Linux Audio

Check our new training course

Loading...
v6.2
  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, 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(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(e);
 66	len = next_ip - regs->ip;
 67	if (len > MAX_INSN_SIZE)
 68		return false;
 69
 70	if (insn_decode(&insn, (void *) regs->ip, len, 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, 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, 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(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(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(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
133static bool ex_handler_uaccess(const struct exception_table_entry *fixup,
134			       struct pt_regs *regs, int trapnr)
135{
136	WARN_ONCE(trapnr == X86_TRAP_GP, "General protection fault in user access. Non-canonical address?");
137	return ex_handler_default(fixup, regs);
 
 
138}
 
139
140static bool ex_handler_copy(const struct exception_table_entry *fixup,
141			    struct pt_regs *regs, int trapnr)
142{
143	WARN_ONCE(trapnr == X86_TRAP_GP, "General protection fault in user access. Non-canonical address?");
144	return ex_handler_fault(fixup, regs, trapnr);
 
 
 
 
 
 
 
145}
 
146
147static bool ex_handler_msr(const struct exception_table_entry *fixup,
148			   struct pt_regs *regs, bool wrmsr, bool safe, int reg)
149{
150	if (__ONCE_LITE_IF(!safe && wrmsr)) {
151		pr_warn("unchecked MSR access error: WRMSR to 0x%x (tried to write 0x%08x%08x) at rIP: 0x%lx (%pS)\n",
152			(unsigned int)regs->cx, (unsigned int)regs->dx,
153			(unsigned int)regs->ax,  regs->ip, (void *)regs->ip);
154		show_stack_regs(regs);
155	}
156
157	if (__ONCE_LITE_IF(!safe && !wrmsr)) {
158		pr_warn("unchecked MSR access error: RDMSR from 0x%x at rIP: 0x%lx (%pS)\n",
159			(unsigned int)regs->cx, regs->ip, (void *)regs->ip);
160		show_stack_regs(regs);
161	}
162
163	if (!wrmsr) {
164		/* Pretend that the read succeeded and returned 0. */
165		regs->ax = 0;
166		regs->dx = 0;
167	}
168
169	if (safe)
170		*pt_regs_nr(regs, reg) = -EIO;
171
172	return ex_handler_default(fixup, regs);
173}
 
174
175static bool ex_handler_clear_fs(const struct exception_table_entry *fixup,
176				struct pt_regs *regs)
177{
178	if (static_cpu_has(X86_BUG_NULL_SEG))
179		asm volatile ("mov %0, %%fs" : : "rm" (__USER_DS));
180	asm volatile ("mov %0, %%fs" : : "rm" (0));
181	return ex_handler_default(fixup, regs);
182}
183
184static bool ex_handler_imm_reg(const struct exception_table_entry *fixup,
185			       struct pt_regs *regs, int reg, int imm)
186{
187	*pt_regs_nr(regs, reg) = (long)imm;
188	return ex_handler_default(fixup, regs);
189}
 
190
191static bool ex_handler_ucopy_len(const struct exception_table_entry *fixup,
192				  struct pt_regs *regs, int trapnr, int reg, int imm)
193{
194	regs->cx = imm * regs->cx + *pt_regs_nr(regs, reg);
195	return ex_handler_uaccess(fixup, regs, trapnr);
196}
197
198int ex_get_fixup_type(unsigned long ip)
199{
200	const struct exception_table_entry *e = search_exception_tables(ip);
 
