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 else if (vma->vm_start <= address)
52 else if (!(vma->vm_flags & VM_GROWSDOWN))
54 else if (is_user && !ARCH_IS_STACKGROW(address))
56 else if (expand_stack(vma, address))
60 *code_out = SEGV_ACCERR;
62 if (!(vma->vm_flags & VM_WRITE))
64 flags |= FAULT_FLAG_WRITE;
66 /* Don't require VM_READ|VM_EXEC for write faults! */
67 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
74 fault = handle_mm_fault(vma, address, flags, NULL);
76 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
79 if (unlikely(fault & VM_FAULT_ERROR)) {
80 if (fault & VM_FAULT_OOM) {
82 } else if (fault & VM_FAULT_SIGSEGV) {
84 } else if (fault & VM_FAULT_SIGBUS) {
90 if (fault & VM_FAULT_RETRY) {
91 flags |= FAULT_FLAG_TRIED;
96 pmd = pmd_off(mm, address);
97 pte = pte_offset_kernel(pmd, address);
98 } while (!pte_present(*pte));
101 * The below warning was added in place of
102 * pte_mkyoung(); if (is_write) pte_mkdirty();
103 * If it's triggered, we'd see normally a hang here (a clean pte is
104 * marked read-only to emulate the dirty bit).
105 * However, the generic code can mark a PTE writable but clean on a
106 * concurrent read fault, triggering this harmlessly. So comment it out.
109 WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
111 flush_tlb_page(vma, address);
113 mmap_read_unlock(mm);
119 * We ran out of memory, call the OOM killer, and return the userspace
120 * (which will retry the fault, or kill us if we got oom-killed).
122 mmap_read_unlock(mm);
124 goto out_nosemaphore;
125 pagefault_out_of_memory();
129 static void show_segv_info(struct uml_pt_regs *regs)
131 struct task_struct *tsk = current;
132 struct faultinfo *fi = UPT_FAULTINFO(regs);
134 if (!unhandled_signal(tsk, SIGSEGV))
137 if (!printk_ratelimit())
140 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %x",
141 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
142 tsk->comm, task_pid_nr(tsk), FAULT_ADDRESS(*fi),
143 (void *)UPT_IP(regs), (void *)UPT_SP(regs),
146 print_vma_addr(KERN_CONT " in ", UPT_IP(regs));
147 printk(KERN_CONT "\n");
150 static void bad_segv(struct faultinfo fi, unsigned long ip)
152 current->thread.arch.faultinfo = fi;
153 force_sig_fault(SIGSEGV, SEGV_ACCERR, (void __user *) FAULT_ADDRESS(fi));
156 void fatal_sigsegv(void)
158 force_fatal_sig(SIGSEGV);
159 do_signal(¤t->thread.regs);
161 * This is to tell gcc that we're not returning - do_signal
162 * can, in general, return, but in this case, it's not, since
163 * we just got a fatal SIGSEGV queued.
169 * segv_handler() - the SIGSEGV handler
170 * @sig: the signal number
171 * @unused_si: the signal info struct; unused in this handler
172 * @regs: the ptrace register information
174 * The handler first extracts the faultinfo from the UML ptrace regs struct.
175 * If the userfault did not happen in an UML userspace process, bad_segv is called.
176 * Otherwise the signal did happen in a cloned userspace process, handle it.
178 void segv_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
180 struct faultinfo * fi = UPT_FAULTINFO(regs);
182 if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
183 show_segv_info(regs);
184 bad_segv(*fi, UPT_IP(regs));
187 segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
191 * We give a *copy* of the faultinfo in the regs to segv.
192 * This must be done, since nesting SEGVs could overwrite
193 * the info in the regs. A pointer to the info then would
196 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
197 struct uml_pt_regs *regs)
202 int is_write = FAULT_WRITE(fi);
203 unsigned long address = FAULT_ADDRESS(fi);
205 if (!is_user && regs)
206 current->thread.segv_regs = container_of(regs, struct pt_regs, regs);
208 if (!is_user && (address >= start_vm) && (address < end_vm)) {
209 flush_tlb_kernel_vm();
212 else if (current->mm == NULL) {
213 show_regs(container_of(regs, struct pt_regs, regs));
214 panic("Segfault with no mm");
216 else if (!is_user && address > PAGE_SIZE && address < TASK_SIZE) {
217 show_regs(container_of(regs, struct pt_regs, regs));
218 panic("Kernel tried to access user memory at addr 0x%lx, ip 0x%lx",
222 if (SEGV_IS_FIXABLE(&fi))
223 err = handle_page_fault(address, ip, is_write, is_user,
228 * A thread accessed NULL, we get a fault, but CR2 is invalid.
229 * This code is used in __do_copy_from_user() of TT mode.
230 * XXX tt mode is gone, so maybe this isn't needed any more
235 catcher = current->thread.fault_catcher;
238 else if (catcher != NULL) {
239 current->thread.fault_addr = (void *) address;
240 UML_LONGJMP(catcher, 1);
242 else if (current->thread.fault_addr != NULL)
243 panic("fault_addr set but no fault catcher");
244 else if (!is_user && arch_fixup(ip, regs))
248 show_regs(container_of(regs, struct pt_regs, regs));
249 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
253 show_segv_info(regs);
255 if (err == -EACCES) {
256 current->thread.arch.faultinfo = fi;
257 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
259 BUG_ON(err != -EFAULT);
260 current->thread.arch.faultinfo = fi;
261 force_sig_fault(SIGSEGV, si_code, (void __user *) address);
266 current->thread.segv_regs = NULL;
271 void relay_signal(int sig, struct siginfo *si, struct uml_pt_regs *regs)
274 if (!UPT_IS_USER(regs)) {
276 printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
277 "mount likely just ran out of space\n");
278 panic("Kernel mode signal %d", sig);
281 arch_examine_signal(sig, regs);
283 /* Is the signal layout for the signal known?
284 * Signal data must be scrubbed to prevent information leaks.
288 if ((err == 0) && (siginfo_layout(sig, code) == SIL_FAULT)) {
289 struct faultinfo *fi = UPT_FAULTINFO(regs);
290 current->thread.arch.faultinfo = *fi;
291 force_sig_fault(sig, code, (void __user *)FAULT_ADDRESS(*fi));
293 printk(KERN_ERR "Attempted to relay unknown signal %d (si_code = %d) with errno %d\n",
299 void bus_handler(int sig, struct siginfo *si, struct uml_pt_regs *regs)
301 if (current->thread.fault_catcher != NULL)
302 UML_LONGJMP(current->thread.fault_catcher, 1);
304 relay_signal(sig, si, regs);
307 void winch(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
309 do_IRQ(WINCH_IRQ, regs);