Linux Audio

Check our new training course

Loading...
v6.2
  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/vmalloc.h>
 22#include <linux/highmem.h>
 23#include <linux/printk.h>
 24#include <linux/memblock.h>
 25#include <linux/init.h>
 26#include <linux/slab.h>
 27#include <linux/uaccess.h>
 28#include <asm/io.h>
 29#include <linux/list.h>
 30#include <linux/ioport.h>
 31#include <linux/memory.h>
 32#include <linux/sched/task.h>
 33#include <linux/security.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
 43static 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
 53static LIST_HEAD(kclist_head);
 54static DECLARE_RWSEM(kclist_lock);
 55static int kcore_need_update = 1;
 56
 57/*
 58 * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
 59 * Same as oldmem_pfn_is_ram in vmcore
 60 */
 61static int (*mem_pfn_is_ram)(unsigned long pfn);
 62
 63int __init register_mem_pfn_is_ram(int (*fn)(unsigned long pfn))
 64{
 65	if (mem_pfn_is_ram)
 66		return -EBUSY;
 67	mem_pfn_is_ram = fn;
 68	return 0;
 69}
 70
 71static int pfn_is_ram(unsigned long pfn)
 72{
 73	if (mem_pfn_is_ram)
 74		return mem_pfn_is_ram(pfn);
 75	else
 76		return 1;
 77}
 78
 79/* This doesn't grab kclist_lock, so it should only be used at init time. */
 80void __init kclist_add(struct kcore_list *new, void *addr, size_t size,
 81		       int type)
 82{
 83	new->addr = (unsigned long)addr;
 84	new->size = size;
 85	new->type = type;
 86
 87	list_add_tail(&new->list, &kclist_head);
 88}
 89
 90static size_t get_kcore_size(int *nphdr, size_t *phdrs_len, size_t *notes_len,
 91			     size_t *data_offset)
 92{
 93	size_t try, size;
 94	struct kcore_list *m;
 95
 96	*nphdr = 1; /* PT_NOTE */
 97	size = 0;
 98
 99	list_for_each_entry(m, &kclist_head, list) {
100		try = kc_vaddr_to_offset((size_t)m->addr + m->size);
101		if (try > size)
102			size = try;
103		*nphdr = *nphdr + 1;
104	}
105
106	*phdrs_len = *nphdr * sizeof(struct elf_phdr);
107	*notes_len = (4 * sizeof(struct elf_note) +
108		      3 * ALIGN(sizeof(CORE_STR), 4) +
109		      VMCOREINFO_NOTE_NAME_BYTES +
110		      ALIGN(sizeof(struct elf_prstatus), 4) +
111		      ALIGN(sizeof(struct elf_prpsinfo), 4) +
112		      ALIGN(arch_task_struct_size, 4) +
113		      ALIGN(vmcoreinfo_size, 4));
114	*data_offset = PAGE_ALIGN(sizeof(struct elfhdr) + *phdrs_len +
115				  *notes_len);
116	return *data_offset + size;
117}
118
119#ifdef CONFIG_HIGHMEM
120/*
121 * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
122 * because memory hole is not as big as !HIGHMEM case.
123 * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
124 */
125static int kcore_ram_list(struct list_head *head)
126{
127	struct kcore_list *ent;
128
129	ent = kmalloc(sizeof(*ent), GFP_KERNEL);
130	if (!ent)
131		return -ENOMEM;
132	ent->addr = (unsigned long)__va(0);
133	ent->size = max_low_pfn << PAGE_SHIFT;
134	ent->type = KCORE_RAM;
135	list_add(&ent->list, head);
136	return 0;
137}
138
139#else /* !CONFIG_HIGHMEM */
140
141#ifdef CONFIG_SPARSEMEM_VMEMMAP
142/* calculate vmemmap's address from given system ram pfn and register it */
143static int
144get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
145{
146	unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
147	unsigned long nr_pages = ent->size >> PAGE_SHIFT;
148	unsigned long start, end;
149	struct kcore_list *vmm, *tmp;
150
151
152	start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
153	end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
154	end = PAGE_ALIGN(end);
155	/* overlap check (because we have to align page */
156	list_for_each_entry(tmp, head, list) {
157		if (tmp->type != KCORE_VMEMMAP)
158			continue;
159		if (start < tmp->addr + tmp->size)
160			if (end > tmp->addr)
161				end = tmp->addr;
162	}
163	if (start < end) {
164		vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
165		if (!vmm)
166			return 0;
167		vmm->addr = start;
168		vmm->size = end - start;
169		vmm->type = KCORE_VMEMMAP;
170		list_add_tail(&vmm->list, head);
171	}
172	return 1;
173
174}
175#else
176static int
177get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
178{
179	return 1;
180}
181
182#endif
183
184static int
185kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
186{
187	struct list_head *head = (struct list_head *)arg;
188	struct kcore_list *ent;
189	struct page *p;
190
191	if (!pfn_valid(pfn))
192		return 1;
193
194	p = pfn_to_page(pfn);
 
