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);
v4.17
  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/mm.h>
 14#include <linux/proc_fs.h>
 15#include <linux/kcore.h>
 16#include <linux/user.h>
 17#include <linux/capability.h>
 18#include <linux/elf.h>
 19#include <linux/elfcore.h>
 20#include <linux/notifier.h>
 21#include <linux/vmalloc.h>
 22#include <linux/highmem.h>
 23#include <linux/printk.h>
 24#include <linux/bootmem.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 <asm/sections.h>
 34#include "internal.h"
 35
 36#define CORE_STR "CORE"
 37
 38#ifndef ELF_CORE_EFLAGS
 39#define ELF_CORE_EFLAGS	0
 40#endif
 41
 42static struct proc_dir_entry *proc_root_kcore;
 43
 44
 45#ifndef kc_vaddr_to_offset
 46#define	kc_vaddr_to_offset(v) ((v) - PAGE_OFFSET)
 47#endif
 48#ifndef	kc_offset_to_vaddr
 49#define	kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET)
 50#endif
 51
 52/* An ELF note in memory */
 53struct memelfnote
 
 
 
 
 
 
 
 
 
 54{
 55	const char *name;
 56	int type;
 57	unsigned int datasz;
 58	void *data;
 59};
 60
 61static LIST_HEAD(kclist_head);
 62static DEFINE_RWLOCK(kclist_lock);
 63static int kcore_need_update = 1;
 
 
 
 
 64
 65void
 66kclist_add(struct kcore_list *new, void *addr, size_t size, int type)
 
 67{
 68	new->addr = (unsigned long)addr;
 69	new->size = size;
 70	new->type = type;
 71
 72	write_lock(&kclist_lock);
 73	list_add_tail(&new->list, &kclist_head);
 74	write_unlock(&kclist_lock);
 75}
 76
 77static size_t get_kcore_size(int *nphdr, size_t *elf_buflen)
 
