Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/extable.h>
3#include <linux/uaccess.h>
4#include <linux/sched/debug.h>
5#include <xen/xen.h>
6
7#include <asm/fpu/internal.h>
8#include <asm/traps.h>
9#include <asm/kdebug.h>
10
11typedef bool (*ex_handler_t)(const struct exception_table_entry *,
12 struct pt_regs *, int, unsigned long,
13 unsigned long);
14
15static inline unsigned long
16ex_fixup_addr(const struct exception_table_entry *x)
17{
18 return (unsigned long)&x->fixup + x->fixup;
19}
20static inline ex_handler_t
21ex_fixup_handler(const struct exception_table_entry *x)
22{
23 return (ex_handler_t)((unsigned long)&x->handler + x->handler);
24}
25
26__visible bool ex_handler_default(const struct exception_table_entry *fixup,
27 struct pt_regs *regs, int trapnr,
28 unsigned long error_code,
29 unsigned long fault_addr)
30{
31 regs->ip = ex_fixup_addr(fixup);
32 return true;
33}
34EXPORT_SYMBOL(ex_handler_default);
35
36__visible bool ex_handler_fault(const struct exception_table_entry *fixup,
37 struct pt_regs *regs, int trapnr,
38 unsigned long error_code,
39 unsigned long fault_addr)
40{
41 regs->ip = ex_fixup_addr(fixup);
42 regs->ax = trapnr;
43 return true;
44}
45EXPORT_SYMBOL_GPL(ex_handler_fault);
46
47/*
48 * Handler for UD0 exception following a failed test against the
49 * result of a refcount inc/dec/add/sub.
50 */
51__visible bool ex_handler_refcount(const struct exception_table_entry *fixup,
52 struct pt_regs *regs, int trapnr,
53 unsigned long error_code,
54 unsigned long fault_addr)
55{
56 /* First unconditionally saturate the refcount. */
57 *(int *)regs->cx = INT_MIN / 2;
58
59 /*
60 * Strictly speaking, this reports the fixup destination, not
61 * the fault location, and not the actually overflowing
62 * instruction, which is the instruction before the "js", but
63 * since that instruction could be a variety of lengths, just
64 * report the location after the overflow, which should be close
65 * enough for finding the overflow, as it's at least back in
66 * the function, having returned from .text.unlikely.
67 */
68 regs->ip = ex_fixup_addr(fixup);
69
70 /*
71 * This function has been called because either a negative refcount
72 * value was seen by any of the refcount functions, or a zero
73 * refcount value was seen by refcount_dec().
74 *
75 * If we crossed from INT_MAX to INT_MIN, OF (Overflow Flag: result
76 * wrapped around) will be set. Additionally, seeing the refcount
77 * reach 0 will set ZF (Zero Flag: result was zero). In each of
78 * these cases we want a report, since it's a boundary condition.
79 * The SF case is not reported since it indicates post-boundary
80 * manipulations below zero or above INT_MAX. And if none of the
81 * flags are set, something has gone very wrong, so report it.
82 */
83 if (regs->flags & (X86_EFLAGS_OF | X86_EFLAGS_ZF)) {
84 bool zero = regs->flags & X86_EFLAGS_ZF;
85
86 refcount_error_report(regs, zero ? "hit zero" : "overflow");
87 } else if ((regs->flags & X86_EFLAGS_SF) == 0) {
88 /* Report if none of OF, ZF, nor SF are set. */
89 refcount_error_report(regs, "unexpected saturation");
90 }
91
92 return true;
93}
94EXPORT_SYMBOL(ex_handler_refcount);
95
96/*
97 * Handler for when we fail to restore a task's FPU state. We should never get
98 * here because the FPU state of a task using the FPU (task->thread.fpu.state)
99 * should always be valid. However, past bugs have allowed userspace to set
100 * reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn().
101 * These caused XRSTOR to fail when switching to the task, leaking the FPU
102 * registers of the task previously executing on the CPU. Mitigate this class
103 * of vulnerability by restoring from the initial state (essentially, zeroing
104 * out all the FPU registers) if we can't restore from the task's FPU state.
