1 /* 2 * This file is subject to the terms and conditions of the GNU General Public 3 * License. See the file "COPYING" in the main directory of this archive 4 * for more details. 5 * 6 * Copyright (C) 1995 - 2000 by Ralf Baechle 7 */ 8 #include <linux/context_tracking.h> 9 #include <linux/signal.h> 10 #include <linux/sched.h> 11 #include <linux/interrupt.h> 12 #include <linux/kernel.h> 13 #include <linux/errno.h> 14 #include <linux/string.h> 15 #include <linux/types.h> 16 #include <linux/ptrace.h> 17 #include <linux/ratelimit.h> 18 #include <linux/mman.h> 19 #include <linux/mm.h> 20 #include <linux/smp.h> 21 #include <linux/kprobes.h> 22 #include <linux/perf_event.h> 23 #include <linux/uaccess.h> 24 25 #include <asm/branch.h> 26 #include <asm/mmu_context.h> 27 #include <asm/ptrace.h> 28 #include <asm/highmem.h> /* For VMALLOC_END */ 29 #include <linux/kdebug.h> 30 31 int show_unhandled_signals = 1; 32 33 /* 34 * This routine handles page faults. It determines the address, 35 * and the problem, and then passes it off to one of the appropriate 36 * routines. 37 */ 38 static void __kprobes __do_page_fault(struct pt_regs *regs, unsigned long write, 39 unsigned long address) 40 { 41 struct vm_area_struct * vma = NULL; 42 struct task_struct *tsk = current; 43 struct mm_struct *mm = tsk->mm; 44 const int field = sizeof(unsigned long) * 2; 45 int si_code; 46 vm_fault_t fault; 47 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; 48 49 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10); 50 51 #if 0 52 printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_smp_processor_id(), 53 current->comm, current->pid, field, address, write, 54 field, regs->cp0_epc); 55 #endif 56 57 #ifdef CONFIG_KPROBES 58 /* 59 * This is to notify the fault handler of the kprobes. 60 */ 61 if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1, 62 current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP) 63 return; 64 #endif 65 66 si_code = SEGV_MAPERR; 67 68 /* 69 * We fault-in kernel-space virtual memory on-demand. The 70 * 'reference' page table is init_mm.pgd. 71 * 72 * NOTE! We MUST NOT take any locks for this case. We may 73 * be in an interrupt or a critical region, and should 74 * only copy the information from the master page table, 75 * nothing more. 76 */ 77 #ifdef CONFIG_64BIT 78 # define VMALLOC_FAULT_TARGET no_context 79 #else 80 # define VMALLOC_FAULT_TARGET vmalloc_fault 81 #endif 82 83 if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END)) 84 goto VMALLOC_FAULT_TARGET; 85 #ifdef MODULE_START 86 if (unlikely(address >= MODULE_START && address < MODULE_END)) 87 goto VMALLOC_FAULT_TARGET; 88 #endif 89 90 /* 91 * If we're in an interrupt or have no user 92 * context, we must not take the fault.. 93 */ 94 if (faulthandler_disabled() || !mm) 95 goto bad_area_nosemaphore; 96 97 if (user_mode(regs)) 98 flags |= FAULT_FLAG_USER; 99 retry: 100 down_read(&mm->mmap_sem); 101 vma = find_vma(mm, address); 102 if (!vma) 103 goto bad_area; 104 if (vma->vm_start <= address) 105 goto good_area; 106 if (!(vma->vm_flags & VM_GROWSDOWN)) 107 goto bad_area; 108 if (expand_stack(vma, address)) 109 goto bad_area; 110 /* 111 * Ok, we have a good vm_area for this memory access, so 112 * we can handle it.. 113 */ 114 good_area: 115 si_code = SEGV_ACCERR; 116 117 if (write) { 118 if (!(vma->vm_flags & VM_WRITE)) 119 goto bad_area; 120 flags |= FAULT_FLAG_WRITE; 121 } else { 122 if (cpu_has_rixi) { 123 if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) { 124 #if 0 125 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n", 126 raw_smp_processor_id(), 127 current->comm, current->pid, 128 field, address, write, 129 field, regs->cp0_epc); 130 #endif 131 goto bad_area; 132 } 133 if (!(vma->vm_flags & VM_READ) && 134 exception_epc(regs) != address) { 135 #if 0 136 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n", 137 raw_smp_processor_id(), 138 current->comm, current->pid, 139 field, address, write, 140 field, regs->cp0_epc); 141 #endif 142 goto bad_area; 143 } 144 } else { 145 if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))) 146 goto bad_area; 147 } 148 } 149 150 /* 151 * If for any reason at all we couldn't handle the fault, 152 * make sure we exit gracefully rather than endlessly redo 153 * the fault. 