 78{
 79	size_t try, size;
 80	struct kcore_list *m;
 81
 82	*nphdr = 1; /* PT_NOTE */
 83	size = 0;
 84
 85	list_for_each_entry(m, &kclist_head, list) {
 86		try = kc_vaddr_to_offset((size_t)m->addr + m->size);
 87		if (try > size)
 88			size = try;
 89		*nphdr = *nphdr + 1;
 90	}
 91	*elf_buflen =	sizeof(struct elfhdr) + 
 92			(*nphdr + 2)*sizeof(struct elf_phdr) + 
 93			3 * ((sizeof(struct elf_note)) +
 94			     roundup(sizeof(CORE_STR), 4)) +
 95			roundup(sizeof(struct elf_prstatus), 4) +
 96			roundup(sizeof(struct elf_prpsinfo), 4) +
 97			roundup(arch_task_struct_size, 4);
 98	*elf_buflen = PAGE_ALIGN(*elf_buflen);
 99	return size + *elf_buflen;
100}
101
102static void free_kclist_ents(struct list_head *head)
103{
104	struct kcore_list *tmp, *pos;
105
106	list_for_each_entry_safe(pos, tmp, head, list) {
107		list_del(&pos->list);
108		kfree(pos);
109	}
110}
111/*
112 * Replace all KCORE_RAM/KCORE_VMEMMAP information with passed list.
113 */
114static void __kcore_update_ram(struct list_head *list)
115{
116	int nphdr;
117	size_t size;
118	struct kcore_list *tmp, *pos;
119	LIST_HEAD(garbage);
120
121	write_lock(&kclist_lock);
122	if (kcore_need_update) {
123		list_for_each_entry_safe(pos, tmp, &kclist_head, list) {
124			if (pos->type == KCORE_RAM
125				|| pos->type == KCORE_VMEMMAP)
126				list_move(&pos->list, &garbage);
127		}
128		list_splice_tail(list, &kclist_head);
129	} else
130		list_splice(list, &garbage);
131	kcore_need_update = 0;
132	proc_root_kcore->size = get_kcore_size(&nphdr, &size);
133	write_unlock(&kclist_lock);
134
135	free_kclist_ents(&garbage);
136}
137
138
139#ifdef CONFIG_HIGHMEM
140/*
141 * If no highmem, we can assume [0...max_low_pfn) continuous range of memory
142 * because memory hole is not as big as !HIGHMEM case.
143 * (HIGHMEM is special because part of memory is _invisible_ from the kernel.)
144 */
145static int kcore_update_ram(void)
146{
147	LIST_HEAD(head);
148	struct kcore_list *ent;
149	int ret = 0;
150
151	ent = kmalloc(sizeof(*ent), GFP_KERNEL);
152	if (!ent)
153		return -ENOMEM;
154	ent->addr = (unsigned long)__va(0);
155	ent->size = max_low_pfn << PAGE_SHIFT;
156	ent->type = KCORE_RAM;
157	list_add(&ent->list, &head);
158	__kcore_update_ram(&head);
159	return ret;
160}
161
162#else /* !CONFIG_HIGHMEM */
163
164#ifdef CONFIG_SPARSEMEM_VMEMMAP
165/* calculate vmemmap's address from given system ram pfn and register it */
166static int
167get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
168{
169	unsigned long pfn = __pa(ent->addr) >> PAGE_SHIFT;
170	unsigned long nr_pages = ent->size >> PAGE_SHIFT;
171	unsigned long start, end;
172	struct kcore_list *vmm, *tmp;
173
174
175	start = ((unsigned long)pfn_to_page(pfn)) & PAGE_MASK;
176	end = ((unsigned long)pfn_to_page(pfn + nr_pages)) - 1;
177	end = PAGE_ALIGN(end);
178	/* overlap check (because we have to align page */
179	list_for_each_entry(tmp, head, list) {
180		if (tmp->type != KCORE_VMEMMAP)
181			continue;
182		if (start < tmp->addr + tmp->size)
183			if (end > tmp->addr)
184				end = tmp->addr;
185	}
186	if (start < end) {
187		vmm = kmalloc(sizeof(*vmm), GFP_KERNEL);
188		if (!vmm)
189			return 0;
190		vmm->addr = start;
191		vmm->size = end - start;
192		vmm->type = KCORE_VMEMMAP;
193		list_add_tail(&vmm->list, head);
194	}
195	return 1;
196
197}
198#else
199static int
200get_sparsemem_vmemmap_info(struct kcore_list *ent, struct list_head *head)
201{
202	return 1;
203}
204
205#endif
206
207static int
208kclist_add_private(unsigned long pfn, unsigned long nr_pages, void *arg)
209{
210	struct list_head *head = (struct list_head *)arg;
211	struct kcore_list *ent;
212	struct page *p;
213
214	if (!pfn_valid(pfn))
215		return 1;
216
217	p = pfn_to_page(pfn);
218	if (!memmap_valid_within(pfn, p, page_zone(p)))
219		return 1;
220
221	ent = kmalloc(sizeof(*ent), GFP_KERNEL);
222	if (!ent)
223		return -ENOMEM;
224	ent->addr = (unsigned long)page_to_virt(p);
225	ent->size = nr_pages << PAGE_SHIFT;
226
227	if (!virt_addr_valid(ent->addr))
228		goto free_out;
229
230	/* cut not-mapped area. ....from ppc-32 code. */
231	if (ULONG_MAX - ent->addr < ent->size)
232		ent->size = ULONG_MAX - ent->addr;
233
234	/*
235	 * We've already checked virt_addr_valid so we know this address
236	 * is a valid pointer, therefore we can check against it to determine
237	 * if we need to trim
238	 */
239	if (VMALLOC_START > ent->addr) {
240		if (VMALLOC_START - ent->addr < ent->size)
241			ent->size = VMALLOC_START - ent->addr;
242	}
243
244	ent->type = KCORE_RAM;
245	list_add_tail(&ent->list, head);
246
247	if (!get_sparsemem_vmemmap_info(ent, head)) {
248		list_del(&ent->list);
249		goto free_out;
250	}
251
252	return 0;
253free_out:
254	kfree(ent);
255	return 1;
256}
257
258static int kcore_update_ram(void)
259{
260	int nid, ret;
261	unsigned long end_pfn;
262	LIST_HEAD(head);
263
264	/* Not inialized....update now */
265	/* find out "max pfn" */
266	end_pfn = 0;
267	for_each_node_state(nid, N_MEMORY) {
268		unsigned long node_end;
269		node_end = node_end_pfn(nid);
270		if (end_pfn < node_end)
271			end_pfn = node_end;
272	}
273	/* scan 0 to max_pfn */
274	ret = walk_system_ram_range(0, end_pfn, &head, kclist_add_private);
275	if (ret) {
276		free_kclist_ents(&head);
277		return -ENOMEM;
278	}
279	__kcore_update_ram(&head);
280	return ret;
281}
282#endif /* CONFIG_HIGHMEM */
283
284/*****************************************************************************/
285/*
286 * determine size of ELF note
287 */
288static int notesize(struct memelfnote *en)
289{
290	int sz;
 
