arch/um/kernel/trap.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
   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, -EACCES, -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         p4d_t *p4d;
  32         pud_t *pud;
  33         pmd_t *pmd;
  34         pte_t *pte;
  35         int err = -EFAULT;
  36         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
  37
  38         *code_out = SEGV_MAPERR;
  39
  40         /*
  41          * If the fault was with pagefaults disabled, don't take the fault, just
  42          * fail.
  43          */
  44         if (faulthandler_disabled())
  45                 goto out_nosemaphore;
  46
  47         if (is_user)
  48                 flags |= FAULT_FLAG_USER;
  49 retry:
  50         down_read(&mm->mmap_sem);
  51         vma = find_vma(mm, address);
  52         if (!vma)
  53                 goto out;
  54         else if (vma->vm_start <= address)
  55                 goto good_area;
  56         else if (!(vma->vm_flags & VM_GROWSDOWN))
  57                 goto out;
  58         else if (is_user && !ARCH_IS_STACKGROW(address))
  59                 goto out;
  60         else if (expand_stack(vma, address))
  61                 goto out;
  62
  63 good_area:
  64         *code_out = SEGV_ACCERR;
  65         if (is_write) {
  66                 if (!(vma->vm_flags & VM_WRITE))
  67                         goto out;
  68                 flags |= FAULT_FLAG_WRITE;
  69         } else {
  70                 /* Don't require VM_READ|VM_EXEC for write faults! */
  71                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
  72                         goto out;
  73         }
  74
  75         do {
  76                 vm_fault_t fault;
  77
  78                 fault = handle_mm_fault(vma, address, flags);
  79
  80                 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
  81                         goto out_nosemaphore;
  82
  83                 if (unlikely(fault & VM_FAULT_ERROR)) {
  84                         if (fault & VM_FAULT_OOM) {
  85                                 goto out_of_memory;
  86                         } else if (fault & VM_FAULT_SIGSEGV) {
  87                                 goto out;
  88                         } else if (fault & VM_FAULT_SIGBUS) {
  89                                 err = -EACCES;
  90                                 goto out;
  91                         }
  92                         BUG();
  93                 }
  94                 if (flags & FAULT_FLAG_ALLOW_RETRY) {
  95                         if (fault & VM_FAULT_MAJOR)
  96                                 current->maj_flt++;
  97                         else
  98                                 current->min_flt++;
  99                         if (fault & VM_FAULT_RETRY) {
 100                                 flags &= ~FAULT_FLAG_ALLOW_RETRY;
 101                                 flags |= FAULT_FLAG_TRIED;
 102
 103                                 goto retry;
 104                         }
 105                 }
 106
 107                 pgd = pgd_offset(mm, address);
 108                 p4d = p4d_offset(pgd, address);
 109                 pud = pud_offset(p4d, address);
 110                 pmd = pmd_offset(pud, address);
 111                 pte = pte_offset_kernel(pmd, address);
 112         } while (!pte_present(*pte));
 113         err = 0;
 114         /*
 115          * The below warning was added in place of
 116          *      pte_mkyoung(); if (is_write) pte_mkdirty();
 117          * If it's triggered, we'd see normally a hang here (a clean pte is
 118          * marked read-only to emulate the dirty bit).
 119          * However, the generic code can mark a PTE writable but clean on a
 120          * concurrent read fault, triggering this harmlessly. So comment it out.
 121          */
 122 #if 0
 123         WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
 124 #endif
 125         flush_tlb_page(vma, address);
 126 out:
 127         up_read(&mm->mmap_sem);
 128 out_nosemaphore:
 129         return err;
 130
 131 out_of_memory:
 132         /*
 133          * We ran out of memory, call the OOM killer, and return the userspace
 134          * (which will retry the fault, or kill us if we got oom-killed).
 135          */
 136         up_read(&mm->mmap_sem);
 137         if (!is_user)
 138                 goto out_nosemaphore;
 139         pagefault_out_of_memory();
 140         return 0;
 141 }
 142 EXPORT_SYMBOL(handle_page_fault);
 143
 144 static void show_segv_info(struct uml_pt_regs *regs)
 145 {
 146         struct task_struct *tsk = current;
 147         struct faultinfo *fi = UPT_FAULTINFO(regs);
 148
 149         if (!unhandled_signal(tsk, SIGSEGV))
 150                 return;
 151
 152         if (!printk_ratelimit())
 153                 return;
 154
 155         printk("%s%s[%d]: segfault at %lx ip %px sp %px error %x",
 156                 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
 157                 tsk->comm, task_pid_nr(tsk), FAULT_ADDRESS(*fi),
 158                 (void *)UPT_IP(regs), (void *)UPT_SP(regs),
 159                 fi->error_code);
 160
 161         print_vma_addr(KERN_CONT " in ", UPT_IP(regs));
 162         printk(KERN_CONT "\n");
 163 }
 164
 165 static void bad_segv(struct faultinfo fi, unsigned long ip)
 166 {
 167         current->thread.arch.faultinfo = fi;
 168         force_sig_fault(SIGSEGV, SEGV_ACCERR, (void __user *) FAULT_ADDRESS(fi));
 169 }
 170
 171 void fatal_sigsegv(void)
 172 {
 173         force_sigsegv(SIGSEGV);
 174         do_signal(&current->thread.regs);
 175         /*
 176          * This is to tell gcc that we're not returning - do_signal
 177          * can, in general, return, but in this case, it's not, since
 178          * we just got a fatal SIGSEGV queued.
