph_ret_addr(state->task, &state->graph_idx, addr,
72 return __kernel_text_address(addr) ? addr : 0;
74 EXPORT_SYMBOL_GPL(unwind_get_return_address);
76 static size_t regs_size(struct pt_regs *regs)
78 /* x86_32 regs from kernel mode are two words shorter: */
79 if (IS_ENABLED(CONFIG_X86_32) && !user_mode(regs))
80 return sizeof(*regs) - 2*sizeof(long);
85 static bool is_last_task_frame(struct unwind_state *state)
87 unsigned long bp = (unsigned long)state->bp;
88 unsigned long regs = (unsigned long)task_pt_regs(state->task);
91 * We have to check for the last task frame at two different locations
92 * because gcc can occasionally decide to realign the stack pointer and
93 * change the offset of the stack frame by a word in the prologue of a
94 * function called by head/entry code.
96 return bp == regs - FRAME_HEADER_SIZE ||
97 bp == regs - FRAME_HEADER_SIZE - sizeof(long);
101 * This determines if the frame pointer actually contains an encoded pointer to
102 * pt_regs on the stack. See ENCODE_FRAME_POINTER.
104 static struct pt_regs *decode_frame_pointer(unsigned long *bp)
106 unsigned long regs = (unsigned long)bp;
111 return (struct pt_regs *)(regs & ~0x1);
114 static bool update_stack_state(struct unwind_state *state, void *addr,
117 struct stack_info *info = &state->stack_info;
118 enum stack_type orig_type = info->type;
121 * If addr isn't on the current stack, switch to the next one.
123 * We may have to traverse multiple stacks to deal with the possibility
124 * that 'info->next_sp' could point to an empty stack and 'addr' could
125 * be on a subsequent stack.
127 while (!on_stack(info, addr, len))
128 if (get_stack_info(info->next_sp, state->task, info,
132 if (!state->orig_sp || info->type != orig_type)
133 state->orig_sp = addr;
138 bool unwind_next_frame(struct unwind_state *state)
140 struct pt_regs *regs;
141 unsigned long *next_bp, *next_frame;
143 enum stack_type prev_type = state->stack_info.type;
145 if (unwind_done(state))
148 /* have we reached the end? */
149 if (state->regs && user_mode(state->regs))
152 if (is_last_task_frame(state)) {
153 regs = task_pt_regs(state->task);
156 * kthreads (other than the boot CPU's idle thread) have some
157 * partial regs at the end of their stack which were placed
158 * there by copy_thread_tls(). But the regs don't have any
159 * useful information, so we can skip them.
161 * This user_mode() check is slightly broader than a PF_KTHREAD
162 * check because it also catches the awkward situation where a
163 * newly forked kthread transitions into a user task by calling
164 * do_execve(), which eventually clears PF_KTHREAD.
166 if (!user_mode(regs))
170 * We're almost at the end, but not quite: there's still the
171 * syscall regs frame. Entry code doesn't encode the regs
172 * pointer for syscalls, so we have to set it manually.
179 /* get the next frame pointer */
181 next_bp = (unsigned long *)state->regs->bp;
183 next_bp = (unsigned long *)READ_ONCE_TASK_STACK(state->task,*state->bp);
185 /* is the next frame pointer an encoded pointer to pt_regs? */
186 regs = decode_frame_pointer(next_bp);
188 next_frame = (unsigned long *)regs;
189 next_len = sizeof(*regs);
191 next_frame = next_bp;
192 next_len = FRAME_HEADER_SIZE;
195 /* make sure the next frame's data is accessible */
196 if (!update_stack_state(state, next_frame, next_len)) {
198 * Don't warn on bad regs->bp. An interrupt in entry code
199 * might cause a false positive warning.
207 /* Make sure it only unwinds up and doesn't overlap the last frame: */
208 if (state->stack_info.type == prev_type) {
209 if (state->regs && (void *)next_frame < (void *)state->regs + regs_size(state->regs))
212 if (state->bp && (void *)next_frame < (void *)state->bp + FRAME_HEADER_SIZE)
216 /* move to the next frame */
229 * When unwinding a non-current task, the task might actually be
230 * running on another CPU, in which case it could be modifying its
231 * stack while we're reading it. This is generally not a problem and
232 * can be ignored as long as the caller understands that unwinding
233 * another task will not always succeed.
235 if (state->task != current)
239 printk_deferred_once(KERN_WARNING
240 "WARNING: kernel stack regs at %p in %s:%d has bad 'bp' value %p\n",
241 state->regs, state->task->comm,
242 state->task->pid, next_frame);
243 unwind_dump(state, (unsigned long *)state->regs);
245 printk_deferred_once(KERN_WARNING
246 "WARNING: kernel stack frame pointer at %p in %s:%d has bad value %p\n",
247 state->bp, state->task->comm,
248 state->task->pid, next_frame);
249 unwind_dump(state, state->bp);
252 state->stack_info.type = STACK_TYPE_UNKNOWN;
255 EXPORT_SYMBOL_GPL(unwind_next_frame);
257 void __unwind_start(struct unwind_state *state, struct task_struct *task,
258 struct pt_regs *regs, unsigned long *first_frame)
260 unsigned long *bp, *frame;
263 memset(state, 0, sizeof(*state));
266 /* don't even attempt to start from user mode regs */
267 if (regs && user_mode(regs)) {
268 state->stack_info.type = STACK_TYPE_UNKNOWN;
272 /* set up the starting stack frame */
273 bp = get_frame_pointer(task, regs);
274 regs = decode_frame_pointer(bp);
277 frame = (unsigned long *)regs;
282 len = FRAME_HEADER_SIZE;
285 /* initialize stack info and make sure the frame data is accessible */
286 get_stack_info(frame, state->task, &state->stack_info,
288 update_stack_state(state, frame, len);
291 * The caller can provide the address of the first frame directly
292 * (first_frame) or indirectly (regs->sp) to indicate which stack frame
293 * to start unwinding at. Skip ahead until we reach it.
295 while (!unwind_done(state) &&
296 (!on_stack(&state->stack_info, first_frame, sizeof(long)) ||
297 state->bp < first_frame))
298 unwind_next_frame(state);
300 EXPORT_SYMBOL_GPL(__unwind_start);