arch/um/kernel/trap.c

   1 /*
   2  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
   3  * Licensed under the GPL
   4  */
   5
   6 #include <linux/mm.h>
   7 #include <linux/sched/signal.h>
   8 #include <linux/hardirq.h>
   9 #include <linux/module.h>
  10 #include <linux/uaccess.h>
  11 #include <linux/sched/debug.h>
  12 #include <asm/current.h>
  13 #include <asm/pgtable.h>
  14 #include <asm/tlbflush.h>
  15 #include <arch.h>
  16 #include <as-layout.h>
  17 #include <kern_util.h>
  18 #include <os.h>
  19 #include <skas.h>
  20
  21 /*
  22  * Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by
  23  * segv().
  24  */
  25 int handle_page_fault(unsigned long address, unsigned long ip,
  26                       int is_write, int is_user, int *code_out)
  27 {
  28         struct mm_struct *mm = current->mm;
  29         struct vm_area_struct *vma;
  30         pgd_t *pgd;
  31         pud_t *pud;
  32         pmd_t *pmd;
  33         pte_t *pte;
  34         int err = -EFAULT;
  35         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
  36
  37         *code_out = SEGV_MAPERR;
  38
  39         /*
  40          * If the fault was with pagefaults disabled, don't take the fault, just
  41          * fail.
  42          */
  43         if (faulthandler_disabled())
  44                 goto out_nosemaphore;
  45
  46         if (is_user)
  47                 flags |= FAULT_FLAG_USER;
  48 retry:
  49         down_read(&mm->mmap_sem);
  50         vma = find_vma(mm, address);
  51         if (!vma)
  52                 goto out;
  53         else if (vma->vm_start <= address)
  54                 goto good_area;
  55         else if (!(vma->vm_flags & VM_GROWSDOWN))
  56                 goto out;
  57         else if (is_user && !ARCH_IS_STACKGROW(address))
  58                 goto out;
  59         else if (expand_stack(vma, address))
  60                 goto out;
  61
  62 good_area:
  63         *code_out = SEGV_ACCERR;
  64         if (is_write) {
  65                 if (!(vma->vm_flags & VM_WRITE))
  66                         goto out;
  67                 flags |= FAULT_FLAG_WRITE;
  68         } else {
  69                 /* Don't require VM_READ|VM_EXEC for write faults! */
  70                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
  71                         goto out;
  72         }
  73
  74         do {
  75                 int fault;
  76
  77                 fault = handle_mm_fault(vma, address, flags);
  78
  79                 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
  80                         goto out_nosemaphore;
  81
  82                 if (unlikely(fault & VM_FAULT_ERROR)) {
  83                         if (fault & VM_FAULT_OOM) {
  84                                 goto out_of_memory;
  85                         } else if (fault & VM_FAULT_SIGSEGV) {
  86                                 goto out;
  87                         } else if (fault & VM_FAULT_SIGBUS) {
  88                                 err = -EACCES;
  89                                 goto out;
  90                         }
  91                         BUG();
  92                 }
  93                 if (flags & FAULT_FLAG_ALLOW_RETRY) {
  94                         if (fault & VM_FAULT_MAJOR)
  95                                 current->maj_flt++;
  96                         else
  97                                 current->min_flt++;
  98                         if (fault & VM_FAULT_RETRY) {
  99                                 flags &= ~FAULT_FLAG_ALLOW_RETRY;
 100                                 flags |= FAULT_FLAG_TRIED;
 101
 102                                 goto retry;
 103                         }
 104                 }
 105
 106                 pgd = pgd_offset(mm, address);
 107                 pud = pud_offset(pgd, address);
 108                 pmd = pmd_offset(pud, address);
 109                 pte = pte_offset_kernel(pmd, address);
 110         } while (!pte_present(*pte));
 111         err = 0;
 112         /*
 113          * The below warning was added in place of
 114          *      pte_mkyoung(); if (is_write) pte_mkdirty();
 115          * If it's triggered, we'd see normally a hang here (a clean pte is
 116          * marked read-only to emulate the dirty bit).
 117          * However, the generic code can mark a PTE writable but clean on a
 118          * concurrent read fault, triggering this harmlessly. So comment it out.
