xref: /openbmc/linux/fs/proc/kcore.c (revision 8ec90bfd) 
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *	fs/proc/kcore.c kernel ELF core dumper
4  *
5  *	Modelled on fs/exec.c:aout_core_dump()
6  *	Jeremy Fitzhardinge <jeremy@sw.oz.au>
7  *	ELF version written by David Howells <David.Howells@nexor.co.uk>
8  *	Modified and incorporated into 2.3.x by Tigran Aivazian <tigran@veritas.com>
9  *	Support to dump vmalloc'd areas (ELF only), Tigran Aivazian <tigran@veritas.com>
10  *	Safe accesses to vmalloc/direct-mapped discontiguous areas, Kanoj Sarcar <kanoj@sgi.com>
11  */
12 
13 #include <linux/crash_core.h>
14 #include <linux/mm.h>
15 #include <linux/proc_fs.h>
16 #include <linux/kcore.h>
17 #include <linux/user.h>
18 #include <linux/capability.h>
19 #include <linux/elf.h>
20 #include <linux/elfcore.h>
21 #include <linux/notifier.h>
22 #include <linux/vmalloc.h>
23 #include <linux/highmem.h>
24 #include <linux/printk.h>
25 #include <linux/memblock.h>
26 #include <linux/init.h>
27 #include <linux/slab.h>
28 #include <linux/uaccess.h>
29 #include <asm/io.h>
30 #include <linux/list.h>
31 #include <linux/ioport.h>
32 #include <linux/memory.h>
33 #include <linux/sched/task.h>
34 #include <linux/security.h>
35 #include <asm/sections.h>
36 #include "internal.h"
37 
38 #define CORE_STR "CORE"
39 
40 #ifndef ELF_CORE_EFLAGS
41 #define ELF_CORE_EFLAGS	0
42 #endif
43 
44 static struct proc_dir_entry *proc_root_kcore;
45 
46 
47 #ifndef kc_vaddr_to_offset
48 #define	kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
49 #endif
50 #ifndef	kc_offset_to_vaddr
51 #define	kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
52 #endif
53 
54 static LIST_HEAD(kclist_head);
55 static DECLARE_RWSEM(kclist_lock);
56 static int kcore_need_update = 1;
57 
58 /*
59  * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
60  * Same as oldmem_pfn_is_ram in vmcore
61  */
62 static int (*mem_pfn_is_ram)(unsigned long pfn);
63 
64 int __init register_mem_pfn_is_ram(int (*fn)(unsigned long pfn))
65 {
66 	if (mem_pfn_is_ram)
67 		return -EBUSY;
68 	mem_pfn_is_ram = fn;
69 	return 0;
70 }
71 
72 static int pfn_is_ram(unsigned long pfn)
73 {
74 	if (mem_pfn_is_ram)
75 		return mem_pfn_is_ram(pfn);
76 	else
77 		return 1;
78 }
79 
80 /* This doesn't grab kclist_lock, so it should only be used at init time. */
81 void __init kclist_add(struct kcore_list *new, void *addr, size_t size,
82 		       int type)
83 {
84 	new->addr = (unsigned long)addr;
85 	new->size = size;
86 	new->type = type;
87 
88 	list_add_tail(&new->list, &kclist_head);
89 }
90 
91 static size_t get_kcore_size(int *nphdr, size_t *phdrs_len, size_t *notes_len,
92 			     size_t *data_offset)
93 {
94 	size_t try, size;
95 	struct kcore_list *m;
96 
97 	*nphdr = 1; /* PT_NOTE */
98 	size = 0;
99 
100 	list_for_each_entry(m, &kclist_head, list) {
101 		try = kc_vaddr_to_offset((size_t)m->addr + m->size);
102 		if (try > size)
103 			size = try;
104 		*nphdr = *nphdr + 1;
105 	}
106 
107 	*phdrs_len = *nphdr * sizeof(struct elf_phdr);
108 	*notes_len = (4 * sizeof(struct elf_note) +
109 		      3 * ALIGN(sizeof(CORE_STR), 4) +
110 		      VMCOREINFO_NOTE_NAME_BYTES +
111 		      ALIGN(sizeof(struct elf_prstatus), 4) +
112 		      ALIGN(sizeof(struct elf_prpsinfo), 4) +
113 		      ALIGN(arch_task_struct_size, 4) +
114 		      ALIGN(vmcoreinfo_size, 4));
115 	*data_offset = PAGE_ALIGN(sizeof(struct elfhdr) + *phdrs_len +
116 				  *notes_len);
117 	return *data_offset + size;
118 }
119 
120 #ifdef CONFIG_HIGHMEM
121 /*
122  * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
123  * because memory hole is not as big as !HIGHMEM case.