 
 
 
 
291
292	sz = sizeof(struct elf_note);
293	sz += roundup((strlen(en->name) + 1), 4);
294	sz += roundup(en->datasz, 4);
295
296	return sz;
297} /* end notesize() */
 
 
 
 
 
298
299/*****************************************************************************/
300/*
301 * store a note in the header buffer
302 */
303static char *storenote(struct memelfnote *men, char *bufp)
304{
305	struct elf_note en;
306
307#define DUMP_WRITE(addr,nr) do { memcpy(bufp,addr,nr); bufp += nr; } while(0)
 
308
309	en.n_namesz = strlen(men->name) + 1;
310	en.n_descsz = men->datasz;
311	en.n_type = men->type;
 
 
 
 
 
312
313	DUMP_WRITE(&en, sizeof(en));
314	DUMP_WRITE(men->name, en.n_namesz);
 
 
 
 
 
 
 
 
 
 
 
 
 
315
316	/* XXX - cast from long long to long to avoid need for libgcc.a */
317	bufp = (char*) roundup((unsigned long)bufp,4);
318	DUMP_WRITE(men->data, men->datasz);
319	bufp = (char*) roundup((unsigned long)bufp,4);
320
321#undef DUMP_WRITE
322
323	return bufp;
324} /* end storenote() */
325
326/*
327 * store an ELF coredump header in the supplied buffer
328 * nphdr is the number of elf_phdr to insert
329 */
330static void elf_kcore_store_hdr(char *bufp, int nphdr, int dataoff)
331{
332	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
333	struct elf_prpsinfo prpsinfo;	/* NT_PRPSINFO */
334	struct elf_phdr *nhdr, *phdr;
335	struct elfhdr *elf;
336	struct memelfnote notes[3];
337	off_t offset = 0;
338	struct kcore_list *m;
 
 
 
 
 