 
195
196	ent = kmalloc(sizeof(*ent), GFP_KERNEL);
197	if (!ent)
198		return -ENOMEM;
199	ent->addr = (unsigned long)page_to_virt(p);
200	ent->size = nr_pages << PAGE_SHIFT;
201
202	if (!virt_addr_valid(ent->addr))
203		goto free_out;
204
205	/* cut not-mapped area. ....from ppc-32 code. */
206	if (ULONG_MAX - ent->addr < ent->size)
207		ent->size = ULONG_MAX - ent->addr;
208
209	/*
210	 * We've already checked virt_addr_valid so we know this address
211	 * is a valid pointer, therefore we can check against it to determine
212	 * if we need to trim
213	 */
214	if (VMALLOC_START > ent->addr) {
215		if (VMALLOC_START - ent->addr < ent->size)
216			ent->size = VMALLOC_START - ent->addr;
217	}
218
219	ent->type = KCORE_RAM;
220	list_add_tail(&ent->list, head);
221
222	if (!get_sparsemem_vmemmap_info(ent, head)) {
223		list_del(&ent->list);
224		goto free_out;
225	}
226
227	return 0;
228free_out:
229	kfree(ent);
230	return 1;
231}
232
233static int kcore_ram_list(struct list_head *list)
234{
235	int nid, ret;
236	unsigned long end_pfn;
237
238	/* Not inialized....update now */
239	/* find out "max pfn" */
240	end_pfn = 0;
241	for_each_node_state(nid, N_MEMORY) {
242		unsigned long node_end;
243		node_end = node_end_pfn(nid);
244		if (end_pfn < node_end)
245			end_pfn = node_end;
246	}
247	/* scan 0 to max_pfn */
248	ret = walk_system_ram_range(0, end_pfn, list, kclist_add_private);
249	if (ret)
250		return -ENOMEM;
251	return 0;
252}
253#endif /* CONFIG_HIGHMEM */
254
255static int kcore_update_ram(void)
256{
257	LIST_HEAD(list);
258	LIST_HEAD(garbage);
259	int nphdr;
260	size_t phdrs_len, notes_len, data_offset;
261	struct kcore_list *tmp, *pos;
262	int ret = 0;
263
264	down_write(&kclist_lock);
265	if (!xchg(&kcore_need_update, 0))
266		goto out;
267
268	ret = kcore_ram_list(&list);
269	if (ret) {
270		/* Couldn't get the RAM list, try again next time. */
271		WRITE_ONCE(kcore_need_update, 1);
272		list_splice_tail(&list, &garbage);
273		goto out;
274	}
275
276	list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
277		if (pos->type == KCORE_RAM || pos->type == KCORE_VMEMMAP)
278			list_move(&pos->list, &garbage);
279	}
280	list_splice_tail(&list, &kclist_head);
281
282	proc_root_kcore->size = get_kcore_size(&nphdr, &phdrs_len, &notes_len,
283					       &data_offset);
284
285out:
286	up_write(&kclist_lock);
287	list_for_each_entry_safe(pos, tmp, &garbage, list) {
288		list_del(&pos->list);
289		kfree(pos);
290	}
291	return ret;
292}
293
294static void append_kcore_note(char *notes, size_t *i, const char *name,
295			      unsigned int type, const void *desc,