124  * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
125  */
126 static int kcore_ram_list(struct list_head *head)
127 {
128 	struct kcore_list *ent;
129 
130 	ent = kmalloc(sizeof(*ent), GFP_KERNEL);
131 	if (!ent)
132 		return -ENOMEM;
133 	ent->addr = (unsigned long)__va(0);
134 	ent->size = max_low_pfn << PAGE_SHIFT;
135 	ent->type = KCORE_RAM;
136 	list_add(&ent->list, head);
137 	return 0;
138 }
139 
140 #else /* !CONFIG_HIGHMEM */
141 
142 #ifdef CONFIG_SPARSEMEM_VMEMMAP
143 /* calculate vmemmap's address from given system ram pfn and register it */
144 static int
145 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
146 {
147 	unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
148 	unsigned long nr_pages = ent->size >> PAGE_SHIFT;
149 	unsigned long start, end;
150 	struct kcore_list *vmm, *tmp;
151 
152 
153 	start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
154 	end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
155 	end = PAGE_ALIGN(end);
156 	/* overlap check (because we have to align page */
157 	list_for_each_entry(tmp, head, list) {
158 		if (tmp->type != KCORE_VMEMMAP)
159 			continue;
160 		if (start < tmp->addr + tmp->size)
161 			if (end > tmp->addr)
162 				end = tmp->addr;
163 	}
164 	if (start < end) {
165 		vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
166 		if (!vmm)
167 			return 0;
168 		vmm->addr = start;
169 		vmm->size = end - start;
170 		vmm->type = KCORE_VMEMMAP;
171 		list_add_tail(&vmm->list, head);
172 	}
173 	return 1;
174 
175 }
176 #else
177 static int
178 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
179 {
180 	return 1;
181 }
182 
183 #endif
184 
185 static int
186 kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
187 {
188 	struct list_head *head = (struct list_head *)arg;
189 	struct kcore_list *ent;
190 	struct page *p;
191 
192 	if (!pfn_valid(pfn))
193 		return 1;
194 
195 	p = pfn_to_page(pfn);
196 	if (!memmap_valid_within(pfn, p, page_zone(p)))
197 		return 1;
198 
199 	ent = kmalloc(sizeof(*ent), GFP_KERNEL);
200 	if (!ent)
201 		return -ENOMEM;
202 	ent->addr = (unsigned long)page_to_virt(p);
203 	ent->size = nr_pages << PAGE_SHIFT;
204 
205 	if (!virt_addr_valid(ent->addr))
206 		goto free_out;
207 
208 	/* cut not-mapped area. ....from ppc-32 code. */
209 	if (ULONG_MAX - ent->addr < ent->size)
210 		ent->size = ULONG_MAX - ent->addr;
211 
212 	/*
213 	 * We've already checked virt_addr_valid so we know this address
214 	 * is a valid pointer, therefore we can check against it to determine
215 	 * if we need to trim
216 	 */
217 	if (VMALLOC_START > ent->addr) {
218 		if (VMALLOC_START - ent->addr < ent->size)
219 			ent->size = VMALLOC_START - ent->addr;
220 	}
221 
222 	ent->type = KCORE_RAM;
223 	list_add_tail(&ent->list, head);
224 
225 	if (!get_sparsemem_vmemmap_info(ent, head)) {
226 		list_del(&ent->list);
227 		goto free_out;
228 	}
229 
230 	return 0;
231 free_out:
232 	kfree(ent);
233 	return 1;
234 }
235 
236 static int kcore_ram_list(struct list_head *list)
