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