296			      size_t descsz)
297{
298	struct elf_note *note = (struct elf_note *)&notes[*i];
299
300	note->n_namesz = strlen(name) + 1;
301	note->n_descsz = descsz;
302	note->n_type = type;
303	*i += sizeof(*note);
304	memcpy(&notes[*i], name, note->n_namesz);
305	*i = ALIGN(*i + note->n_namesz, 4);
306	memcpy(&notes[*i], desc, descsz);
307	*i = ALIGN(*i + descsz, 4);
308}
309
310static ssize_t
311read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
312{
313	char *buf = file->private_data;
314	size_t phdrs_offset, notes_offset, data_offset;
315	size_t page_offline_frozen = 1;
316	size_t phdrs_len, notes_len;
317	struct kcore_list *m;
318	size_t tsz;
319	int nphdr;
320	unsigned long start;
321	size_t orig_buflen = buflen;
322	int ret = 0;
323
324	down_read(&kclist_lock);
325	/*
326	 * Don't race against drivers that set PageOffline() and expect no
327	 * further page access.
328	 */
329	page_offline_freeze();
330
331	get_kcore_size(&nphdr, &phdrs_len, &notes_len, &data_offset);
332	phdrs_offset = sizeof(struct elfhdr);
333	notes_offset = phdrs_offset + phdrs_len;
334
335	/* ELF file header. */
336	if (buflen && *fpos < sizeof(struct elfhdr)) {
337		struct elfhdr ehdr = {
338			.e_ident = {
339				[EI_MAG0] = ELFMAG0,
340				[EI_MAG1] = ELFMAG1,
341				[EI_MAG2] = ELFMAG2,
342				[EI_MAG3] = ELFMAG3,
343				[EI_CLASS] = ELF_CLASS,
344				[EI_DATA] = ELF_DATA,
345				[EI_VERSION] = EV_CURRENT,
346				[EI_OSABI] = ELF_OSABI,
347			},
348			.e_type = ET_CORE,
349			.e_machine = ELF_ARCH,
350			.e_version = EV_CURRENT,
351			.e_phoff = sizeof(struct elfhdr),
352			.e_flags = ELF_CORE_EFLAGS,
353			.e_ehsize = sizeof(struct elfhdr),
354			.e_phentsize = sizeof(struct elf_phdr),
355			.e_phnum = nphdr,
356		};
357
358		tsz = min_t(size_t, buflen, sizeof(struct elfhdr) - *fpos);
359		if (copy_to_user(buffer, (char *)&ehdr + *fpos, tsz)) {
360			ret = -EFAULT;
361			goto out;
362		}
363
364		buffer += tsz;
365		buflen -= tsz;
366		*fpos += tsz;
367	}
368
369	/* ELF program headers. */
370	if (buflen && *fpos < phdrs_offset + phdrs_len) {
371		struct elf_phdr *phdrs, *phdr;
372
373		phdrs = kzalloc(phdrs_len, GFP_KERNEL);
374		if (!phdrs) {
375			ret = -ENOMEM;
376			goto out;
377		}
378
379		phdrs[0].p_type = PT_NOTE;
380		phdrs[0].p_offset = notes_offset;
381		phdrs[0].p_filesz = notes_len;
382
383		phdr = &phdrs[1];
384		list_for_each_entry(m, &kclist_head, list) {
385			phdr->p_type = PT_LOAD;
386			phdr->p_flags = PF_R | PF_W | PF_X;
387			phdr->p_offset = kc_vaddr_to_offset(m->addr) + data_offset;
388			phdr->p_vaddr = (size_t)m->addr;
389			if (m->type == KCORE_RAM)
 
