xref: /openbmc/linux/fs/proc/kcore.c (revision b60a5b8d) 
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 <asm/sections.h>
35 #include "internal.h"
36 
37 #define CORE_STR "CORE"
38 
39 #ifndef ELF_CORE_EFLAGS
40 #define ELF_CORE_EFLAGS	0
41 #endif
42 
43 static struct proc_dir_entry *proc_root_kcore;
44 
45 
46 #ifndef kc_vaddr_to_offset
47 #define	kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
48 #endif
49 #ifndef	kc_offset_to_vaddr
50 #define	kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
51 #endif
52 
53 static LIST_HEAD(kclist_head);
54 static DECLARE_RWSEM(kclist_lock);
55 static int kcore_need_update = 1;
56 
57 /* This doesn't grab kclist_lock, so it should only be used at init time. */
58 void __init kclist_add(struct kcore_list *new, void *addr, size_t size,
59 		       int type)
60 {
61 	new->addr = (unsigned long)addr;
62 	new->size = size;
63 	new->type = type;
64 
65 	list_add_tail(&new->list, &kclist_head);
66 }
67 
68 static size_t get_kcore_size(int *nphdr, size_t *phdrs_len, size_t *notes_len,
69 			     size_t *data_offset)
70 {
71 	size_t try, size;
72 	struct kcore_list *m;
73 
74 	*nphdr = 1; /* PT_NOTE */
75 	size = 0;
76 
77 	list_for_each_entry(m, &kclist_head, list) {
78 		try = kc_vaddr_to_offset((size_t)m->addr + m->size);
79 		if (try > size)
80 			size = try;
81 		*nphdr = *nphdr + 1;
82 	}
83 
84 	*phdrs_len = *nphdr * sizeof(struct elf_phdr);
85 	*notes_len = (4 * sizeof(struct elf_note) +
86 		      3 * ALIGN(sizeof(CORE_STR), 4) +
87 		      VMCOREINFO_NOTE_NAME_BYTES +
88 		      ALIGN(sizeof(struct elf_prstatus), 4) +
89 		      ALIGN(sizeof(struct elf_prpsinfo), 4) +
90 		      ALIGN(arch_task_struct_size, 4) +
91 		      ALIGN(vmcoreinfo_size, 4));
92 	*data_offset = PAGE_ALIGN(sizeof(struct elfhdr) + *phdrs_len +
93 				  *notes_len);
94 	return *data_offset + size;
95 }
96 
97 #ifdef CONFIG_HIGHMEM
98 /*
99  * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
100  * because memory hole is not as big as !HIGHMEM case.
101  * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
102  */
103 static int kcore_ram_list(struct list_head *head)
104 {
105 	struct kcore_list *ent;
106 
107 	ent = kmalloc(sizeof(*ent), GFP_KERNEL);
108 	if (!ent)
109 		return -ENOMEM;
110 	ent->addr = (unsigned long)__va(0);
111 	ent->size = max_low_pfn << PAGE_SHIFT;
112 	ent->type = KCORE_RAM;
113 	list_add(&ent->list, head);
114 	return 0;
115 }
116 
117 #else /* !CONFIG_HIGHMEM */
118 
119 #ifdef CONFIG_SPARSEMEM_VMEMMAP
120 /* calculate vmemmap's address from given system ram pfn and register it */
121 static int
122 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
123 {
124 	unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
125 	unsigned long nr_pages = ent->size >> PAGE_SHIFT;
126 	unsigned long start, end;
127 	struct kcore_list *vmm, *tmp;
128 
129 
130 	start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
131 	end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
132 	end = PAGE_ALIGN(end);
133 	/* overlap check (because we have to align page */
134 	list_for_each_entry(tmp, head, list) {
135 		if (tmp->type != KCORE_VMEMMAP)
136 			continue;
137 		if (start < tmp->addr + tmp->size)
138 			if (end > tmp->addr)