 119          */
 120 #if 0
 121         WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
 122 #endif
 123         flush_tlb_page(vma, address);
 124 out:
 125         up_read(&mm->mmap_sem);
 126 out_nosemaphore:
 127         return err;
 128
 129 out_of_memory:
 130         /*
 131          * We ran out of memory, call the OOM killer, and return the userspace
 132          * (which will retry the fault, or kill us if we got oom-killed).
 133          */
 134         up_read(&mm->mmap_sem);
 135         if (!is_user)
 136                 goto out_nosemaphore;
 137         pagefault_out_of_memory();
 138         return 0;
 139 }
 140 EXPORT_SYMBOL(handle_page_fault);
 141
 142 static void show_segv_info(struct uml_pt_regs *regs)
 143 {
 144         struct task_struct *tsk = current;
 145         struct faultinfo *fi = UPT_FAULTINFO(regs);
 146
 147         if (!unhandled_signal(tsk, SIGSEGV))
 148                 return;
 149
 150         if (!printk_ratelimit())
 151                 return;
 152
 153         printk("%s%s[%d]: segfault at %lx ip %px sp %px error %x",
 154                 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
 155                 tsk->comm, task_pid_nr(tsk), FAULT_ADDRESS(*fi),
 156                 (void *)UPT_IP(regs), (void *)UPT_SP(regs),
 157                 fi->error_code);
 158
 159         print_vma_addr(KERN_CONT " in ", UPT_IP(regs));
 160         printk(KERN_CONT "\n");
 161 }
 162
 163 static void bad_segv(struct faultinfo fi, unsigned long ip)
 164 {
 165         struct siginfo si;
 166
 167         si.si_signo = SIGSEGV;
 168         si.si_code = SEGV_ACCERR;
 169         si.si_addr = (void __user *) FAULT_ADDRESS(fi);
 170         current->thread.arch.faultinfo = fi;
 171         force_sig_info(SIGSEGV, &si, current);
 172 }
 173
 174 void fatal_sigsegv(void)
 175 {
 176         force_sigsegv(SIGSEGV, current);
 177         do_signal(&current->thread.regs);
 178         /*
 179          * This is to tell gcc that we're not returning - do_signal
 180          * can, in general, return, but in this case, it's not, since
 181          * we just got a fatal SIGSEGV queued.
 182          */
 183         os_dump_core();
 184 }
 185
 186 /**
 187  * segv_handler() - the SIGSEGV handler
 188  * @sig:        the signal number
 189  * @unused_si:  the signal info struct; unused in this handler
 190  * @regs:       the ptrace register information
 191  *
 192  * The handler first extracts the faultinfo from the UML ptrace regs struct.
 193  * If the userfault did not happen in an UML userspace process, bad_segv is called.
 194  * Otherwise the signal did happen in a cloned userspace process, handle it.
 195  */
 196 void segv_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
 197 {
 198         struct faultinfo * fi = UPT_FAULTINFO(regs);
 199
 200         if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
 201                 show_segv_info(regs);
 202                 bad_segv(*fi, UPT_IP(regs));
 203                 return;
 204         }
 205         segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
 206 }
 207
 208 /*
 209  * We give a *copy* of the faultinfo in the regs to segv.
 210  * This must be done, since nesting SEGVs could overwrite
 211  * the info in the regs. A pointer to the info then would
 212  * give us bad data!
 213  */
 214 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
 215                    struct uml_pt_regs *regs)
 216 {
 217         struct siginfo si;
 218         jmp_buf *catcher;
 219         int err;
 220         int is_write = FAULT_WRITE(fi);
 221         unsigned long address = FAULT_ADDRESS(fi);
 222
 223         if (!is_user && regs)
 224                 current->thread.segv_regs = container_of(regs, struct pt_regs, regs);
 225
 226         if (!is_user && (address >= start_vm) && (address < end_vm)) {
 227                 flush_tlb_kernel_vm();
 228                 goto out;
 229         }
 230         else if (current->mm == NULL) {
 231                 show_regs(container_of(regs, struct pt_regs, regs));
 232                 panic("Segfault with no mm");
 233         }
 234         else if (!is_user && address > PAGE_SIZE && address < TASK_SIZE) {
 235                 show_regs(container_of(regs, struct pt_regs, regs));
 236                 panic("Kernel tried to access user memory at addr 0x%lx, ip 0x%lx",
 237                        address, ip);
 238         }
 239
 240         if (SEGV_IS_FIXABLE(&fi))
 241                 err = handle_page_fault(address, ip, is_write, is_user,
 242                                         &si.si_code);
 243         else {
 244                 err = -EFAULT;
 245                 /*
 246                  * A thread accessed NULL, we get a fault, but CR2 is invalid.
