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 10 typedef bool (*ex_handler_t)(const struct exception_table_entry *, 11 struct pt_regs *, int, unsigned long, 12 unsigned long); 13 14 static inline unsigned long 15 ex_fixup_addr(const struct exception_table_entry *x) 16 { 17 return (unsigned long)&x->fixup + x->fixup; 18 } 19 static inline ex_handler_t 20 ex_fixup_handler(const struct exception_table_entry *x) 21 { 22 return (ex_handler_t)((unsigned long)&x->handler + x->handler); 23 } 24 25 __visible bool ex_handler_default(const struct exception_table_entry *fixup, 26 struct pt_regs *regs, int trapnr, 27 unsigned long error_code, 28 unsigned long fault_addr) 29 { 30 regs->ip = ex_fixup_addr(fixup); 31 return true; 32 } 33 EXPORT_SYMBOL(ex_handler_default); 34 35 __visible bool ex_handler_fault(const struct exception_table_entry *fixup, 36 struct pt_regs *regs, int trapnr, 37 unsigned long error_code, 38 unsigned long fault_addr) 39 { 40 regs->ip = ex_fixup_addr(fixup); 41 regs->ax = trapnr; 42 return true; 43 } 44 EXPORT_SYMBOL_GPL(ex_handler_fault); 45 46 /* 47 * Handler for UD0 exception following a failed test against the 48 * result of a refcount inc/dec/add/sub. 49 */ 50 __visible bool ex_handler_refcount(const struct exception_table_entry *fixup, 51 struct pt_regs *regs, int trapnr, 52 unsigned long error_code, 53 unsigned long fault_addr) 54 { 55 /* First unconditionally saturate the refcount. */ 56 *(int *)regs->cx = INT_MIN / 2; 57 58 /* 59 * Strictly speaking, this reports the fixup destination, not 60 * the fault location, and not the actually overflowing 61 * instruction, which is the instruction before the "js", but 62 * since that instruction could be a variety of lengths, just 63 * report the location after the overflow, which should be close 64 * enough for finding the overflow, as it's at least back in 65 * the function, having returned from .text.unlikely. 66 */ 67 regs->ip = ex_fixup_addr(fixup); 68 69 /* 70 * This function has been called because either a negative refcount 71 * value was seen by any of the refcount functions, or a zero 72 * refcount value was seen by refcount_dec(). 73 * 74 * If we crossed from INT_MAX to INT_MIN, OF (Overflow Flag: result 75 * wrapped around) will be set. Additionally, seeing the refcount 76 * reach 0 will set ZF (Zero Flag: result was zero). In each of 77 * these cases we want a report, since it's a boundary condition. 78 * The SF case is not reported since it indicates post-boundary 79 * manipulations below zero or above INT_MAX. And if none of the 80 * flags are set, something has gone very wrong, so report it. 81 */ 82 if (regs->flags & (X86_EFLAGS_OF | X86_EFLAGS_ZF)) { 83 bool zero = regs->flags & X86_EFLAGS_ZF; 84 85 refcount_error_report(regs, zero ? "hit zero" : "overflow"); 86 } else if ((regs->flags & X86_EFLAGS_SF) == 0) { 87 /* Report if none of OF, ZF, nor SF are set. */ 88 refcount_error_report(regs, "unexpected saturation"); 89 } 90 91 return true; 92 } 93 EXPORT_SYMBOL(ex_handler_refcount); 94 95 /* 96 * Handler for when we fail to restore a task's FPU state. We should never get 97 * here because the FPU state of a task using the FPU (task->thread.fpu.state) 98 * should always be valid. However, past bugs have allowed userspace to set 99 * reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn(). 100 * These caused XRSTOR to fail when switching to the task, leaking the FPU 101 * registers of the task previously executing on the CPU. Mitigate this class 102 * of vulnerability by restoring from the initial state (essentially, zeroing 103 * out all the FPU registers) if we can't restore from the task's FPU state. 