xref: /openbmc/linux/arch/um/kernel/trap.c (revision 20e2fc42)
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  */
25 int 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;
48 retry:
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 
62 good_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);
124 out:
125 	up_read(&mm->mmap_sem);
126 out_nosemaphore:
127 	return err;
128 
129 out_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 }
140 EXPORT_SYMBOL(handle_page_fault);
141 
142 static 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 
163 static 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 
169 void fatal_sigsegv(void)
170 {
171 	force_sigsegv(SIGSEGV);
172 	do_signal(&current->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  */
191 void 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  */
209 unsigned 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 
277 out:
278 	if (regs)
279 		current->thread.segv_regs = NULL;
280 
281 	return 0;
282 }
283 
284 void 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 
312 void 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 
320 void winch(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
321 {
322 	do_IRQ(WINCH_IRQ, regs);
323 }
324 
325 void trap_init(void)
326 {
327 }
328