237 {
238 	int nid, ret;
239 	unsigned long end_pfn;
240 
241 	/* Not inialized....update now */
242 	/* find out "max pfn" */
243 	end_pfn = 0;
244 	for_each_node_state(nid, N_MEMORY) {
245 		unsigned long node_end;
246 		node_end = node_end_pfn(nid);
247 		if (end_pfn < node_end)
248 			end_pfn = node_end;
249 	}
250 	/* scan 0 to max_pfn */
251 	ret = walk_system_ram_range(0, end_pfn, list, kclist_add_private);
252 	if (ret)
253 		return -ENOMEM;
254 	return 0;
255 }
256 #endif /* CONFIG_HIGHMEM */
257 
258 static int kcore_update_ram(void)
259 {
260 	LIST_HEAD(list);
261 	LIST_HEAD(garbage);
262 	int nphdr;
263 	size_t phdrs_len, notes_len, data_offset;
264 	struct kcore_list *tmp, *pos;
265 	int ret = 0;
266 
267 	down_write(&kclist_lock);
268 	if (!xchg(&kcore_need_update, 0))
269 		goto out;
270 
271 	ret = kcore_ram_list(&list);
272 	if (ret) {
273 		/* Couldn't get the RAM list, try again next time. */
274 		WRITE_ONCE(kcore_need_update, 1);
275 		list_splice_tail(&list, &garbage);
276 		goto out;
277 	}
278 
279 	list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
280 		if (pos->type == KCORE_RAM || pos->type == KCORE_VMEMMAP)
281 			list_move(&pos->list, &garbage);
282 	}
283 	list_splice_tail(&list, &kclist_head);
284 
285 	proc_root_kcore->size = get_kcore_size(&nphdr, &phdrs_len, &notes_len,
286 					       &data_offset);
287 
288 out:
289 	up_write(&kclist_lock);
290 	list_for_each_entry_safe(pos, tmp, &garbage, list) {
291 		list_del(&pos->list);
292 		kfree(pos);
293 	}
294 	return ret;
295 }
296 
297 static void append_kcore_note(char *notes, size_t *i, const char *name,
298 			      unsigned int type, const void *desc,
299 			      size_t descsz)
300 {
301 	struct elf_note *note = (struct elf_note *)&notes[*i];
302 
303 	note->n_namesz = strlen(name) + 1;
304 	note->n_descsz = descsz;
305 	note->n_type = type;
306 	*i += sizeof(*note);
307 	memcpy(&notes[*i], name, note->n_namesz);
308 	*i = ALIGN(*i + note->n_namesz, 4);
309 	memcpy(&notes[*i], desc, descsz);
310 	*i = ALIGN(*i + descsz, 4);
311 }
312 
313 static ssize_t
314 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
315 {
316 	char *buf = file->private_data;
317 	size_t phdrs_offset, notes_offset, data_offset;
318 	size_t phdrs_len, notes_len;
319 	struct kcore_list *m;
320 	size_t tsz;
321 	int nphdr;
322 	unsigned long start;
323 	size_t orig_buflen = buflen;
324 	int ret = 0;
325 
326 	down_read(&kclist_lock);
327 
328 	get_kcore_size(&nphdr, &phdrs_len, &notes_len, &data_offset);
329 	phdrs_offset = sizeof(struct elfhdr);
330 	notes_offset = phdrs_offset + phdrs_len;
331 
332 	/* ELF file header. */
333 	if (buflen && *fpos < sizeof(struct elfhdr)) {
334 		struct elfhdr ehdr = {
335 			.e_ident = {
336 				[EI_MAG0] = ELFMAG0,
337 				[EI_MAG1] = ELFMAG1,
338 				[EI_MAG2] = ELFMAG2,
339 				[EI_MAG3] = ELFMAG3,
340 				[EI_CLASS] = ELF_CLASS,
341 				[EI_DATA] = ELF_DATA,
342 				[EI_VERSION] = EV_CURRENT,
343 				[EI_OSABI] = ELF_OSABI,
344 			},
345 			.e_type = ET_CORE,