 247                  * This code is used in __do_copy_from_user() of TT mode.
 248                  * XXX tt mode is gone, so maybe this isn't needed any more
 249                  */
 250                 address = 0;
 251         }
 252
 253         catcher = current->thread.fault_catcher;
 254         if (!err)
 255                 goto out;
 256         else if (catcher != NULL) {
 257                 current->thread.fault_addr = (void *) address;
 258                 UML_LONGJMP(catcher, 1);
 259         }
 260         else if (current->thread.fault_addr != NULL)
 261                 panic("fault_addr set but no fault catcher");
 262         else if (!is_user && arch_fixup(ip, regs))
 263                 goto out;
 264
 265         if (!is_user) {
 266                 show_regs(container_of(regs, struct pt_regs, regs));
 267                 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
 268                       address, ip);
 269         }
 270
 271         show_segv_info(regs);
 272
 273         if (err == -EACCES) {
 274                 si.si_signo = SIGBUS;
 275                 si.si_errno = 0;
 276                 si.si_code = BUS_ADRERR;
 277                 si.si_addr = (void __user *)address;
 278                 current->thread.arch.faultinfo = fi;
 279                 force_sig_info(SIGBUS, &si, current);
 280         } else {
 281                 BUG_ON(err != -EFAULT);
 282                 si.si_signo = SIGSEGV;
 283                 si.si_addr = (void __user *) address;
 284                 current->thread.arch.faultinfo = fi;
 285                 force_sig_info(SIGSEGV, &si, current);
 286         }
 287
 288 out:
 289         if (regs)
 290                 current->thread.segv_regs = NULL;
 291
 292         return 0;
 293 }
 294
 295 void relay_signal(int sig, struct siginfo *si, struct uml_pt_regs *regs)
 296 {
 297         struct faultinfo *fi;
 298         struct siginfo clean_si;
 299
 300         if (!UPT_IS_USER(regs)) {
 301                 if (sig == SIGBUS)
 302                         printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
 303                                "mount likely just ran out of space\n");
 304                 panic("Kernel mode signal %d", sig);
 305         }
 306
 307         arch_examine_signal(sig, regs);
 308
 309         clear_siginfo(&clean_si);
 310         clean_si.si_signo = si->si_signo;
 311         clean_si.si_errno = si->si_errno;
 312         clean_si.si_code = si->si_code;
 313         switch (sig) {
 314         case SIGILL:
 315         case SIGFPE:
 316         case SIGSEGV:
 317         case SIGBUS:
 318         case SIGTRAP:
 319                 fi = UPT_FAULTINFO(regs);
 320                 clean_si.si_addr = (void __user *) FAULT_ADDRESS(*fi);
 321                 current->thread.arch.faultinfo = *fi;
 322 #ifdef __ARCH_SI_TRAPNO
 323                 clean_si.si_trapno = si->si_trapno;
 324 #endif
 325                 break;
 326         default:
 327                 printk(KERN_ERR "Attempted to relay unknown signal %d (si_code = %d)\n",
 328                         sig, si->si_code);
 329         }
 330
 331         force_sig_info(sig, &clean_si, current);
 332 }
 333
 334 void bus_handler(int sig, struct siginfo *si, struct uml_pt_regs *regs)
 335 {
 336         if (current->thread.fault_catcher != NULL)
 337                 UML_LONGJMP(current->thread.fault_catcher, 1);
 338         else
 339                 relay_signal(sig, si, regs);
 340 }
 341
 342 void winch(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
 343 {
 344         do_IRQ(WINCH_IRQ, regs);
 345 }
 346
 347 void trap_init(void)
 348 {
 349 }