 
 
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		struct page *page;
474		unsigned long pfn;
475
476		/*
477		 * If this is the first iteration or the address is not within
478		 * the previous entry, search for a matching entry.
479		 */
480		if (!m || start < m->addr || start >= m->addr + m->size) {
481			struct kcore_list *iter;
482
483			m = NULL;
484			list_for_each_entry(iter, &kclist_head, list) {
485				if (start >= iter->addr &&
486				    start < iter->addr + iter->size) {
487					m = iter;
488					break;
489				}
490			}
491		}
492
493		if (page_offline_frozen++ % MAX_ORDER_NR_PAGES == 0) {
494			page_offline_thaw();
495			cond_resched();
496			page_offline_freeze();
497		}
498
499		if (!m) {
500			if (clear_user(buffer, tsz)) {
501				ret = -EFAULT;
502				goto out;
503			}
504			goto skip;
505		}
506
507		switch (m->type) {
508		case KCORE_VMALLOC:
 
 
509			vread(buf, (char *)start, tsz);
510			/* we have to zero-fill user buffer even if no read */
511			if (copy_to_user(buffer, buf, tsz)) {
512				ret = -EFAULT;
513				goto out;
514			}
515			break;
516		case KCORE_USER:
517			/* User page is handled prior to normal kernel page: */
518			if (copy_to_user(buffer, (char *)start, tsz)) {
519				ret = -EFAULT;
520				goto out;
521			}
522			break;
523		case KCORE_RAM:
524			pfn = __pa(start) >> PAGE_SHIFT;
525			page = pfn_to_online_page(pfn);
526
527			/*
528			 * Don't read offline sections, logically offline pages
529			 * (e.g., inflated in a balloon), hwpoisoned pages,
530			 * and explicitly excluded physical ranges.
531			 */
532			if (!page || PageOffline(page) ||
533			    is_page_hwpoison(page) || !pfn_is_ram(pfn)) {
534				if (clear_user(buffer, tsz)) {
535					ret = -EFAULT;
536					goto out;
537				}
538				break;
539			}
540			fallthrough;
541		case KCORE_VMEMMAP:
542		case KCORE_TEXT:
543			/*
544			 * Using bounce buffer to bypass the
545			 * hardened user copy kernel text checks.
546			 */
547			if (copy_from_kernel_nofault(buf, (void *)start, tsz)) {
548				if (clear_user(buffer, tsz)) {
549					ret = -EFAULT;
550					goto out;
551				}
552			} else {
553				if (copy_to_user(buffer, buf, tsz)) {
554					ret = -EFAULT;
555					goto out;
556				}
557			}
558			break;
559		default:
560			pr_warn_once("Unhandled KCORE type: %d\n", m->type);
561			if (clear_user(buffer, tsz)) {
562				ret = -EFAULT;
563				goto out;
564			}
565		}
566skip:
567		buflen -= tsz;
568		*fpos += tsz;
569		buffer += tsz;
570		start += tsz;
571		tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
572	}
573
574out:
575	page_offline_thaw();
576	up_read(&kclist_lock);
577	if (ret)
578		return ret;
579	return orig_buflen - buflen;
580}
581
582static int open_kcore(struct inode *inode, struct file *filp)
583{
584	int ret = security_locked_down(LOCKDOWN_KCORE);
585
586	if (!capable(CAP_SYS_RAWIO))
587		return -EPERM;
588
589	if (ret)
590		return ret;
591
592	filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
593	if (!filp->private_data)
594		return -ENOMEM;
595
596	if (kcore_need_update)
597		kcore_update_ram();
598	if (i_size_read(inode) != proc_root_kcore->size) {
599		inode_lock(inode);
600		i_size_write(inode, proc_root_kcore->size);
601		inode_unlock(inode);
602	}
603	return 0;
604}
605
606static int release_kcore(struct inode *inode, struct file *file)
607{
608	kfree(file->private_data);
609	return 0;
610}
611
612static const struct proc_ops kcore_proc_ops = {
613	.proc_read	= read_kcore,
614	.proc_open	= open_kcore,
615	.proc_release	= release_kcore,
616	.proc_lseek	= default_llseek,
617};
618
619/* just remember that we have to update kcore */
620static int __meminit kcore_callback(struct notifier_block *self,
621				    unsigned long action, void *arg)
622{
623	switch (action) {
624	case MEM_ONLINE:
625	case MEM_OFFLINE:
626		kcore_need_update = 1;
627		break;
628	}
629	return NOTIFY_OK;
630}
631
 
 
 
 
632
633static struct kcore_list kcore_vmalloc;
634
635#ifdef CONFIG_ARCH_PROC_KCORE_TEXT
636static struct kcore_list kcore_text;
637/*
638 * If defined, special segment is used for mapping kernel text instead of
639 * direct-map area. We need to create special TEXT section.
640 */
641static void __init proc_kcore_text_init(void)
642{
643	kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
644}
645#else
646static void __init proc_kcore_text_init(void)
647{
648}
649#endif
650
651#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
652/*
653 * MODULES_VADDR has no intersection with VMALLOC_ADDR.
654 */
655static struct kcore_list kcore_modules;
656static void __init add_modules_range(void)
657{
658	if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) {
659		kclist_add(&kcore_modules, (void *)MODULES_VADDR,
660			MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
661	}
662}
663#else
664static void __init add_modules_range(void)
665{
666}
667#endif
668
669static int __init proc_kcore_init(void)
670{
671	proc_root_kcore = proc_create("kcore", S_IRUSR, NULL, &kcore_proc_ops);
 