105 */
106__visible bool ex_handler_fprestore(const struct exception_table_entry *fixup,
107 struct pt_regs *regs, int trapnr,
108 unsigned long error_code,
109 unsigned long fault_addr)
110{
111 regs->ip = ex_fixup_addr(fixup);
112
113 WARN_ONCE(1, "Bad FPU state detected at %pB, reinitializing FPU registers.",
114 (void *)instruction_pointer(regs));
115
116 __copy_kernel_to_fpregs(&init_fpstate, -1);
117 return true;
118}
119EXPORT_SYMBOL_GPL(ex_handler_fprestore);
120
121__visible bool ex_handler_uaccess(const struct exception_table_entry *fixup,
122 struct pt_regs *regs, int trapnr,
123 unsigned long error_code,
124 unsigned long fault_addr)
125{
126 WARN_ONCE(trapnr == X86_TRAP_GP, "General protection fault in user access. Non-canonical address?");
127 regs->ip = ex_fixup_addr(fixup);
128 return true;
129}
130EXPORT_SYMBOL(ex_handler_uaccess);
131
132__visible bool ex_handler_ext(const struct exception_table_entry *fixup,
133 struct pt_regs *regs, int trapnr,
134 unsigned long error_code,
135 unsigned long fault_addr)
136{
137 /* Special hack for uaccess_err */
138 current->thread.uaccess_err = 1;
139 regs->ip = ex_fixup_addr(fixup);
140 return true;
141}
142EXPORT_SYMBOL(ex_handler_ext);
143
144__visible bool ex_handler_rdmsr_unsafe(const struct exception_table_entry *fixup,
145 struct pt_regs *regs, int trapnr,
146 unsigned long error_code,
147 unsigned long fault_addr)
148{
149 if (pr_warn_once("unchecked MSR access error: RDMSR from 0x%x at rIP: 0x%lx (%pS)\n",
150 (unsigned int)regs->cx, regs->ip, (void *)regs->ip))
151 show_stack_regs(regs);
152
153 /* Pretend that the read succeeded and returned 0. */
154 regs->ip = ex_fixup_addr(fixup);
155 regs->ax = 0;
156 regs->dx = 0;
157 return true;
158}
159EXPORT_SYMBOL(ex_handler_rdmsr_unsafe);
160
161__visible bool ex_handler_wrmsr_unsafe(const struct exception_table_entry *fixup,
162 struct pt_regs *regs, int trapnr,
163 unsigned long error_code,
164 unsigned long fault_addr)
165{
166 if (pr_warn_once("unchecked MSR access error: WRMSR to 0x%x (tried to write 0x%08x%08x) at rIP: 0x%lx (%pS)\n",
167 (unsigned int)regs->cx, (unsigned int)regs->dx,
168 (unsigned int)regs->ax, regs->ip, (void *)regs->ip))
169 show_stack_regs(regs);
170
171 /* Pretend that the write succeeded. */
172 regs->ip = ex_fixup_addr(fixup);
173 return true;
174}
175EXPORT_SYMBOL(ex_handler_wrmsr_unsafe);
176
177__visible bool ex_handler_clear_fs(const struct exception_table_entry *fixup,
178 struct pt_regs *regs, int trapnr,
179 unsigned long error_code,
180 unsigned long fault_addr)
181{
182 if (static_cpu_has(X86_BUG_NULL_SEG))
183 asm volatile ("mov %0, %%fs" : : "rm" (__USER_DS));
184 asm volatile ("mov %0, %%fs" : : "rm" (0));
185 return ex_handler_default(fixup, regs, trapnr, error_code, fault_addr);
186}
187EXPORT_SYMBOL(ex_handler_clear_fs);
188
189__visible bool ex_has_fault_handler(unsigned long ip)
190{
191 const struct exception_table_entry *e;
192 ex_handler_t handler;
193
194 e = search_exception_tables(ip);
195 if (!e)
196 return false;
197 handler = ex_fixup_handler(e);
198
199 return handler == ex_handler_fault;
200}
201
202int fixup_exception(struct pt_regs *regs, int trapnr, unsigned long error_code,
203 unsigned long fault_addr)
204{
205 const struct exception_table_entry *e;