 179          */
 180         os_dump_core();
 181 }
 182
 183 /**
 184  * segv_handler() - the SIGSEGV handler
 185  * @sig:        the signal number
 186  * @unused_si:  the signal info struct; unused in this handler
 187  * @regs:       the ptrace register information
 188  *
 189  * The handler first extracts the faultinfo from the UML ptrace regs struct.
 190  * If the userfault did not happen in an UML userspace process, bad_segv is called.
 191  * Otherwise the signal did happen in a cloned userspace process, handle it.
 192  */
 193 void segv_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
 194 {
 195         struct faultinfo * fi = UPT_FAULTINFO(regs);
 196
 197         if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
 198                 show_segv_info(regs);
 199                 bad_segv(*fi, UPT_IP(regs));
 200                 return;
 201         }
 202         segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
 203 }
 204
 205 /*
 206  * We give a *copy* of the faultinfo in the regs to segv.
 207  * This must be done, since nesting SEGVs could overwrite
 208  * the info in the regs. A pointer to the info then would
 209  * give us bad data!
 210  */
 211 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
 212                    struct uml_pt_regs *regs)
 213 {
 214         jmp_buf *catcher;
 215         int si_code;
 216         int err;
 217         int is_write = FAULT_WRITE(fi);
 218         unsigned long address = FAULT_ADDRESS(fi);
 219
 220         if (!is_user && regs)
 221                 current->thread.segv_regs = container_of(regs, struct pt_regs, regs);
 222
 223         if (!is_user && (address >= start_vm) && (address < end_vm)) {
 224                 flush_tlb_kernel_vm();
 225                 goto out;
 226         }
 227         else if (current->mm == NULL) {
 228                 show_regs(container_of(regs, struct pt_regs, regs));
 229                 panic("Segfault with no mm");
 230         }
 231         else if (!is_user && address > PAGE_SIZE && address < TASK_SIZE) {
 232                 show_regs(container_of(regs, struct pt_regs, regs));
 233                 panic("Kernel tried to access user memory at addr 0x%lx, ip 0x%lx",
 234                        address, ip);
 235         }
 236
 237         if (SEGV_IS_FIXABLE(&fi))
 238                 err = handle_page_fault(address, ip, is_write, is_user,
 239                                         &si_code);
 240         else {
 241                 err = -EFAULT;
 242                 /*
 243                  * A thread accessed NULL, we get a fault, but CR2 is invalid.
 244                  * This code is used in __do_copy_from_user() of TT mode.
 245                  * XXX tt mode is gone, so maybe this isn't needed any more
 246                  */
 247                 address = 0;
 248         }
 249
 250         catcher = current->thread.fault_catcher;
 251         if (!err)
 252                 goto out;
 253         else if (catcher != NULL) {
 254                 current->thread.fault_addr = (void *) address;
 255                 UML_LONGJMP(catcher, 1);
 256         }
 257         else if (current->thread.fault_addr != NULL)
 258                 panic("fault_addr set but no fault catcher");
 259         else if (!is_user && arch_fixup(ip, regs))
 260                 goto out;
 261
 262         if (!is_user) {
 263                 show_regs(container_of(regs, struct pt_regs, regs));
 264                 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
 265                       address, ip);
 266         }
 267
 268         show_segv_info(regs);
 269
 270         if (err == -EACCES) {
 271                 current->thread.arch.faultinfo = fi;
 272                 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
 273         } else {
 274                 BUG_ON(err != -EFAULT);
 275                 current->thread.arch.faultinfo = fi;
 276                 force_sig_fault(SIGSEGV, si_code, (void __user *) address);
 277         }
 278
 279 out:
 280         if (regs)
 281                 current->thread.segv_regs = NULL;
 282
 283         return 0;
 284 }
 285
 286 void relay_signal(int sig, struct siginfo *si, struct uml_pt_regs *regs)
 287 {
 288         int code, err;
 289         if (!UPT_IS_USER(regs)) {
 290                 if (sig == SIGBUS)
 291                         printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
 292                                "mount likely just ran out of space\n");
 293                 panic("Kernel mode signal %d", sig);
 294         }
 295
 296         arch_examine_signal(sig, regs);
 297
 298         /* Is the signal layout for the signal known?
 299          * Signal data must be scrubbed to prevent information leaks.
 300          */
 301         code = si->si_code;
 302         err = si->si_errno;
 303         if ((err == 0) && (siginfo_layout(sig, code) == SIL_FAULT)) {
 304                 struct faultinfo *fi = UPT_FAULTINFO(regs);
 305                 current->thread.arch.faultinfo = *fi;
 306                 force_sig_fault(sig, code, (void __user *)FAULT_ADDRESS(*fi));
 307         } else {
 308                 printk(KERN_ERR "Attempted to relay unknown signal %d (si_code = %d) with errno %d\n",
 309                        sig, code, err);
 310                 force_sig(sig);
 311         }
 312 }
 313
 314 void bus_handler(int sig, struct siginfo *si, struct uml_pt_regs *regs)
 315 {
 316         if (current->thread.fault_catcher != NULL)
 317                 UML_LONGJMP(current->thread.fault_catcher, 1);
 318         else
 319                 relay_signal(sig, si, regs);
 320 }
 321
 322 void winch(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
 323 {
 324         do_IRQ(WINCH_IRQ, regs);
 325 }
 326
 327 void trap_init(void)
 328 {
 329 }