1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
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>
15 #include <as-layout.h>
16 #include <kern_util.h>
21 * Note this is constrained to return 0, -EFAULT, -EACCES, -ENOMEM by
24 int handle_page_fault(unsigned long address, unsigned long ip,
25 int is_write, int is_user, int *code_out)
27 struct mm_struct *mm = current->mm;
28 struct vm_area_struct *vma;
32 unsigned int flags = FAULT_FLAG_DEFAULT;
34 *code_out = SEGV_MAPERR;
37 * If the fault was with pagefaults disabled, don't take the fault, just
40 if (faulthandler_disabled())
44 flags |= FAULT_FLAG_USER;
47 vma = find_vma(mm, address);
50 if (vma->vm_start <= address)
52 if (!(vma->vm_flags & VM_GROWSDOWN))
54 if (is_user && !ARCH_IS_STACKGROW(address))
56 vma = expand_stack(mm, address);
61 *code_out = SEGV_ACCERR;
63 if (!(vma->vm_flags & VM_WRITE))
65 flags |= FAULT_FLAG_WRITE;
67 /* Don't require VM_READ|VM_EXEC for write faults! */
68 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
75 fault = handle_mm_fault(vma, address, flags, NULL);
77 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
80 /* The fault is fully completed (including releasing mmap lock) */
81 if (fault & VM_FAULT_COMPLETED)
84 if (unlikely(fault & VM_FAULT_ERROR)) {
85 if (fault & VM_FAULT_OOM) {
87 } else if (fault & VM_FAULT_SIGSEGV) {
89 } else if (fault & VM_FAULT_SIGBUS) {
95 if (fault & VM_FAULT_RETRY) {
96 flags |= FAULT_FLAG_TRIED;
101 pmd = pmd_off(mm, address);
102 pte = pte_offset_kernel(pmd, address);
103 } while (!pte_present(*pte));
106 * The below warning was added in place of
107 * pte_mkyoung(); if (is_write) pte_mkdirty();
108 * If it's triggered, we'd see normally a hang here (a clean pte is
109 * marked read-only to emulate the dirty bit).
110 * However, the generic code can mark a PTE writable but clean on a
111 * concurrent read fault, triggering this harmlessly. So comment it out.
114 WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
116 flush_tlb_page(vma, address);
118 mmap_read_unlock(mm);
124 * We ran out of memory, call the OOM killer, and return the userspace
125 * (which will retry the fault, or kill us if we got oom-killed).
127 mmap_read_unlock(mm);
129 goto out_nosemaphore;
130 pagefault_out_of_memory();
134 static void show_segv_info(struct uml_pt_regs *regs)
136 struct task_struct *tsk = current;
137 struct faultinfo *fi = UPT_FAULTINFO(regs);
139 if (!unhandled_signal(tsk, SIGSEGV))
142 if (!printk_ratelimit())
145 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %x",
146 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
147 tsk->comm, task_pid_nr(tsk), FAULT_ADDRESS(*fi),
148 (void *)UPT_IP(regs), (void *)UPT_SP(regs),
151 print_vma_addr(KERN_CONT " in ", UPT_IP(regs));
152 printk(KERN_CONT "\n");
155 static void bad_segv(struct faultinfo fi, unsigned long ip)
157 current->thread.arch.faultinfo = fi;
158 force_sig_fault(SIGSEGV, SEGV_ACCERR, (void __user *) FAULT_ADDRESS(fi));
161 void fatal_sigsegv(void)
163 force_fatal_sig(SIGSEGV);
164 do_signal(¤t->thread.regs);
166 * This is to tell gcc that we're not returning - do_signal
167 * can, in general, return, but in this case, it's not, since
168 * we just got a fatal SIGSEGV queued.
174 * segv_handler() - the SIGSEGV handler
175 * @sig: the signal number
176 * @unused_si: the signal info struct; unused in this handler
177 * @regs: the ptrace register information
179 * The handler first extracts the faultinfo from the UML ptrace regs struct.
180 * If the userfault did not happen in an UML userspace process, bad_segv is called.
181 * Otherwise the signal did happen in a cloned userspace process, handle it.
183 void segv_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
185 struct faultinfo * fi = UPT_FAULTINFO(regs);
187 if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
188 show_segv_info(regs);
189 bad_segv(*fi, UPT_IP(regs));
192 segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
196 * We give a *copy* of the faultinfo in the regs to segv.
197 * This must be done, since nesting SEGVs could overwrite
198 * the info in the regs. A pointer to the info then would
201 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
202 struct uml_pt_regs *regs)
207 int is_write = FAULT_WRITE(fi);
208 unsigned long address = FAULT_ADDRESS(fi);
210 if (!is_user && regs)
211 current->thread.segv_regs = container_of(regs, struct pt_regs, regs);
213 if (!is_user && (address >= start_vm) && (address < end_vm)) {
214 flush_tlb_kernel_vm();
217 else if (current->mm == NULL) {
218 show_regs(container_of(regs, struct pt_regs, regs));
219 panic("Segfault with no mm");
221 else if (!is_user && address > PAGE_SIZE && address < TASK_SIZE) {
222 show_regs(container_of(regs, struct pt_regs, regs));
223 panic("Kernel tried to access user memory at addr 0x%lx, ip 0x%lx",
227 if (SEGV_IS_FIXABLE(&fi))
228 err = handle_page_fault(address, ip, is_write, is_user,
233 * A thread accessed NULL, we get a fault, but CR2 is invalid.
234 * This code is used in __do_copy_from_user() of TT mode.
235 * XXX tt mode is gone, so maybe this isn't needed any more
240 catcher = current->thread.fault_catcher;
243 else if (catcher != NULL) {
244 current->thread.fault_addr = (void *) address;
245 UML_LONGJMP(catcher, 1);
247 else if (current->thread.fault_addr != NULL)
248 panic("fault_addr set but no fault catcher");
249 else if (!is_user && arch_fixup(ip, regs))
253 show_regs(container_of(regs, struct pt_regs, regs));
254 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
258 show_segv_info(regs);
260 if (err == -EACCES) {
261 current->thread.arch.faultinfo = fi;
262 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
264 BUG_ON(err != -EFAULT);
265 current->thread.arch.faultinfo = fi;
266 force_sig_fault(SIGSEGV, si_code, (void __user *) address);
271 current->thread.segv_regs = NULL;
276 void relay_signal(int sig, struct siginfo *si, struct uml_pt_regs *regs)
279 if (!UPT_IS_USER(regs)) {
281 printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
282 "mount likely just ran out of space\n");
283 panic("Kernel mode signal %d", sig);
286 arch_examine_signal(sig, regs);
288 /* Is the signal layout for the signal known?
289 * Signal data must be scrubbed to prevent information leaks.
293 if ((err == 0) && (siginfo_layout(sig, code) == SIL_FAULT)) {
294 struct faultinfo *fi = UPT_FAULTINFO(regs);
295 current->thread.arch.faultinfo = *fi;
296 force_sig_fault(sig, code, (void __user *)FAULT_ADDRESS(*fi));
298 printk(KERN_ERR "Attempted to relay unknown signal %d (si_code = %d) with errno %d\n",
304 void bus_handler(int sig, struct siginfo *si, struct uml_pt_regs *regs)
306 if (current->thread.fault_catcher != NULL)
307 UML_LONGJMP(current->thread.fault_catcher, 1);
309 relay_signal(sig, si, regs);
312 void winch(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
314 do_IRQ(WINCH_IRQ, regs);