339
340	/* setup ELF header */
341	elf = (struct elfhdr *) bufp;
342	bufp += sizeof(struct elfhdr);
343	offset += sizeof(struct elfhdr);
344	memcpy(elf->e_ident, ELFMAG, SELFMAG);
345	elf->e_ident[EI_CLASS]	= ELF_CLASS;
346	elf->e_ident[EI_DATA]	= ELF_DATA;
347	elf->e_ident[EI_VERSION]= EV_CURRENT;
348	elf->e_ident[EI_OSABI] = ELF_OSABI;
349	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
350	elf->e_type	= ET_CORE;
351	elf->e_machine	= ELF_ARCH;
352	elf->e_version	= EV_CURRENT;
353	elf->e_entry	= 0;
354	elf->e_phoff	= sizeof(struct elfhdr);
355	elf->e_shoff	= 0;
356	elf->e_flags	= ELF_CORE_EFLAGS;
357	elf->e_ehsize	= sizeof(struct elfhdr);
358	elf->e_phentsize= sizeof(struct elf_phdr);
359	elf->e_phnum	= nphdr;
360	elf->e_shentsize= 0;
361	elf->e_shnum	= 0;
362	elf->e_shstrndx	= 0;
363
364	/* setup ELF PT_NOTE program header */
365	nhdr = (struct elf_phdr *) bufp;
366	bufp += sizeof(struct elf_phdr);
367	offset += sizeof(struct elf_phdr);
368	nhdr->p_type	= PT_NOTE;
369	nhdr->p_offset	= 0;
370	nhdr->p_vaddr	= 0;
371	nhdr->p_paddr	= 0;
372	nhdr->p_filesz	= 0;
373	nhdr->p_memsz	= 0;
374	nhdr->p_flags	= 0;
375	nhdr->p_align	= 0;
376
377	/* setup ELF PT_LOAD program header for every area */
378	list_for_each_entry(m, &kclist_head, list) {
379		phdr = (struct elf_phdr *) bufp;
380		bufp += sizeof(struct elf_phdr);
381		offset += sizeof(struct elf_phdr);
382
383		phdr->p_type	= PT_LOAD;
384		phdr->p_flags	= PF_R|PF_W|PF_X;
385		phdr->p_offset	= kc_vaddr_to_offset(m->addr) + dataoff;
386		phdr->p_vaddr	= (size_t)m->addr;
387		if (m->type == KCORE_RAM || m->type == KCORE_TEXT)
388			phdr->p_paddr	= __pa(m->addr);
389		else
390			phdr->p_paddr	= (elf_addr_t)-1;
391		phdr->p_filesz	= phdr->p_memsz	= m->size;
392		phdr->p_align	= PAGE_SIZE;
393	}
394
395	/*
396	 * Set up the notes in similar form to SVR4 core dumps made
397	 * with info from their /proc.
398	 */
399	nhdr->p_offset	= offset;
400
401	/* set up the process status */
402	notes[0].name = CORE_STR;
403	notes[0].type = NT_PRSTATUS;
404	notes[0].datasz = sizeof(struct elf_prstatus);
405	notes[0].data = &prstatus;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
406
407	memset(&prstatus, 0, sizeof(struct elf_prstatus));
 
 
 
408
409	nhdr->p_filesz	= notesize(&notes[0]);
410	bufp = storenote(&notes[0], bufp);
 
 
 
 
 
 
 
411
412	/* set up the process info */
413	notes[1].name	= CORE_STR;
414	notes[1].type	= NT_PRPSINFO;
415	notes[1].datasz	= sizeof(struct elf_prpsinfo);
416	notes[1].data	= &prpsinfo;
417
418	memset(&prpsinfo, 0, sizeof(struct elf_prpsinfo));
419	prpsinfo.pr_state	= 0;
420	prpsinfo.pr_sname	= 'R';
421	prpsinfo.pr_zomb	= 0;
 
 
 
 
 
 
 
 
 
 
 
 
422
423	strcpy(prpsinfo.pr_fname, "vmlinux");
424	strlcpy(prpsinfo.pr_psargs, saved_command_line, sizeof(prpsinfo.pr_psargs));
 
 
 
 
 
 
425
426	nhdr->p_filesz	+= notesize(&notes[1]);
427	bufp = storenote(&notes[1], bufp);
 
 
428
429	/* set up the task structure */
430	notes[2].name	= CORE_STR;
431	notes[2].type	= NT_TASKSTRUCT;
432	notes[2].datasz	= arch_task_struct_size;
433	notes[2].data	= current;
 
 
 
 
 
 
 
 
 
 
 
 
 