104 */ 105 __visible bool ex_handler_fprestore(const struct exception_table_entry *fixup, 106 struct pt_regs *regs, int trapnr, 107 unsigned long error_code, 108 unsigned long fault_addr) 109 { 110 regs->ip = ex_fixup_addr(fixup); 111 112 WARN_ONCE(1, "Bad FPU state detected at %pB, reinitializing FPU registers.", 113 (void *)instruction_pointer(regs)); 114 115 __copy_kernel_to_fpregs(&init_fpstate, -1); 116 return true; 117 } 118 EXPORT_SYMBOL_GPL(ex_handler_fprestore); 119 120 /* Helper to check whether a uaccess fault indicates a kernel bug. */ 121 static bool bogus_uaccess(struct pt_regs *regs, int trapnr, 122 unsigned long fault_addr) 123 { 124 /* This is the normal case: #PF with a fault address in userspace. */ 125 if (trapnr == X86_TRAP_PF && fault_addr < TASK_SIZE_MAX) 126 return false; 127 128 /* 129 * This code can be reached for machine checks, but only if the #MC 130 * handler has already decided that it looks like a candidate for fixup. 131 * This e.g. happens when attempting to access userspace memory which 132 * the CPU can't access because of uncorrectable bad memory. 133 */ 134 if (trapnr == X86_TRAP_MC) 135 return false; 136 137 /* 138 * There are two remaining exception types we might encounter here: 139 * - #PF for faulting accesses to kernel addresses 140 * - #GP for faulting accesses to noncanonical addresses 141 * Complain about anything else. 142 */ 143 if (trapnr != X86_TRAP_PF && trapnr != X86_TRAP_GP) { 144 WARN(1, "unexpected trap %d in uaccess\n", trapnr); 145 return false; 146 } 147 148 /* 149 * This is a faulting memory access in kernel space, on a kernel 150 * address, in a usercopy function. This can e.g. be caused by improper 151 * use of helpers like __put_user and by improper attempts to access 152 * userspace addresses in KERNEL_DS regions. 153 * The one (semi-)legitimate exception are probe_kernel_{read,write}(), 154 * which can be invoked from places like kgdb, /dev/mem (for reading) 155 * and privileged BPF code (for reading). 156 * The probe_kernel_*() functions set the kernel_uaccess_faults_ok flag 157 * to tell us that faulting on kernel addresses, and even noncanonical 158 * addresses, in a userspace accessor does not necessarily imply a 159 * kernel bug, root might just be doing weird stuff. 160 */ 161 if (current->kernel_uaccess_faults_ok) 162 return false; 163 164 /* This is bad. Refuse the fixup so that we go into die(). */ 165 if (trapnr == X86_TRAP_PF) { 166 pr_emerg("BUG: pagefault on kernel address 0x%lx in non-whitelisted uaccess\n", 167 fault_addr); 168 } else { 169 pr_emerg("BUG: GPF in non-whitelisted uaccess (non-canonical address?)\n"); 170 } 171 return true; 172 } 173 174 __visible bool ex_handler_uaccess(const struct exception_table_entry *fixup, 175 struct pt_regs *regs, int trapnr, 176 unsigned long error_code, 177 unsigned long fault_addr) 178 { 179 if (bogus_uaccess(regs, trapnr, fault_addr)) 180 return false; 181 regs->ip = ex_fixup_addr(fixup); 182 return true; 183 } 184 EXPORT_SYMBOL(ex_handler_uaccess); 185 186 __visible bool ex_handler_ext(const struct exception_table_entry *fixup, 187 struct pt_regs *regs, int trapnr, 188 unsigned long error_code, 189 unsigned long fault_addr) 190 { 191 if (bogus_uaccess(regs, trapnr, fault_addr)) 192 return false; 193 /* Special hack for uaccess_err */ 194 current->thread.uaccess_err = 1; 195 regs->ip = ex_fixup_addr(fixup); 196 return true; 197 } 198 EXPORT_SYMBOL(ex_handler_ext); 199 200 __visible bool ex_handler_rdmsr_unsafe(const struct exception_table_entry *fixup, 201 struct pt_regs *regs, int trapnr, 202 unsigned long error_code, 203 unsigned long fault_addr) 204 { 205 if (pr_warn_once("unchecked MSR access error: RDMSR from 0x%x at rIP: 0x%lx (%pF)\n", 206 (unsigned int)regs->cx, regs->ip, (void *)regs->ip)) 207 show_stack_regs(regs); 208 209 /* Pretend that the read succeeded and returned 0. */ 210 regs->ip = ex_fixup_addr(fixup); 211 regs->ax = 0; 212 regs->dx = 0; 213 return true; 214 } 215 