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