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 <linux/bitfield.h>
6#include <xen/xen.h>
7
8#include <asm/fpu/api.h>
9#include <asm/sev.h>
10#include <asm/traps.h>
11#include <asm/kdebug.h>
12#include <asm/insn-eval.h>
13#include <asm/sgx.h>
14
15static inline unsigned long *pt_regs_nr(struct pt_regs *regs, int nr)
16{
17 int reg_offset = pt_regs_offset(regs, regno: nr);
18 static unsigned long __dummy;
19
20 if (WARN_ON_ONCE(reg_offset < 0))
21 return &__dummy;
22
23 return (unsigned long *)((unsigned long)regs + reg_offset);
24}
25
26static inline unsigned long
27ex_fixup_addr(const struct exception_table_entry *x)
28{
29 return (unsigned long)&x->fixup + x->fixup;
30}
31
32static bool ex_handler_default(const struct exception_table_entry *e,
33 struct pt_regs *regs)
34{
35 if (e->data & EX_FLAG_CLEAR_AX)
36 regs->ax = 0;
37 if (e->data & EX_FLAG_CLEAR_DX)
38 regs->dx = 0;
39
40 regs->ip = ex_fixup_addr(x: e);
41 return true;
42}
43
44/*
45 * This is the *very* rare case where we do a "load_unaligned_zeropad()"
46 * and it's a page crosser into a non-existent page.
47 *
48 * This happens when we optimistically load a pathname a word-at-a-time
49 * and the name is less than the full word and the next page is not
50 * mapped. Typically that only happens for CONFIG_DEBUG_PAGEALLOC.
51 *
52 * NOTE! The faulting address is always a 'mov mem,reg' type instruction
53 * of size 'long', and the exception fixup must always point to right
54 * after the instruction.
55 */
56static bool ex_handler_zeropad(const struct exception_table_entry *e,
57 struct pt_regs *regs,
58 unsigned long fault_addr)
59{
60 struct insn insn;
61 const unsigned long mask = sizeof(long) - 1;
62 unsigned long offset, addr, next_ip, len;
63 unsigned long *reg;
64
65 next_ip = ex_fixup_addr(x: e);
66 len = next_ip - regs->ip;
67 if (len > MAX_INSN_SIZE)
68 return false;
69
70 if (insn_decode(insn: &insn, kaddr: (void *) regs->ip, buf_len: len, m: INSN_MODE_KERN))
71 return false;
72 if (insn.length != len)
73 return false;
74
75 if (insn.opcode.bytes[0] != 0x8b)
76 return false;
77 if (insn.opnd_bytes != sizeof(long))
78 return false;
79
80 addr = (unsigned long) insn_get_addr_ref(insn: &insn, regs);
81 if (addr == ~0ul)
82 return false;
83
84 offset = addr & mask;
85 addr = addr & ~mask;
86 if (fault_addr != addr + sizeof(long))
87 return false;
88
89 reg = insn_get_modrm_reg_ptr(insn: &insn, regs);
90 if (!reg)
91 return false;
92
93 *reg = *(unsigned long *)addr >> (offset * 8);
94 return ex_handler_default(e, regs);
95}
96
97static bool ex_handler_fault(const struct exception_table_entry *fixup,
98 struct pt_regs *regs, int trapnr)
99{
100 regs->ax = trapnr;
101 return ex_handler_default(e: fixup, regs);
102}
103
104static bool ex_handler_sgx(const struct exception_table_entry *fixup,
105 struct pt_regs *regs, int trapnr)
106{
107 regs->ax = trapnr | SGX_ENCLS_FAULT_FLAG;
108 return ex_handler_default(e: fixup, regs);
109}
110
111/*
112 * Handler for when we fail to restore a task's FPU state. We should never get
113 * here because the FPU state of a task using the FPU (task->thread.fpu.state)
114 * should always be valid. However, past bugs have allowed userspace to set
115 * reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn().
116 * These caused XRSTOR to fail when switching to the task, leaking the FPU
117 * registers of the task previously executing on the CPU. Mitigate this class
118 * of vulnerability by restoring from the initial state (essentially, zeroing
119 * out all the FPU registers) if we can't restore from the task's FPU state.
120 */
121static bool ex_handler_fprestore(const struct exception_table_entry *fixup,
122 struct pt_regs *regs)
123{
124 regs->ip = ex_fixup_addr(x: fixup);
125
126 WARN_ONCE(1, "Bad FPU state detected at %pB, reinitializing FPU registers.",
127 (void *)instruction_pointer(regs));
128
129 fpu_reset_from_exception_fixup();
130 return true;
131}
132
133/*
134 * On x86-64, we end up being imprecise with 'access_ok()', and allow
135 * non-canonical user addresses to make the range comparisons simpler,
136 * and to not have to worry about LAM being enabled.
137 *
138 * In fact, we allow up to one page of "slop" at the sign boundary,
139 * which means that we can do access_ok() by just checking the sign
140 * of the pointer for the common case of having a small access size.