346 			.e_machine = ELF_ARCH,
347 			.e_version = EV_CURRENT,
348 			.e_phoff = sizeof(struct elfhdr),
349 			.e_flags = ELF_CORE_EFLAGS,
350 			.e_ehsize = sizeof(struct elfhdr),
351 			.e_phentsize = sizeof(struct elf_phdr),
352 			.e_phnum = nphdr,
353 		};
354 
355 		tsz = min_t(size_t, buflen, sizeof(struct elfhdr) - *fpos);
356 		if (copy_to_user(buffer, (char *)&ehdr + *fpos, tsz)) {
357 			ret = -EFAULT;
358 			goto out;
359 		}
360 
361 		buffer += tsz;
362 		buflen -= tsz;
363 		*fpos += tsz;
364 	}
365 
366 	/* ELF program headers. */
367 	if (buflen && *fpos < phdrs_offset + phdrs_len) {
368 		struct elf_phdr *phdrs, *phdr;
369 
370 		phdrs = kzalloc(phdrs_len, GFP_KERNEL);
371 		if (!phdrs) {
372 			ret = -ENOMEM;
373 			goto out;
374 		}
375 
376 		phdrs[0].p_type = PT_NOTE;
377 		phdrs[0].p_offset = notes_offset;
378 		phdrs[0].p_filesz = notes_len;
379 
380 		phdr = &phdrs[1];
381 		list_for_each_entry(m, &kclist_head, list) {
382 			phdr->p_type = PT_LOAD;
383 			phdr->p_flags = PF_R | PF_W | PF_X;
384 			phdr->p_offset = kc_vaddr_to_offset(m->addr) + data_offset;
385 			if (m->type == KCORE_REMAP)
386 				phdr->p_vaddr = (size_t)m->vaddr;
387 			else
388 				phdr->p_vaddr = (size_t)m->addr;
389 			if (m->type == KCORE_RAM || m->type == KCORE_REMAP)
390 				phdr->p_paddr = __pa(m->addr);
391 			else if (m->type == KCORE_TEXT)
392 				phdr->p_paddr = __pa_symbol(m->addr);
393 			else
394 				phdr->p_paddr = (elf_addr_t)-1;
395 			phdr->p_filesz = phdr->p_memsz = m->size;
396 			phdr->p_align = PAGE_SIZE;
397 			phdr++;
398 		}
399 
400 		tsz = min_t(size_t, buflen, phdrs_offset + phdrs_len - *fpos);
401 		if (copy_to_user(buffer, (char *)phdrs + *fpos - phdrs_offset,
402 				 tsz)) {
403 			kfree(phdrs);
404 			ret = -EFAULT;
405 			goto out;
406 		}
407 		kfree(phdrs);
408 
409 		buffer += tsz;
410 		buflen -= tsz;
411 		*fpos += tsz;
412 	}
413 
414 	/* ELF note segment. */
415 	if (buflen && *fpos < notes_offset + notes_len) {
416 		struct elf_prstatus prstatus = {};
417 		struct elf_prpsinfo prpsinfo = {
418 			.pr_sname = 'R',
419 			.pr_fname = "vmlinux",
420 		};
421 		char *notes;
422 		size_t i = 0;
423 
424 		strlcpy(prpsinfo.pr_psargs, saved_command_line,
425 			sizeof(prpsinfo.pr_psargs));
426 
427 		notes = kzalloc(notes_len, GFP_KERNEL);
428 		if (!notes) {
429 			ret = -ENOMEM;
430 			goto out;
431 		}
432 
433 		append_kcore_note(notes, &i, CORE_STR, NT_PRSTATUS, &prstatus,
434 				  sizeof(prstatus));
435 		append_kcore_note(notes, &i, CORE_STR, NT_PRPSINFO, &prpsinfo,
436 				  sizeof(prpsinfo));
437 		append_kcore_note(notes, &i, CORE_STR, NT_TASKSTRUCT, current,
438 				  arch_task_struct_size);
439 		/*
440 		 * vmcoreinfo_size is mostly constant after init time, but it
441 		 * can be changed by crash_save_vmcoreinfo(). Racing here with a
442 		 * panic on another CPU before the machine goes down is insanely
443 		 * unlikely, but it's better to not leave potential buffer
444 		 * overflows lying around, regardless.