672	if (!proc_root_kcore) {
673		pr_err("couldn't create /proc/kcore\n");
674		return 0; /* Always returns 0. */
675	}
676	/* Store text area if it's special */
677	proc_kcore_text_init();
678	/* Store vmalloc area */
679	kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
680		VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
681	add_modules_range();
682	/* Store direct-map area from physical memory map */
683	kcore_update_ram();
684	hotplug_memory_notifier(kcore_callback, DEFAULT_CALLBACK_PRI);
685
686	return 0;
687}
688fs_initcall(proc_kcore_init);
v5.4
  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
 44static 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
 54static LIST_HEAD(kclist_head);
 55static DECLARE_RWSEM(kclist_lock);
 56static 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 */
 62static int (*mem_pfn_is_ram)(unsigned long pfn);
 63
 64int __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
 72static 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. */
 81void __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
 91static 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 */
126static 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 */
144static int
145get_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
177static int
178get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
179{
180	return 1;
181}
182
183#endif
184
185static int
186kclist_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;
231free_out:
232	kfree(ent);
233	return 1;
234}
235
236static 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
258static 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
288out:
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
297static 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
313static ssize_t
314read_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 (probe_kernel_read(buf, (void *) start, tsz)) {
516					if (clear_user(buffer, tsz)) {
517						ret = -EFAULT;
518						goto out;
519					}
520				} else {
521					if (copy_to_user(buffer, buf, tsz)) {
522						ret = -EFAULT;
523						goto out;
524					}
 
 
 
 
 
 
 
 
 
 
 
 
 
525				}
526			} else {
527				if (clear_user(buffer, tsz)) {
528					ret = -EFAULT;
529					goto out;
530				}
531			}
 
 
 
 
 
 
 
532		}
 
533		buflen -= tsz;
534		*fpos += tsz;
535		buffer += tsz;
536		start += tsz;
537		tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
538	}
539
540out:
 
541	up_read(&kclist_lock);
542	if (ret)
543		return ret;
544	return orig_buflen - buflen;
545}
546
547static int open_kcore(struct inode *inode, struct file *filp)
548{
549	int ret = security_locked_down(LOCKDOWN_KCORE);
550
551	if (!capable(CAP_SYS_RAWIO))
552		return -EPERM;
553
554	if (ret)
555		return ret;
556
557	filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
558	if (!filp->private_data)
559		return -ENOMEM;
560
561	if (kcore_need_update)
562		kcore_update_ram();
563	if (i_size_read(inode) != proc_root_kcore->size) {
564		inode_lock(inode);
565		i_size_write(inode, proc_root_kcore->size);
566		inode_unlock(inode);
567	}
568	return 0;
569}
570
571static int release_kcore(struct inode *inode, struct file *file)
572{
573	kfree(file->private_data);
574	return 0;
575}
576
577static const struct file_operations proc_kcore_operations = {
578	.read		= read_kcore,
579	.open		= open_kcore,
580	.release	= release_kcore,
581	.llseek		= default_llseek,
582};
583
584/* just remember that we have to update kcore */
585static int __meminit kcore_callback(struct notifier_block *self,
586				    unsigned long action, void *arg)
587{
588	switch (action) {
589	case MEM_ONLINE:
590	case MEM_OFFLINE:
591		kcore_need_update = 1;
592		break;
593	}
594	return NOTIFY_OK;
595}
596
597static struct notifier_block kcore_callback_nb __meminitdata = {
598	.notifier_call = kcore_callback,
599	.priority = 0,
600};
601
602static struct kcore_list kcore_vmalloc;
603
604#ifdef CONFIG_ARCH_PROC_KCORE_TEXT
605static struct kcore_list kcore_text;
606/*
607 * If defined, special segment is used for mapping kernel text instead of
608 * direct-map area. We need to create special TEXT section.
609 */
610static void __init proc_kcore_text_init(void)
611{
612	kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
613}
614#else
615static void __init proc_kcore_text_init(void)
616{
617}
618#endif
619
620#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
621/*
622 * MODULES_VADDR has no intersection with VMALLOC_ADDR.
623 */
624static struct kcore_list kcore_modules;
625static void __init add_modules_range(void)
626{
627	if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) {
628		kclist_add(&kcore_modules, (void *)MODULES_VADDR,
629			MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
630	}
631}
632#else
633static void __init add_modules_range(void)
634{
635}
636#endif
637
638static int __init proc_kcore_init(void)
639{
640	proc_root_kcore = proc_create("kcore", S_IRUSR, NULL,
641				      &proc_kcore_operations);
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}
658fs_initcall(proc_kcore_init);