201
202	return e ? FIELD_GET(EX_DATA_TYPE_MASK, e->data) : EX_TYPE_NONE;
203}
204
205int fixup_exception(struct pt_regs *regs, int trapnr, unsigned long error_code,
206		    unsigned long fault_addr)
207{
208	const struct exception_table_entry *e;
209	int type, reg, imm;
210
211#ifdef CONFIG_PNPBIOS
212	if (unlikely(SEGMENT_IS_PNP_CODE(regs->cs))) {
213		extern u32 pnp_bios_fault_eip, pnp_bios_fault_esp;
214		extern u32 pnp_bios_is_utter_crap;
215		pnp_bios_is_utter_crap = 1;
216		printk(KERN_CRIT "PNPBIOS fault.. attempting recovery.\n");
217		__asm__ volatile(
218			"movl %0, %%esp\n\t"
219			"jmp *%1\n\t"
220			: : "g" (pnp_bios_fault_esp), "g" (pnp_bios_fault_eip));
221		panic("do_trap: can't hit this");
222	}
223#endif
224
225	e = search_exception_tables(regs->ip);
226	if (!e)
227		return 0;
228
229	type = FIELD_GET(EX_DATA_TYPE_MASK, e->data);
230	reg  = FIELD_GET(EX_DATA_REG_MASK,  e->data);
231	imm  = FIELD_GET(EX_DATA_IMM_MASK,  e->data);
232
233	switch (type) {
234	case EX_TYPE_DEFAULT:
235	case EX_TYPE_DEFAULT_MCE_SAFE:
236		return ex_handler_default(e, regs);
237	case EX_TYPE_FAULT:
238	case EX_TYPE_FAULT_MCE_SAFE:
239		return ex_handler_fault(e, regs, trapnr);
240	case EX_TYPE_UACCESS:
241		return ex_handler_uaccess(e, regs, trapnr);
242	case EX_TYPE_COPY:
243		return ex_handler_copy(e, regs, trapnr);
244	case EX_TYPE_CLEAR_FS:
245		return ex_handler_clear_fs(e, regs);
246	case EX_TYPE_FPU_RESTORE:
247		return ex_handler_fprestore(e, regs);
248	case EX_TYPE_BPF:
249		return ex_handler_bpf(e, regs);
250	case EX_TYPE_WRMSR:
251		return ex_handler_msr(e, regs, true, false, reg);
252	case EX_TYPE_RDMSR:
253		return ex_handler_msr(e, regs, false, false, reg);
254	case EX_TYPE_WRMSR_SAFE:
255		return ex_handler_msr(e, regs, true, true, reg);
256	case EX_TYPE_RDMSR_SAFE:
257		return ex_handler_msr(e, regs, false, true, reg);
258	case EX_TYPE_WRMSR_IN_MCE:
259		ex_handler_msr_mce(regs, true);
260		break;
261	case EX_TYPE_RDMSR_IN_MCE:
262		ex_handler_msr_mce(regs, false);
263		break;
264	case EX_TYPE_POP_REG:
265		regs->sp += sizeof(long);
266		fallthrough;
267	case EX_TYPE_IMM_REG:
268		return ex_handler_imm_reg(e, regs, reg, imm);
269	case EX_TYPE_FAULT_SGX:
270		return ex_handler_sgx(e, regs, trapnr);
271	case EX_TYPE_UCOPY_LEN:
272		return ex_handler_ucopy_len(e, regs, trapnr, reg, imm);
273	case EX_TYPE_ZEROPAD:
274		return ex_handler_zeropad(e, regs, fault_addr);
275	}
276	BUG();
277}
278
279extern unsigned int early_recursion_flag;
280
281/* Restricted version used during very early boot */
282void __init early_fixup_exception(struct pt_regs *regs, int trapnr)
283{
284	/* Ignore early NMIs. */
285	if (trapnr == X86_TRAP_NMI)
286		return;
287
288	if (early_recursion_flag > 2)
289		goto halt_loop;
290
291	/*
292	 * Old CPUs leave the high bits of CS on the stack
293	 * undefined.  I'm not sure which CPUs do this, but at least
294	 * the 486 DX works this way.
295	 * Xen pv domains are not using the default __KERNEL_CS.
296	 */
297	if (!xen_pv_domain() && regs->cs != __KERNEL_CS)
298		goto fail;
299
300	/*
301	 * The full exception fixup machinery is available as soon as
302	 * the early IDT is loaded.  This means that it is the
303	 * responsibility of extable users to either function correctly
304	 * when handlers are invoked early or to simply avoid causing
305	 * exceptions before they're ready to handle them.
306	 *
307	 * This is better than filtering which handlers can be used,
308	 * because refusing to call a handler here is guaranteed to
309	 * result in a hard-to-debug panic.
310	 *
311	 * Keep in mind that not all vectors actually get here.  Early
312	 * page faults, for example, are special.
313	 */
314	if (fixup_exception(regs, trapnr, regs->orig_ax, 0))
315		return;
316
317	if (trapnr == X86_TRAP_UD) {
318		if (report_bug(regs->ip, regs) == BUG_TRAP_TYPE_WARN) {
319			/* Skip the ud2. */
320			regs->ip += LEN_UD2;
321			return;
322		}
323
324		/*
325		 * If this was a BUG and report_bug returns or if this
326		 * was just a normal #UD, we want to continue onward and
327		 * crash.
328		 */
329	}
330
331fail:
332	early_printk("PANIC: early exception 0x%02x IP %lx:%lx error %lx cr2 0x%lx\n",
333		     (unsigned)trapnr, (unsigned long)regs->cs, regs->ip,
334		     regs->orig_ax, read_cr2());
335
336	show_regs(regs);
337
338halt_loop:
339	while (true)
340		halt();
341}
v4.17
 