445 		 */
446 		append_kcore_note(notes, &i, VMCOREINFO_NOTE_NAME, 0,
447 				  vmcoreinfo_data,
448 				  min(vmcoreinfo_size, notes_len - i));
449 
450 		tsz = min_t(size_t, buflen, notes_offset + notes_len - *fpos);
451 		if (copy_to_user(buffer, notes + *fpos - notes_offset, tsz)) {
452 			kfree(notes);
453 			ret = -EFAULT;
454 			goto out;
455 		}
456 		kfree(notes);
457 
458 		buffer += tsz;
459 		buflen -= tsz;
460 		*fpos += tsz;
461 	}
462 
463 	/*
464 	 * Check to see if our file offset matches with any of
465 	 * the addresses in the elf_phdr on our list.
466 	 */
467 	start = kc_offset_to_vaddr(*fpos - data_offset);
468 	if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
469 		tsz = buflen;
470 
471 	m = NULL;
472 	while (buflen) {
473 		/*
474 		 * If this is the first iteration or the address is not within
475 		 * the previous entry, search for a matching entry.
476 		 */
477 		if (!m || start < m->addr || start >= m->addr + m->size) {
478 			list_for_each_entry(m, &kclist_head, list) {
479 				if (start >= m->addr &&
480 				    start < m->addr + m->size)
481 					break;
482 			}
483 		}
484 
485 		if (&m->list == &kclist_head) {
486 			if (clear_user(buffer, tsz)) {
487 				ret = -EFAULT;
488 				goto out;
489 			}
490 			m = NULL;	/* skip the list anchor */
491 		} else if (!pfn_is_ram(__pa(start) >> PAGE_SHIFT)) {
492 			if (clear_user(buffer, tsz)) {
493 				ret = -EFAULT;
494 				goto out;
495 			}
496 		} else if (m->type == KCORE_VMALLOC) {
497 			vread(buf, (char *)start, tsz);
498 			/* we have to zero-fill user buffer even if no read */
499 			if (copy_to_user(buffer, buf, tsz)) {
500 				ret = -EFAULT;
501 				goto out;
502 			}
503 		} else if (m->type == KCORE_USER) {
504 			/* User page is handled prior to normal kernel page: */
505 			if (copy_to_user(buffer, (char *)start, tsz)) {
506 				ret = -EFAULT;
507 				goto out;
508 			}
509 		} else {
510 			if (kern_addr_valid(start)) {
511 				/*
512 				 * Using bounce buffer to bypass the
513 				 * hardened user copy kernel text checks.