139 				end = tmp->addr;
140 	}
141 	if (start < end) {
142 		vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
143 		if (!vmm)
144 			return 0;
145 		vmm->addr = start;
146 		vmm->size = end - start;
147 		vmm->type = KCORE_VMEMMAP;
148 		list_add_tail(&vmm->list, head);
149 	}
150 	return 1;
151 
152 }
153 #else
154 static int
155 get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
156 {
157 	return 1;
158 }
159 
160 #endif
161 
162 static int
163 kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
164 {
165 	struct list_head *head = (struct list_head *)arg;
166 	struct kcore_list *ent;
167 	struct page *p;
168 
169 	if (!pfn_valid(pfn))
170 		return 1;
171 
172 	p = pfn_to_page(pfn);
173 	if (!memmap_valid_within(pfn, p, page_zone(p)))
174 		return 1;
175 
176 	ent = kmalloc(sizeof(*ent), GFP_KERNEL);
177 	if (!ent)
178 		return -ENOMEM;
179 	ent->addr = (unsigned long)page_to_virt(p);
180 	ent->size = nr_pages << PAGE_SHIFT;
181 
182 	if (!virt_addr_valid(ent->addr))
183 		goto free_out;
184 
185 	/* cut not-mapped area. ....from ppc-32 code. */
186 	if (ULONG_MAX - ent->addr < ent->size)
187 		ent->size = ULONG_MAX - ent->addr;
188 
189 	/*
190 	 * We've already checked virt_addr_valid so we know this address
191 	 * is a valid pointer, therefore we can check against it to determine
192 	 * if we need to trim
193 	 */
194 	if (VMALLOC_START > ent->addr) {
195 		if (VMALLOC_START - ent->addr < ent->size)
196 			ent->size = VMALLOC_START - ent->addr;
197 	}
198 
199 	ent->type = KCORE_RAM;
200 	list_add_tail(&ent->list, head);
201 
202 	if (!get_sparsemem_vmemmap_info(ent, head)) {
203 		list_del(&ent->list);
204 		goto free_out;
205 	}
206 
207 	return 0;
208 free_out:
209 	kfree(ent);
210 	return 1;
211 }
212 
213 static int kcore_ram_list(struct list_head *list)
214 {
215 	int nid, ret;
216 	unsigned long end_pfn;
217 
218 	/* Not inialized....update now */
219 	/* find out "max pfn" */
220 	end_pfn = 0;
221 	for_each_node_state(nid, N_MEMORY) {
222 		unsigned long node_end;
223 		node_end = node_end_pfn(nid);
224 		if (end_pfn < node_end)
225 			end_pfn = node_end;
226 	}
227 	/* scan 0 to max_pfn */
228 	ret = walk_system_ram_range(0, end_pfn, list, kclist_add_private);
229 	if (ret)
230 		return -ENOMEM;
231 	return 0;
232 }
233 #endif /* CONFIG_HIGHMEM */
234 
235 static int kcore_update_ram(void)
236 {
237 	LIST_HEAD(list);
238 	LIST_HEAD(garbage);
239 	int nphdr;
240 	size_t phdrs_len, notes_len, data_offset;
241 	struct kcore_list *tmp, *pos;
242 	int ret = 0;
243 
244 	down_write(&kclist_lock);
245 	if (!xchg(&kcore_need_update, 0))
246 		goto out;
247 
248 	ret = kcore_ram_list(&list);
249 	if (ret) {
250 		/* Couldn't get the RAM list, try again next time. */
251 		WRITE_ONCE(kcore_need_update, 1);
252 		list_splice_tail(&list, &garbage);
253 		goto out;
254 	}
255 
256 	list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
257 		if (pos->type == KCORE_RAM || pos->type == KCORE_VMEMMAP)
258 			list_move(&pos->list, &garbage);
259 	}
260 	list_splice_tail(&list, &kclist_head);
261 
262 	proc_root_kcore->size = get_kcore_size(&nphdr, &phdrs_len, &notes_len,
263 					       &data_offset);
264 
265 out:
266 	up_write(&kclist_lock);
267 	list_for_each_entry_safe(pos, tmp, &garbage, list) {