141 */
142static bool gp_fault_address_ok(unsigned long fault_address)
143{
144#ifdef CONFIG_X86_64
145 /* Is it in the "user space" part of the non-canonical space? */
146 if (valid_user_address(fault_address))
147 return true;
148
149 /* .. or just above it? */
150 fault_address -= PAGE_SIZE;
151 if (valid_user_address(fault_address))
152 return true;
153#endif
154 return false;
155}
156
157static bool ex_handler_uaccess(const struct exception_table_entry *fixup,
158 struct pt_regs *regs, int trapnr,
159 unsigned long fault_address)
160{
161 WARN_ONCE(trapnr == X86_TRAP_GP && !gp_fault_address_ok(fault_address),
162 "General protection fault in user access. Non-canonical address?");
163 return ex_handler_default(e: fixup, regs);
164}
165
166static bool ex_handler_copy(const struct exception_table_entry *fixup,
167 struct pt_regs *regs, int trapnr)
168{
169 WARN_ONCE(trapnr == X86_TRAP_GP, "General protection fault in user access. Non-canonical address?");
170 return ex_handler_fault(fixup, regs, trapnr);
171}
172
173static bool ex_handler_msr(const struct exception_table_entry *fixup,
174 struct pt_regs *regs, bool wrmsr, bool safe, int reg)
175{
176 if (__ONCE_LITE_IF(!safe && wrmsr)) {
177 pr_warn("unchecked MSR access error: WRMSR to 0x%x (tried to write 0x%08x%08x) at rIP: 0x%lx (%pS)\n",
178 (unsigned int)regs->cx, (unsigned int)regs->dx,
179 (unsigned int)regs->ax, regs->ip, (void *)regs->ip);
180 show_stack_regs(regs);
181 }
182
183 if (__ONCE_LITE_IF(!safe && !wrmsr)) {
184 pr_warn("unchecked MSR access error: RDMSR from 0x%x at rIP: 0x%lx (%pS)\n",
185 (unsigned int)regs->cx, regs->ip, (void *)regs->ip);
186 show_stack_regs(regs);
187 }
188
189 if (!wrmsr) {
190 /* Pretend that the read succeeded and returned 0. */
191 regs->ax = 0;
192 regs->dx = 0;
193 }
194
195 if (safe)
196 *pt_regs_nr(regs, nr: reg) = -EIO;
197
198 return ex_handler_default(e: fixup, regs);
199}
200
201static bool ex_handler_clear_fs(const struct exception_table_entry *fixup,
202 struct pt_regs *regs)
203{
204 if (static_cpu_has(X86_BUG_NULL_SEG))
205 asm volatile ("mov %0, %%fs" : : "rm" (__USER_DS));
206 asm volatile ("mov %0, %%fs" : : "rm" (0));
207 return ex_handler_default(e: fixup, regs);
208}
209
210static bool ex_handler_imm_reg(const struct exception_table_entry *fixup,
211 struct pt_regs *regs, int reg, int imm)
212{
213 *pt_regs_nr(regs, nr: reg) = (long)imm;
214 return ex_handler_default(e: fixup, regs);
215}
216
217static bool ex_handler_ucopy_len(const struct exception_table_entry *fixup,
218 struct pt_regs *regs, int trapnr,
219 unsigned long fault_address,
220 int reg, int imm)
221{
222 regs->cx = imm * regs->cx + *pt_regs_nr(regs, nr: reg);
223 return ex_handler_uaccess(fixup, regs, trapnr, fault_address);
224}
225
226int ex_get_fixup_type(unsigned long ip)
227{
228 const struct exception_table_entry *e = search_exception_tables(add: ip);
229
230 return e ? FIELD_GET(EX_DATA_TYPE_MASK, e->data) : EX_TYPE_NONE;
231}
232
233int fixup_exception(struct pt_regs *regs, int trapnr, unsigned long error_code,
234 unsigned long fault_addr)
235{
236 const struct exception_table_entry *e;
237 int type, reg, imm;
238
239#ifdef CONFIG_PNPBIOS
240 if (unlikely(SEGMENT_IS_PNP_CODE(regs->cs))) {
241 extern u32 pnp_bios_fault_eip, pnp_bios_fault_esp;
242 extern u32 pnp_bios_is_utter_crap;
243 pnp_bios_is_utter_crap = 1;
244 printk(KERN_CRIT "PNPBIOS fault.. attempting recovery.\n");
245 __asm__ volatile(
246 "movl %0, %%esp\n\t"
247 "jmp *%1\n\t"
248 : : "g" (pnp_bios_fault_esp), "g" (pnp_bios_fault_eip));
249 panic("do_trap: can't hit this");
250 }
251#endif
252
253 e = search_exception_tables(add: regs->ip);
254 if (!e)
255 return 0;
256
257 type = FIELD_GET(EX_DATA_TYPE_MASK, e->data);