  1#include <linux/extable.h>
  2#include <linux/uaccess.h>
  3#include <linux/sched/debug.h>
 
  4#include <xen/xen.h>
  5
  6#include <asm/fpu/internal.h>
 
  7#include <asm/traps.h>
  8#include <asm/kdebug.h>
 
 
  9
 10typedef bool (*ex_handler_t)(const struct exception_table_entry *,
 11			    struct pt_regs *, int);
 
 
 
 
 
 
 
 
 12
 13static inline unsigned long
 14ex_fixup_addr(const struct exception_table_entry *x)
 15{
 16	return (unsigned long)&x->fixup + x->fixup;
 17}
 18static inline ex_handler_t
 19ex_fixup_handler(const struct exception_table_entry *x)
 20{
 21	return (ex_handler_t)((unsigned long)&x->handler + x->handler);
 22}
 23
 24__visible bool ex_handler_default(const struct exception_table_entry *fixup,
 25				  struct pt_regs *regs, int trapnr)
 26{
 27	regs->ip = ex_fixup_addr(fixup);
 28	return true;
 29}
 30EXPORT_SYMBOL(ex_handler_default);
 31
 32__visible bool ex_handler_fault(const struct exception_table_entry *fixup,
 33				struct pt_regs *regs, int trapnr)
 34{
 35	regs->ip = ex_fixup_addr(fixup);
 36	regs->ax = trapnr;
 37	return true;
 38}
 39EXPORT_SYMBOL_GPL(ex_handler_fault);
 40
 41/*
 42 * Handler for UD0 exception following a failed test against the
 43 * result of a refcount inc/dec/add/sub.
 
 
 
 
 
 
 