268 		list_del(&pos->list);
269 		kfree(pos);
270 	}
271 	return ret;
272 }
273 
274 static void append_kcore_note(char *notes, size_t *i, const char *name,
275 			      unsigned int type, const void *desc,
276 			      size_t descsz)
277 {
278 	struct elf_note *note = (struct elf_note *)&notes[*i];
279 
280 	note->n_namesz = strlen(name) + 1;
281 	note->n_descsz = descsz;
282 	note->n_type = type;
283 	*i += sizeof(*note);
284 	memcpy(&notes[*i], name, note->n_namesz);
285 	*i = ALIGN(*i + note->n_namesz, 4);
286 	memcpy(&notes[*i], desc, descsz);
287 	*i = ALIGN(*i + descsz, 4);
288 }
289 
290 static ssize_t
291 read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
292 {
293 	char *buf = file->private_data;
294 	size_t phdrs_offset, notes_offset, data_offset;
295 	size_t phdrs_len, notes_len;
296 	struct kcore_list *m;
297 	size_t tsz;
298 	int nphdr;
299 	unsigned long start;
300 	size_t orig_buflen = buflen;
301 	int ret = 0;
302 
303 	down_read(&kclist_lock);
304 
305 	get_kcore_size(&nphdr, &phdrs_len, &notes_len, &data_offset);
306 	phdrs_offset = sizeof(struct elfhdr);
307 	notes_offset = phdrs_offset + phdrs_len;
308 
309 	/* ELF file header. */
310 	if (buflen && *fpos < sizeof(struct elfhdr)) {
311 		struct elfhdr ehdr = {
312 			.e_ident = {
313 				[EI_MAG0] = ELFMAG0,
314 				[EI_MAG1] = ELFMAG1,
315 				[EI_MAG2] = ELFMAG2,
316 				[EI_MAG3] = ELFMAG3,
317 				[EI_CLASS] = ELF_CLASS,
318 				[EI_DATA] = ELF_DATA,
319 				[EI_VERSION] = EV_CURRENT,
320 				[EI_OSABI] = ELF_OSABI,
321 			},
322 			.e_type = ET_CORE,
323 			.e_machine = ELF_ARCH,
324 			.e_version = EV_CURRENT,
325 			.e_phoff = sizeof(struct elfhdr),
326 			.e_flags = ELF_CORE_EFLAGS,
327 			.e_ehsize = sizeof(struct elfhdr),
328 			.e_phentsize = sizeof(struct elf_phdr),
329 			.e_phnum = nphdr,
330 		};
331 
332 		tsz = min_t(size_t, buflen, sizeof(struct elfhdr) - *fpos);
333 		if (copy_to_user(buffer, (char *)&ehdr + *fpos, tsz)) {
334 			ret = -EFAULT;
335 			goto out;
336 		}
337 
338 		buffer += tsz;
339 		buflen -= tsz;
340 		*fpos += tsz;
341 	}
342 
343 	/* ELF program headers. */
344 	if (buflen && *fpos < phdrs_offset + phdrs_len) {
345 		struct elf_phdr *phdrs, *phdr;
346 
347 		phdrs = kzalloc(phdrs_len, GFP_KERNEL);
348 		if (!phdrs) {
349 			ret = -ENOMEM;
350 			goto out;
351 		}
352 
353 		phdrs[0].p_type = PT_NOTE;
354 		phdrs[0].p_offset = notes_offset;
355 		phdrs[0].p_filesz = notes_len;
356 
357 		phdr = &phdrs[1];
358 		list_for_each_entry(m, &kclist_head, list) {
359 			phdr->p_type = PT_LOAD;
360 			phdr->p_flags = PF_R | PF_W | PF_X;
361 			phdr->p_offset = kc_vaddr_to_offset(m->addr) + data_offset;
362 			if (m->type == KCORE_REMAP)
363 				phdr->p_vaddr = (size_t)m->vaddr;
364 			else
365 				phdr->p_vaddr = (size_t)m->addr;
366 			if (m->type == KCORE_RAM || m->type == KCORE_REMAP)
367 				phdr->p_paddr = __pa(m->addr);
368 			else if (m->type == KCORE_TEXT)
369 				phdr->p_paddr = __pa_symbol(m->addr);
370 			else
371 				phdr->p_paddr = (elf_addr_t)-1;
372 			phdr->p_filesz = phdr->p_memsz = m->size;
373 			phdr->p_align = PAGE_SIZE;
374 			phdr++;
375 		}
376 
377 		tsz = min_t(size_t, buflen, phdrs_offset + phdrs_len - *fpos);