EXPORT_SYMBOL(ex_handler_rdmsr_unsafe); 216 217 __visible bool ex_handler_wrmsr_unsafe(const struct exception_table_entry *fixup, 218 struct pt_regs *regs, int trapnr, 219 unsigned long error_code, 220 unsigned long fault_addr) 221 { 222 if (pr_warn_once("unchecked MSR access error: WRMSR to 0x%x (tried to write 0x%08x%08x) at rIP: 0x%lx (%pF)\n", 223 (unsigned int)regs->cx, (unsigned int)regs->dx, 224 (unsigned int)regs->ax, regs->ip, (void *)regs->ip)) 225 show_stack_regs(regs); 226 227 /* Pretend that the write succeeded. */ 228 regs->ip = ex_fixup_addr(fixup); 229 return true; 230 } 231 EXPORT_SYMBOL(ex_handler_wrmsr_unsafe); 232 233 __visible bool ex_handler_clear_fs(const struct exception_table_entry *fixup, 234 struct pt_regs *regs, int trapnr, 235 unsigned long error_code, 236 unsigned long fault_addr) 237 { 238 if (static_cpu_has(X86_BUG_NULL_SEG)) 239 asm volatile ("mov %0, %%fs" : : "rm" (__USER_DS)); 240 asm volatile ("mov %0, %%fs" : : "rm" (0)); 241 return ex_handler_default(fixup, regs, trapnr, error_code, fault_addr); 242 } 243 EXPORT_SYMBOL(ex_handler_clear_fs); 244 245 __visible bool ex_has_fault_handler(unsigned long ip) 246 { 247 const struct exception_table_entry *e; 248 ex_handler_t handler; 249 250 e = search_exception_tables(ip); 251 if (!e) 252 return false; 253 handler = ex_fixup_handler(e); 254 255 return handler == ex_handler_fault; 256 } 257 258 int fixup_exception(struct pt_regs *regs, int trapnr, unsigned long error_code, 259 unsigned long fault_addr) 260 { 261 const struct exception_table_entry *e; 262 ex_handler_t handler; 263 264 #ifdef CONFIG_PNPBIOS 265 if (unlikely(SEGMENT_IS_PNP_CODE(regs->cs))) { 266 extern u32 pnp_bios_fault_eip, pnp_bios_fault_esp; 267 extern u32 pnp_bios_is_utter_crap; 268 pnp_bios_is_utter_crap = 1; 269 printk(KERN_CRIT "PNPBIOS fault.. attempting recovery.\n"); 270 __asm__ volatile( 271 "movl %0, %%esp\n\t" 272 "jmp *%1\n\t" 273 : : "g" (pnp_bios_fault_esp), "g" (pnp_bios_fault_eip)); 274 panic("do_trap: can't hit this"); 275 } 276 #endif 277 278 e = search_exception_tables(regs->ip); 279 if (!e) 280 return 0; 281 282 handler = ex_fixup_handler(e); 283 return handler(e, regs, trapnr, error_code, fault_addr); 284 } 285 286 extern unsigned int early_recursion_flag; 287 288 /* Restricted version used during very early boot */ 289 void __init early_fixup_exception(struct pt_regs *regs, int trapnr) 290 { 291 /* Ignore early NMIs. */ 292 if (trapnr == X86_TRAP_NMI) 293 return; 294 295 if (early_recursion_flag > 2) 296 goto halt_loop; 297 298 /* 299 * Old CPUs leave the high bits of CS on the stack 300 * undefined. I'm not sure which CPUs do this, but at least 301 * the 486 DX works this way. 302 * Xen pv domains are not using the default __KERNEL_CS. 303 */ 304 if (!xen_pv_domain() && regs->cs != __KERNEL_CS) 305 goto fail; 306 307 /* 308 * The full exception fixup machinery is available as soon as 309 * the early IDT is loaded. This means that it is the 310 * responsibility of extable users to either function correctly 311 * when handlers are invoked early or to simply avoid causing 312 * exceptions before they're ready to handle them. 313 * 314 * This is better than filtering which handlers can be used, 315 * because refusing to call a handler here is guaranteed to 316 * result in a hard-to-debug panic. 317 * 318 * Keep in mind that not all vectors actually get here. Early 319 * page faults, for example, are special. 320 */ 321 if (fixup_exception(regs, trapnr, regs->orig_ax, 0)) 322 return; 323 324 if (fixup_bug(regs, trapnr)) 325 return; 326 327 fail: 328 early_printk("PANIC: early exception 0x%02x IP %lx:%lx error %lx cr2 0x%lx\n", 329 (unsigned)trapnr, (unsigned long)regs->cs, regs->ip, 330 regs->orig_ax, read_cr2()); 331 332 show_regs(regs); 333 334 halt_loop: 335 while (true) 336 halt(); 337 } 338