206 ex_handler_t handler;
207
208#ifdef CONFIG_PNPBIOS
209 if (unlikely(SEGMENT_IS_PNP_CODE(regs->cs))) {
210 extern u32 pnp_bios_fault_eip, pnp_bios_fault_esp;
211 extern u32 pnp_bios_is_utter_crap;
212 pnp_bios_is_utter_crap = 1;
213 printk(KERN_CRIT "PNPBIOS fault.. attempting recovery.\n");
214 __asm__ volatile(
215 "movl %0, %%esp\n\t"
216 "jmp *%1\n\t"
217 : : "g" (pnp_bios_fault_esp), "g" (pnp_bios_fault_eip));
218 panic("do_trap: can't hit this");
219 }
220#endif
221
222 e = search_exception_tables(regs->ip);
223 if (!e)
224 return 0;
225
226 handler = ex_fixup_handler(e);
227 return handler(e, regs, trapnr, error_code, fault_addr);
228}
229
230extern unsigned int early_recursion_flag;
231
232/* Restricted version used during very early boot */
233void __init early_fixup_exception(struct pt_regs *regs, int trapnr)
234{
235 /* Ignore early NMIs. */
236 if (trapnr == X86_TRAP_NMI)
237 return;
238
239 if (early_recursion_flag > 2)
240 goto halt_loop;
241
242 /*
243 * Old CPUs leave the high bits of CS on the stack
244 * undefined. I'm not sure which CPUs do this, but at least
245 * the 486 DX works this way.
246 * Xen pv domains are not using the default __KERNEL_CS.
247 */
248 if (!xen_pv_domain() && regs->cs != __KERNEL_CS)
249 goto fail;
250
251 /*
252 * The full exception fixup machinery is available as soon as
253 * the early IDT is loaded. This means that it is the
254 * responsibility of extable users to either function correctly
255 * when handlers are invoked early or to simply avoid causing
256 * exceptions before they're ready to handle them.
257 *
258 * This is better than filtering which handlers can be used,
259 * because refusing to call a handler here is guaranteed to
260 * result in a hard-to-debug panic.
261 *
262 * Keep in mind that not all vectors actually get here. Early
263 * page faults, for example, are special.
264 */
265 if (fixup_exception(regs, trapnr, regs->orig_ax, 0))
266 return;
267
268 if (fixup_bug(regs, trapnr))
269 return;
270
271fail:
272 early_printk("PANIC: early exception 0x%02x IP %lx:%lx error %lx cr2 0x%lx\n",
273 (unsigned)trapnr, (unsigned long)regs->cs, regs->ip,
274 regs->orig_ax, read_cr2());
275
276 show_regs(regs);
277
278halt_loop:
279 while (true)
280 halt();
281}
1#include <linux/module.h>
2#include <linux/spinlock.h>
3#include <linux/sort.h>
4#include <asm/uaccess.h>
5
6static inline unsigned long
7ex_insn_addr(const struct exception_table_entry *x)
8{
9 return (unsigned long)&x->insn + x->insn;
10}
11static inline unsigned long
12ex_fixup_addr(const struct exception_table_entry *x)
13{
14 return (unsigned long)&x->fixup + x->fixup;
15}
16
17int fixup_exception(struct pt_regs *regs)
18{
19 const struct exception_table_entry *fixup;
20 unsigned long new_ip;
21
22#ifdef CONFIG_PNPBIOS
23 if (unlikely(SEGMENT_IS_PNP_CODE(regs->cs))) {
24 extern u32 pnp_bios_fault_eip, pnp_bios_fault_esp;
25 extern u32 pnp_bios_is_utter_crap;
26 pnp_bios_is_utter_crap = 1;
27 printk(KERN_CRIT "PNPBIOS fault.. attempting recovery.\n");
28 __asm__ volatile(
29 "movl %0, %%esp\n\t"
30 "jmp *%1\n\t"
31 : : "g" (pnp_bios_fault_esp), "g" (pnp_bios_fault_eip));
32 panic("do_trap: can't hit this");
33 }
34#endif
35
36 fixup = search_exception_tables(regs->ip);
37 if (fixup) {
