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 */
handle_page_fault(unsigned long address,unsigned long ip,int is_write,int is_user,int * code_out)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 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
60 good_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
117 out:
118 mmap_read_unlock(mm);
119 out_nosemaphore:
120 return err;
121
122 out_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
show_segv_info(struct uml_pt_regs * regs)134 static 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
bad_segv(struct faultinfo fi,unsigned long ip)155 static 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
fatal_sigsegv(void)161 void 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 */
segv_handler(int sig,struct siginfo * unused_si,struct uml_pt_regs * regs)183 void 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 */
segv(struct faultinfo fi,unsigned long ip,int is_user,struct uml_pt_regs * regs)201 unsigned 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
270 out:
271 if (regs)
272 current->thread.segv_regs = NULL;
273
274 return 0;
275 }
276
relay_signal(int sig,struct siginfo * si,struct uml_pt_regs * regs)277 void 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
winch(int sig,struct siginfo * unused_si,struct uml_pt_regs * regs)305 void winch(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
306 {
307 do_IRQ(WINCH_IRQ, regs);
308 }
309