258 reg = FIELD_GET(EX_DATA_REG_MASK, e->data);
259 imm = FIELD_GET(EX_DATA_IMM_MASK, e->data);
260
261 switch (type) {
262 case EX_TYPE_DEFAULT:
263 case EX_TYPE_DEFAULT_MCE_SAFE:
264 return ex_handler_default(e, regs);
265 case EX_TYPE_FAULT:
266 case EX_TYPE_FAULT_MCE_SAFE:
267 return ex_handler_fault(fixup: e, regs, trapnr);
268 case EX_TYPE_UACCESS:
269 return ex_handler_uaccess(fixup: e, regs, trapnr, fault_address: fault_addr);
270 case EX_TYPE_COPY:
271 return ex_handler_copy(fixup: e, regs, trapnr);
272 case EX_TYPE_CLEAR_FS:
273 return ex_handler_clear_fs(fixup: e, regs);
274 case EX_TYPE_FPU_RESTORE:
275 return ex_handler_fprestore(fixup: e, regs);
276 case EX_TYPE_BPF:
277 return ex_handler_bpf(x: e, regs);
278 case EX_TYPE_WRMSR:
279 return ex_handler_msr(fixup: e, regs, wrmsr: true, safe: false, reg);
280 case EX_TYPE_RDMSR:
281 return ex_handler_msr(fixup: e, regs, wrmsr: false, safe: false, reg);
282 case EX_TYPE_WRMSR_SAFE:
283 return ex_handler_msr(fixup: e, regs, wrmsr: true, safe: true, reg);
284 case EX_TYPE_RDMSR_SAFE:
285 return ex_handler_msr(fixup: e, regs, wrmsr: false, safe: true, reg);
286 case EX_TYPE_WRMSR_IN_MCE:
287 ex_handler_msr_mce(regs, wrmsr: true);
288 break;
289 case EX_TYPE_RDMSR_IN_MCE:
290 ex_handler_msr_mce(regs, wrmsr: false);
291 break;
292 case EX_TYPE_POP_REG:
293 regs->sp += sizeof(long);
294 fallthrough;
295 case EX_TYPE_IMM_REG:
296 return ex_handler_imm_reg(fixup: e, regs, reg, imm);
297 case EX_TYPE_FAULT_SGX:
298 return ex_handler_sgx(fixup: e, regs, trapnr);
299 case EX_TYPE_UCOPY_LEN:
300 return ex_handler_ucopy_len(fixup: e, regs, trapnr, fault_address: fault_addr, reg, imm);
301 case EX_TYPE_ZEROPAD:
302 return ex_handler_zeropad(e, regs, fault_addr);
303 }
304 BUG();
305}
306
307extern unsigned int early_recursion_flag;
308
309/* Restricted version used during very early boot */
310void __init early_fixup_exception(struct pt_regs *regs, int trapnr)
311{
312 /* Ignore early NMIs. */
313 if (trapnr == X86_TRAP_NMI)
314 return;
315
316 if (early_recursion_flag > 2)
317 goto halt_loop;
318
319 /*
320 * Old CPUs leave the high bits of CS on the stack
321 * undefined. I'm not sure which CPUs do this, but at least
322 * the 486 DX works this way.
323 * Xen pv domains are not using the default __KERNEL_CS.
324 */
325 if (!xen_pv_domain() && regs->cs != __KERNEL_CS)
326 goto fail;
327
328 /*
329 * The full exception fixup machinery is available as soon as
330 * the early IDT is loaded. This means that it is the
331 * responsibility of extable users to either function correctly
332 * when handlers are invoked early or to simply avoid causing
333 * exceptions before they're ready to handle them.
334 *
335 * This is better than filtering which handlers can be used,
336 * because refusing to call a handler here is guaranteed to
337 * result in a hard-to-debug panic.
338 *
339 * Keep in mind that not all vectors actually get here. Early
340 * page faults, for example, are special.
341 */
342 if (fixup_exception(regs, trapnr, error_code: regs->orig_ax, fault_addr: 0))
343 return;
344
345 if (trapnr == X86_TRAP_UD) {
346 if (report_bug(bug_addr: regs->ip, regs) == BUG_TRAP_TYPE_WARN) {
347 /* Skip the ud2. */
348 regs->ip += LEN_UD2;
349 return;
350 }
351
352 /*
353 * If this was a BUG and report_bug returns or if this
354 * was just a normal #UD, we want to continue onward and
355 * crash.
356 */
357 }
358
359fail:
360 early_printk(fmt: "PANIC: early exception 0x%02x IP %lx:%lx error %lx cr2 0x%lx\n",
361 (unsigned)trapnr, (unsigned long)regs->cs, regs->ip,
362 regs->orig_ax, read_cr2());
363
364 show_regs(regs);
365
366halt_loop:
367 while (true)
368 halt();
369}
370

source code of linux/arch/x86/mm/extable.c