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