514 				 */
515 				if (copy_from_kernel_nofault(buf, (void *)start,
516 						tsz)) {
517 					if (clear_user(buffer, tsz)) {
518 						ret = -EFAULT;
519 						goto out;
520 					}
521 				} else {
522 					if (copy_to_user(buffer, buf, tsz)) {
523 						ret = -EFAULT;
524 						goto out;
525 					}
526 				}
527 			} else {
528 				if (clear_user(buffer, tsz)) {
529 					ret = -EFAULT;
530 					goto out;
531 				}
532 			}
533 		}
534 		buflen -= tsz;
535 		*fpos += tsz;
536 		buffer += tsz;
537 		start += tsz;
538 		tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
539 	}
540 
541 out:
542 	up_read(&kclist_lock);
543 	if (ret)
544 		return ret;
545 	return orig_buflen - buflen;
546 }
547 
548 static int open_kcore(struct inode *inode, struct file *filp)
549 {
550 	int ret = security_locked_down(LOCKDOWN_KCORE);
551 
552 	if (!capable(CAP_SYS_RAWIO))
553 		return -EPERM;
554 
555 	if (ret)
556 		return ret;
557 
558 	filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
559 	if (!filp->private_data)
560 		return -ENOMEM;
561 
562 	if (kcore_need_update)
563 		kcore_update_ram();
564 	if (i_size_read(inode) != proc_root_kcore->size) {
565 		inode_lock(inode);
566 		i_size_write(inode, proc_root_kcore->size);
567 		inode_unlock(inode);
568 	}
569 	return 0;
570 }
571 
572 static int release_kcore(struct inode *inode, struct file *file)
573 {
574 	kfree(file->private_data);
575 	return 0;
576 }
577 
578 static const struct proc_ops kcore_proc_ops = {
579 	.proc_read	= read_kcore,
580 	.proc_open	= open_kcore,
581 	.proc_release	= release_kcore,
582 	.proc_lseek	= default_llseek,
583 };
584 
585 /* just remember that we have to update kcore */
586 static int __meminit kcore_callback(struct notifier_block *self,
587 				    unsigned long action, void *arg)
588 {
589 	switch (action) {
590 	case MEM_ONLINE:
591 	case MEM_OFFLINE:
592 		kcore_need_update = 1;
593 		break;
594 	}
595 	return NOTIFY_OK;
596 }
597 
598 static struct notifier_block kcore_callback_nb __meminitdata = {
599 	.notifier_call = kcore_callback,
600 	.priority = 0,
601 };
602 
603 static struct kcore_list kcore_vmalloc;
604 
605 #ifdef CONFIG_ARCH_PROC_KCORE_TEXT
606 static struct kcore_list kcore_text;
607 /*
608  * If defined, special segment is used for mapping kernel text instead of
609  * direct-map area. We need to create special TEXT section.
610  */
611 static void __init proc_kcore_text_init(void)
612 {
613 	kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
614 }
615 #else
616 static void __init proc_kcore_text_init(void)
617 {
618 }
619 #endif
620 
621 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
622 /*
623  * MODULES_VADDR has no intersection with VMALLOC_ADDR.
624  */
625 static struct kcore_list kcore_modules;
626 static void __init add_modules_range(void)
627 {
628 	if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) {
629 		kclist_add(&kcore_modules, (void *)MODULES_VADDR,
630 			MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
631 	}
632 }
633 #else
634 static void __init add_modules_range(void)
635 {
636 }
637 #endif
638 
639 static int __init proc_kcore_init(void)
640 {
641 	proc_root_kcore = proc_create("kcore", S_IRUSR, NULL, &kcore_proc_ops);
642 	if (!proc_root_kcore) {
643 		pr_err("couldn't create /proc/kcore\n");
644 		return 0; /* Always returns 0. */
645 	}
646 	/* Store text area if it's special */
647 	proc_kcore_text_init();
648 	/* Store vmalloc area */
649 	kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
650 		VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
651 	add_modules_range();
652 	/* Store direct-map area from physical memory map */
653 	kcore_update_ram();
654 	register_hotmemory_notifier(&kcore_callback_nb);
655 
656 	return 0;
657 }
658 fs_initcall(proc_kcore_init);
659