378 		if (copy_to_user(buffer, (char *)phdrs + *fpos - phdrs_offset,
379 				 tsz)) {
380 			kfree(phdrs);
381 			ret = -EFAULT;
382 			goto out;
383 		}
384 		kfree(phdrs);
385 
386 		buffer += tsz;
387 		buflen -= tsz;
388 		*fpos += tsz;
389 	}
390 
391 	/* ELF note segment. */
392 	if (buflen && *fpos < notes_offset + notes_len) {
393 		struct elf_prstatus prstatus = {};
394 		struct elf_prpsinfo prpsinfo = {
395 			.pr_sname = 'R',
396 			.pr_fname = "vmlinux",
397 		};
398 		char *notes;
399 		size_t i = 0;
400 
401 		strlcpy(prpsinfo.pr_psargs, saved_command_line,
402 			sizeof(prpsinfo.pr_psargs));
403 
404 		notes = kzalloc(notes_len, GFP_KERNEL);
405 		if (!notes) {
406 			ret = -ENOMEM;
407 			goto out;
408 		}
409 
410 		append_kcore_note(notes, &i, CORE_STR, NT_PRSTATUS, &prstatus,
411 				  sizeof(prstatus));
412 		append_kcore_note(notes, &i, CORE_STR, NT_PRPSINFO, &prpsinfo,
413 				  sizeof(prpsinfo));
414 		append_kcore_note(notes, &i, CORE_STR, NT_TASKSTRUCT, current,
415 				  arch_task_struct_size);
416 		/*
417 		 * vmcoreinfo_size is mostly constant after init time, but it
418 		 * can be changed by crash_save_vmcoreinfo(). Racing here with a
419 		 * panic on another CPU before the machine goes down is insanely
420 		 * unlikely, but it's better to not leave potential buffer
421 		 * overflows lying around, regardless.
422 		 */
423 		append_kcore_note(notes, &i, VMCOREINFO_NOTE_NAME, 0,
424 				  vmcoreinfo_data,
425 				  min(vmcoreinfo_size, notes_len - i));
426 
427 		tsz = min_t(size_t, buflen, notes_offset + notes_len - *fpos);
428 		if (copy_to_user(buffer, notes + *fpos - notes_offset, tsz)) {
429 			kfree(notes);
430 			ret = -EFAULT;
431 			goto out;
432 		}
433 		kfree(notes);
434 
435 		buffer += tsz;
436 		buflen -= tsz;
437 		*fpos += tsz;
438 	}
439 
440 	/*
441 	 * Check to see if our file offset matches with any of
442 	 * the addresses in the elf_phdr on our list.
443 	 */
444 	start = kc_offset_to_vaddr(*fpos - data_offset);
445 	if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
446 		tsz = buflen;
447 
448 	m = NULL;
449 	while (buflen) {
450 		/*
451 		 * If this is the first iteration or the address is not within
452 		 * the previous entry, search for a matching entry.
453 		 */
454 		if (!m || start < m->addr || start >= m->addr + m->size) {
455 			list_for_each_entry(m, &kclist_head, list) {
456 				if (start >= m->addr &&
457 				    start < m->addr + m->size)
458 					break;
459 			}
460 		}
461 
462 		if (&m->list == &kclist_head) {
463 			if (clear_user(buffer, tsz)) {
464 				ret = -EFAULT;
465 				goto out;
466 			}
467 			m = NULL;	/* skip the list anchor */
468 		} else if (m->type == KCORE_VMALLOC) {
469 			vread(buf, (char *)start, tsz);
470 			/* we have to zero-fill user buffer even if no read */
471 			if (copy_to_user(buffer, buf, tsz)) {
472 				ret = -EFAULT;
473 				goto out;
474 			}
475 		} else if (m->type == KCORE_USER) {
476 			/* User page is handled prior to normal kernel page: */
477 			if (copy_to_user(buffer, (char *)start, tsz)) {
478 				ret = -EFAULT;
479 				goto out;
480 			}
481 		} else {
482 			if (kern_addr_valid(start)) {
483 				/*
484 				 * Using bounce buffer to bypass the
485 				 * hardened user copy kernel text checks.