 
 44 */
 45__visible bool ex_handler_refcount(const struct exception_table_entry *fixup,
 46				   struct pt_regs *regs, int trapnr)
 47{
 48	/* First unconditionally saturate the refcount. */
 49	*(int *)regs->cx = INT_MIN / 2;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 50
 51	/*
 52	 * Strictly speaking, this reports the fixup destination, not
 53	 * the fault location, and not the actually overflowing
 54	 * instruction, which is the instruction before the "js", but
 55	 * since that instruction could be a variety of lengths, just
 56	 * report the location after the overflow, which should be close
 57	 * enough for finding the overflow, as it's at least back in
 58	 * the function, having returned from .text.unlikely.
 59	 */
 60	regs->ip = ex_fixup_addr(fixup);
 61
 62	/*
 63	 * This function has been called because either a negative refcount
 64	 * value was seen by any of the refcount functions, or a zero
 65	 * refcount value was seen by refcount_dec().
 66	 *
 67	 * If we crossed from INT_MAX to INT_MIN, OF (Overflow Flag: result
 68	 * wrapped around) will be set. Additionally, seeing the refcount
 69	 * reach 0 will set ZF (Zero Flag: result was zero). In each of
 70	 * these cases we want a report, since it's a boundary condition.
 71	 * The SF case is not reported since it indicates post-boundary
 72	 * manipulations below zero or above INT_MAX. And if none of the
 73	 * flags are set, something has gone very wrong, so report it.
 74	 */
 75	if (regs->flags & (X86_EFLAGS_OF | X86_EFLAGS_ZF)) {
 76		bool zero = regs->flags & X86_EFLAGS_ZF;
 77
 78		refcount_error_report(regs, zero ? "hit zero" : "overflow");
 79	} else if ((regs->flags & X86_EFLAGS_SF) == 0) {
 80		/* Report if none of OF, ZF, nor SF are set. */
 81		refcount_error_report(regs, "unexpected saturation");
 82	}
 
 83
 84	return true;
 
 
 
 
 85}
 86EXPORT_SYMBOL(ex_handler_refcount);
 87
 88/*
 89 * Handler for when we fail to restore a task's FPU state.  We should never get
 90 * here because the FPU state of a task using the FPU (task->thread.fpu.state)
 91 * should always be valid.  However, past bugs have allowed userspace to set
 92 * reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn().
 93 * These caused XRSTOR to fail when switching to the task, leaking the FPU
 94 * registers of the task previously executing on the CPU.  Mitigate this class
 95 * of vulnerability by restoring from the initial state (essentially, zeroing
 96 * out all the FPU registers) if we can't restore from the task's FPU state.
 97 */
 98__visible bool ex_handler_fprestore(const struct exception_table_entry *fixup,
 99				    struct pt_regs *regs, int trapnr)
100{
101	regs->ip = ex_fixup_addr(fixup);
102
103	WARN_ONCE(1, "Bad FPU state detected at %pB, reinitializing FPU registers.",
104		  (void *)instruction_pointer(regs));
105
106	__copy_kernel_to_fpregs(&init_fpstate, -1);
107	return true;
108}
109EXPORT_SYMBOL_GPL(ex_handler_fprestore);
110
111__visible bool ex_handler_ext(const struct exception_table_entry *fixup,
112			      struct pt_regs *regs, int trapnr)
113{
114	/* Special hack for uaccess_err */
115	current->thread.uaccess_err = 1;
116	regs->ip = ex_fixup_addr(fixup);
117	return true;
118}
119EXPORT_SYMBOL(ex_handler_ext);
120
121__visible bool ex_handler_rdmsr_unsafe(const struct exception_table_entry *fixup,
122				       struct pt_regs *regs, int trapnr)
123{
124	if (pr_warn_once("unchecked MSR access error: RDMSR from 0x%x at rIP: 0x%lx (%pF)\n",
125			 (unsigned int)regs->cx, regs->ip, (void *)regs->ip))
126		show_stack_regs(regs);
127
128	/* Pretend that the read succeeded and returned 0. */
129	regs->ip = ex_fixup_addr(fixup);
130	regs->ax = 0;
131	regs->dx = 0;
132	return true;
133}
134EXPORT_SYMBOL(ex_handler_rdmsr_unsafe);
135
136__visible bool ex_handler_wrmsr_unsafe(const struct exception_table_entry *fixup,
137				       struct pt_regs *regs, int trapnr)
138{
139	if (pr_warn_once("unchecked MSR access error: WRMSR to 0x%x (tried to write 0x%08x%08x) at rIP: 0x%lx (%pF)\n",
140			 (unsigned int)regs->cx, (unsigned int)regs->dx,
141			 (unsigned int)regs->ax,  regs->ip, (void *)regs->ip))
 
142		show_stack_regs(regs);
 
 
 
 
 
 
 
143
144	/* Pretend that the write succeeded. */
145	regs->ip = ex_fixup_addr(fixup);
146	return true;
 