38 new_ip = ex_fixup_addr(fixup);
39
40 if (fixup->fixup - fixup->insn >= 0x7ffffff0 - 4) {
41 /* Special hack for uaccess_err */
42 current_thread_info()->uaccess_err = 1;
43 new_ip -= 0x7ffffff0;
44 }
45 regs->ip = new_ip;
46 return 1;
47 }
48
49 return 0;
50}
51
52/* Restricted version used during very early boot */
53int __init early_fixup_exception(unsigned long *ip)
54{
55 const struct exception_table_entry *fixup;
56 unsigned long new_ip;
57
58 fixup = search_exception_tables(*ip);
59 if (fixup) {
60 new_ip = ex_fixup_addr(fixup);
61
62 if (fixup->fixup - fixup->insn >= 0x7ffffff0 - 4) {
63 /* uaccess handling not supported during early boot */
64 return 0;
65 }
66
67 *ip = new_ip;
68 return 1;
69 }
70
71 return 0;
72}
73
74/*
75 * Search one exception table for an entry corresponding to the
76 * given instruction address, and return the address of the entry,
77 * or NULL if none is found.
78 * We use a binary search, and thus we assume that the table is
79 * already sorted.
80 */
81const struct exception_table_entry *
82search_extable(const struct exception_table_entry *first,
83 const struct exception_table_entry *last,
84 unsigned long value)
85{
86 while (first <= last) {
87 const struct exception_table_entry *mid;
88 unsigned long addr;
89
90 mid = ((last - first) >> 1) + first;
91 addr = ex_insn_addr(mid);
92 if (addr < value)
93 first = mid + 1;
94 else if (addr > value)
95 last = mid - 1;
96 else
97 return mid;
98 }
99 return NULL;
100}
101
102/*
103 * The exception table needs to be sorted so that the binary
104 * search that we use to find entries in it works properly.
105 * This is used both for the kernel exception table and for
106 * the exception tables of modules that get loaded.
107 *
108 */
109static int cmp_ex(const void *a, const void *b)
110{
111 const struct exception_table_entry *x = a, *y = b;
112
113 /*
114 * This value will always end up fittin in an int, because on
115 * both i386 and x86-64 the kernel symbol-reachable address
116 * space is < 2 GiB.
117 *
118 * This compare is only valid after normalization.
119 */
120 return x->insn - y->insn;
121}
122
123void sort_extable(struct exception_table_entry *start,
124 struct exception_table_entry *finish)
125{
126 struct exception_table_entry *p;
127 int i;
128
129 /* Convert all entries to being relative to the start of the section */
130 i = 0;
131 for (p = start; p < finish; p++) {
132 p->insn += i;
133 i += 4;
134 p->fixup += i;
135 i += 4;
136 }
137
138 sort(start, finish - start, sizeof(struct exception_table_entry),
139 cmp_ex, NULL);
140
141 /* Denormalize all entries */
142 i = 0;
143 for (p = start; p < finish; p++) {
144 p->insn -= i;
145 i += 4;
146 p->fixup -= i;
147 i += 4;
148 }
149}
150
151#ifdef CONFIG_MODULES
152/*
153 * If the exception table is sorted, any referring to the module init
154 * will be at the beginning or the end.
155 */
156void trim_init_extable(struct module *m)
157{
158 /*trim the beginning*/
159 while (m->num_exentries &&
160 within_module_init(ex_insn_addr(&m->extable[0]), m)) {
161 m->extable++;
162 m->num_exentries--;
163 }
164 /*trim the end*/
165 while (m->num_exentries &&
166 within_module_init(ex_insn_addr(&m->extable[m->num_exentries-1]), m))
167 m->num_exentries--;
168}
169#endif /* CONFIG_MODULES */