486 				 */
487 				if (probe_kernel_read(buf, (void *) start, tsz)) {
488 					if (clear_user(buffer, tsz)) {
489 						ret = -EFAULT;
490 						goto out;
491 					}
492 				} else {
493 					if (copy_to_user(buffer, buf, tsz)) {
494 						ret = -EFAULT;
495 						goto out;
496 					}
497 				}
498 			} else {
499 				if (clear_user(buffer, tsz)) {
500 					ret = -EFAULT;
501 					goto out;
502 				}
503 			}
504 		}
505 		buflen -= tsz;
506 		*fpos += tsz;
507 		buffer += tsz;
508 		start += tsz;
509 		tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
510 	}
511 
512 out:
513 	up_read(&kclist_lock);
514 	if (ret)
515 		return ret;
516 	return orig_buflen - buflen;
517 }
518 
519 static int open_kcore(struct inode *inode, struct file *filp)
520 {
521 	if (!capable(CAP_SYS_RAWIO))
522 		return -EPERM;
523 
524 	filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
525 	if (!filp->private_data)
526 		return -ENOMEM;
527 
528 	if (kcore_need_update)
529 		kcore_update_ram();
530 	if (i_size_read(inode) != proc_root_kcore->size) {
531 		inode_lock(inode);
532 		i_size_write(inode, proc_root_kcore->size);
533 		inode_unlock(inode);
534 	}
535 	return 0;
536 }
537 
538 static int release_kcore(struct inode *inode, struct file *file)
539 {
540 	kfree(file->private_data);
541 	return 0;
542 }
543 
544 static const struct file_operations proc_kcore_operations = {
545 	.read		= read_kcore,
546 	.open		= open_kcore,
547 	.release	= release_kcore,
548 	.llseek		= default_llseek,
549 };
550 
551 /* just remember that we have to update kcore */
552 static int __meminit kcore_callback(struct notifier_block *self,
553 				    unsigned long action, void *arg)
554 {
555 	switch (action) {
556 	case MEM_ONLINE:
557 	case MEM_OFFLINE:
558 		kcore_need_update = 1;
559 		break;
560 	}
561 	return NOTIFY_OK;
562 }
563 
564 static struct notifier_block kcore_callback_nb __meminitdata = {
565 	.notifier_call = kcore_callback,
566 	.priority = 0,
567 };
568 
569 static struct kcore_list kcore_vmalloc;
570 
571 #ifdef CONFIG_ARCH_PROC_KCORE_TEXT
572 static struct kcore_list kcore_text;
573 /*
574  * If defined, special segment is used for mapping kernel text instead of
575  * direct-map area. We need to create special TEXT section.
576  */
577 static void __init proc_kcore_text_init(void)
578 {
579 	kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
580 }
581 #else
582 static void __init proc_kcore_text_init(void)
583 {
584 }
585 #endif
586 
587 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
588 /*
589  * MODULES_VADDR has no intersection with VMALLOC_ADDR.
590  */
591 struct kcore_list kcore_modules;
592 static void __init add_modules_range(void)
593 {
594 	if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) {
595 		kclist_add(&kcore_modules, (void *)MODULES_VADDR,
596 			MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
597 	}
598 }
599 #else
600 static void __init add_modules_range(void)
601 {
602 }
603 #endif
604 
605 static int __init proc_kcore_init(void)
606 {
607 	proc_root_kcore = proc_create("kcore", S_IRUSR, NULL,
608 				      &proc_kcore_operations);
609 	if (!proc_root_kcore) {
610 		pr_err("couldn't create /proc/kcore\n");
611 		return 0; /* Always returns 0. */
612 	}
613 	/* Store text area if it's special */
614 	proc_kcore_text_init();
615 	/* Store vmalloc area */
616 	kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
617 		VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
618 	add_modules_range();
619 	/* Store direct-map area from physical memory map */
620 	kcore_update_ram();
621 	register_hotmemory_notifier(&kcore_callback_nb);
622 
623 	return 0;
624 }
625 fs_initcall(proc_kcore_init);
626