154 */ 155 fault = handle_mm_fault(vma, address, flags); 156 157 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) 158 return; 159 160 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 161 if (unlikely(fault & VM_FAULT_ERROR)) { 162 if (fault & VM_FAULT_OOM) 163 goto out_of_memory; 164 else if (fault & VM_FAULT_SIGSEGV) 165 goto bad_area; 166 else if (fault & VM_FAULT_SIGBUS) 167 goto do_sigbus; 168 BUG(); 169 } 170 if (flags & FAULT_FLAG_ALLOW_RETRY) { 171 if (fault & VM_FAULT_MAJOR) { 172 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 173 regs, address); 174 tsk->maj_flt++; 175 } else { 176 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 177 regs, address); 178 tsk->min_flt++; 179 } 180 if (fault & VM_FAULT_RETRY) { 181 flags &= ~FAULT_FLAG_ALLOW_RETRY; 182 flags |= FAULT_FLAG_TRIED; 183 184 /* 185 * No need to up_read(&mm->mmap_sem) as we would 186 * have already released it in __lock_page_or_retry 187 * in mm/filemap.c. 188 */ 189 190 goto retry; 191 } 192 } 193 194 up_read(&mm->mmap_sem); 195 return; 196 197 /* 198 * Something tried to access memory that isn't in our memory map.. 199 * Fix it, but check if it's kernel or user first.. 200 */ 201 bad_area: 202 up_read(&mm->mmap_sem); 203 204 bad_area_nosemaphore: 205 /* User mode accesses just cause a SIGSEGV */ 206 if (user_mode(regs)) { 207 tsk->thread.cp0_badvaddr = address; 208 tsk->thread.error_code = write; 209 if (show_unhandled_signals && 210 unhandled_signal(tsk, SIGSEGV) && 211 __ratelimit(&ratelimit_state)) { 212 pr_info("do_page_fault(): sending SIGSEGV to %s for invalid %s %0*lx\n", 213 tsk->comm, 214 write ? "write access to" : "read access from", 215 field, address); 216 pr_info("epc = %0*lx in", field, 217 (unsigned long) regs->cp0_epc); 218 print_vma_addr(KERN_CONT " ", regs->cp0_epc); 219 pr_cont("\n"); 220 pr_info("ra = %0*lx in", field, 221 (unsigned long) regs->regs[31]); 222 print_vma_addr(KERN_CONT " ", regs->regs[31]); 223 pr_cont("\n"); 224 } 225 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f; 226 force_sig_fault(SIGSEGV, si_code, (void __user *)address); 227 return; 228 } 229 230 no_context: 231 /* Are we prepared to handle this kernel fault? */ 232 if (fixup_exception(regs)) { 233 current->thread.cp0_baduaddr = address; 234 return; 235 } 236 237 /* 238 * Oops. The kernel tried to access some bad page. We'll have to 239 * terminate things with extreme prejudice. 240 */ 241 bust_spinlocks(1); 242 243 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at " 244 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n", 245 raw_smp_processor_id(), field, address, field, regs->cp0_epc, 246 field, regs->regs[31]); 247 die("Oops", regs); 248 249 out_of_memory: 250 /* 251 * We ran out of memory, call the OOM killer, and return the userspace 252 * (which will retry the fault, or kill us if we got oom-killed). 253 */ 254 up_read(&mm->mmap_sem); 255 if (!user_mode(regs)) 256 goto no_context; 257 pagefault_out_of_memory(); 258 return; 259 260 do_sigbus: 261 up_read(&mm->mmap_sem); 262 263 /* Kernel mode? Handle exceptions or die */ 264 if (!user_mode(regs)) 265 goto no_context; 266 267 /* 268 * Send a sigbus, regardless of whether we were in kernel 269 * or user mode. 270 */ 271 #if 0 272 printk("do_page_fault() #3: sending SIGBUS to %s for " 273 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n", 274 tsk->comm, 275 write ? "write access to" : "read access from", 276 field, address, 277 field, (unsigned long) regs->cp0_epc, 278 field, (unsigned long) regs->regs[31]); 279 #endif 280 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f; 281 tsk->thread.cp0_badvaddr = address; 282 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address); 283 284 return; 285 #ifndef CONFIG_64BIT 286 vmalloc_fault: 287 { 288 /* 289 * Synchronize this task's top level page-table 290 * with the 'reference' page table. 291 * 292 * Do _not_ use "tsk" here. We might be inside 293 * an interrupt in the middle of a task switch.. 294 */ 295 int offset = pgd_index(address); 296 pgd_t *pgd, *pgd_k; 297 p4d_t *p4d, *p4d_k; 298 pud_t *pud, *pud_k; 299 pmd_t *pmd, *pmd_k; 300 pte_t *pte_k; 301 302 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset; 303 pgd_k = init_mm.pgd + offset; 304 305 if (!pgd_present(*pgd_k)) 306 goto no_context; 307 set_pgd(pgd, *pgd_k); 308 309 p4d = p4d_offset(pgd, address); 310 p4d_k = p4d_offset(pgd_k, address); 311 if (!p4d_present(*p4d_k)) 312 goto no_context; 313 314 pud = pud_offset(p4d, address); 315 pud_k = pud_offset(p4d_k, address); 316 if (!pud_present(*pud_k)) 317 goto no_context; 318 319 pmd = pmd_offset(pud, address); 320 pmd_k = pmd_offset(pud_k, address); 321 if (!pmd_present(*pmd_k)) 322 goto no_context; 323 set_pmd(pmd, *pmd_k); 324 325 pte_k = pte_offset_kernel(pmd_k, address); 326 if (!pte_present(*pte_k)) 327 goto no_context; 328 return; 329 } 330 #endif 331 } 332 333 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs, 334 unsigned long write, unsigned long address) 335 { 336 enum ctx_state prev_state; 337 338 prev_state = exception_enter(); 339 __do_page_fault(regs, write, address); 340 exception_exit(prev_state); 341 } 342