 
 
 
 
 
 
147}
148EXPORT_SYMBOL(ex_handler_wrmsr_unsafe);
149
150__visible bool ex_handler_clear_fs(const struct exception_table_entry *fixup,
151				   struct pt_regs *regs, int trapnr)
152{
153	if (static_cpu_has(X86_BUG_NULL_SEG))
154		asm volatile ("mov %0, %%fs" : : "rm" (__USER_DS));
155	asm volatile ("mov %0, %%fs" : : "rm" (0));
156	return ex_handler_default(fixup, regs, trapnr);
 
 
 
 
 
 
 
157}
158EXPORT_SYMBOL(ex_handler_clear_fs);
159
160__visible bool ex_has_fault_handler(unsigned long ip)
 
161{
162	const struct exception_table_entry *e;
163	ex_handler_t handler;
 
164
165	e = search_exception_tables(ip);
166	if (!e)
167		return false;
168	handler = ex_fixup_handler(e);
169
170	return handler == ex_handler_fault;
171}
172
173int fixup_exception(struct pt_regs *regs, int trapnr)
 
174{
175	const struct exception_table_entry *e;
176	ex_handler_t handler;
177
178#ifdef CONFIG_PNPBIOS
179	if (unlikely(SEGMENT_IS_PNP_CODE(regs->cs))) {
180		extern u32 pnp_bios_fault_eip, pnp_bios_fault_esp;
181		extern u32 pnp_bios_is_utter_crap;
182		pnp_bios_is_utter_crap = 1;
183		printk(KERN_CRIT "PNPBIOS fault.. attempting recovery.\n");
184		__asm__ volatile(
185			"movl %0, %%esp\n\t"
186			"jmp *%1\n\t"
187			: : "g" (pnp_bios_fault_esp), "g" (pnp_bios_fault_eip));
188		panic("do_trap: can't hit this");
189	}
190#endif
191
192	e = search_exception_tables(regs->ip);
193	if (!e)
194		return 0;
195
196	handler = ex_fixup_handler(e);
197	return handler(e, regs, trapnr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
198}
199
200extern unsigned int early_recursion_flag;
201
202/* Restricted version used during very early boot */
203void __init early_fixup_exception(struct pt_regs *regs, int trapnr)
204{
205	/* Ignore early NMIs. */
206	if (trapnr == X86_TRAP_NMI)
207		return;
208
209	if (early_recursion_flag > 2)
210		goto halt_loop;
211
212	/*
213	 * Old CPUs leave the high bits of CS on the stack
214	 * undefined.  I'm not sure which CPUs do this, but at least
215	 * the 486 DX works this way.
216	 * Xen pv domains are not using the default __KERNEL_CS.
217	 */
218	if (!xen_pv_domain() && regs->cs != __KERNEL_CS)
219		goto fail;
220
221	/*
222	 * The full exception fixup machinery is available as soon as
223	 * the early IDT is loaded.  This means that it is the
224	 * responsibility of extable users to either function correctly
225	 * when handlers are invoked early or to simply avoid causing
226	 * exceptions before they're ready to handle them.
227	 *
228	 * This is better than filtering which handlers can be used,
229	 * because refusing to call a handler here is guaranteed to
230	 * result in a hard-to-debug panic.
231	 *
232	 * Keep in mind that not all vectors actually get here.  Early
233	 * fage faults, for example, are special.
234	 */
235	if (fixup_exception(regs, trapnr))
236		return;
237
238	if (fixup_bug(regs, trapnr))
239		return;
 
 
 
 
 
 
 
 
 
 
 
240
241fail:
242	early_printk("PANIC: early exception 0x%02x IP %lx:%lx error %lx cr2 0x%lx\n",
243		     (unsigned)trapnr, (unsigned long)regs->cs, regs->ip,
244		     regs->orig_ax, read_cr2());
245
246	show_regs(regs);
247
248halt_loop:
249	while (true)
250		halt();
251}