434
435	nhdr->p_filesz	+= notesize(&notes[2]);
436	bufp = storenote(&notes[2], bufp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
437
438} /* end elf_kcore_store_hdr() */
439
440/*****************************************************************************/
441/*
442 * read from the ELF header and then kernel memory
443 */
444static ssize_t
445read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
446{
447	char *buf = file->private_data;
448	ssize_t acc = 0;
449	size_t size, tsz;
450	size_t elf_buflen;
451	int nphdr;
452	unsigned long start;
453
454	read_lock(&kclist_lock);
455	size = get_kcore_size(&nphdr, &elf_buflen);
456
457	if (buflen == 0 || *fpos >= size) {
458		read_unlock(&kclist_lock);
459		return 0;
460	}
461
462	/* trim buflen to not go beyond EOF */
463	if (buflen > size - *fpos)
464		buflen = size - *fpos;
465
466	/* construct an ELF core header if we'll need some of it */
467	if (*fpos < elf_buflen) {
468		char * elf_buf;
469
470		tsz = elf_buflen - *fpos;
471		if (buflen < tsz)
472			tsz = buflen;
473		elf_buf = kzalloc(elf_buflen, GFP_ATOMIC);
474		if (!elf_buf) {
475			read_unlock(&kclist_lock);
476			return -ENOMEM;
477		}
478		elf_kcore_store_hdr(elf_buf, nphdr, elf_buflen);
479		read_unlock(&kclist_lock);
480		if (copy_to_user(buffer, elf_buf + *fpos, tsz)) {
481			kfree(elf_buf);
482			return -EFAULT;
483		}
484		kfree(elf_buf);
485		buflen -= tsz;
486		*fpos += tsz;
487		buffer += tsz;
488		acc += tsz;
489
490		/* leave now if filled buffer already */
491		if (buflen == 0)
492			return acc;
493	} else
494		read_unlock(&kclist_lock);
495
496	/*
497	 * Check to see if our file offset matches with any of
498	 * the addresses in the elf_phdr on our list.
499	 */
500	start = kc_offset_to_vaddr(*fpos - elf_buflen);
501	if ((tsz = (PAGE_SIZE - (start & ~PAGE_MASK))) > buflen)
502		tsz = buflen;
503		
 
504	while (buflen) {
505		struct kcore_list *m;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
506
507		read_lock(&kclist_lock);
508		list_for_each_entry(m, &kclist_head, list) {
509			if (start >= m->addr && start < (m->addr+m->size))
510				break;
 
 
 
 
 
 
 
 
511		}
512		read_unlock(&kclist_lock);
513
514		if (&m->list == &kclist_head) {
515			if (clear_user(buffer, tsz))
516				return -EFAULT;
517		} else if (m->type == KCORE_VMALLOC) {
518			vread(buf, (char *)start, tsz);
519			/* we have to zero-fill user buffer even if no read */
520			if (copy_to_user(buffer, buf, tsz))
521				return -EFAULT;
522		} else if (m->type == KCORE_USER) {
 
 
 
523			/* User page is handled prior to normal kernel page: */
524			if (copy_to_user(buffer, (char *)start, tsz))
525				return -EFAULT;
526		} else {
527			if (kern_addr_valid(start)) {
528				/*
529				 * Using bounce buffer to bypass the
530				 * hardened user copy kernel text checks.
531				 */
532				if (probe_kernel_read(buf, (void *) start, tsz)) {
533					if (clear_user(buffer, tsz))
534						return -EFAULT;
535				} else {
536					if (copy_to_user(buffer, buf, tsz))
537						return -EFAULT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
538				}
539			} else {
540				if (clear_user(buffer, tsz))
541					return -EFAULT;
 
 
 
 
 
 
 
 
 
542			}
543		}
 
544		buflen -= tsz;
545		*fpos += tsz;
546		buffer += tsz;
547		acc += tsz;
548		start += tsz;
549		tsz = (buflen > PAGE_SIZE ? PAGE_SIZE : buflen);
550	}
551
552	return acc;
 
 
 
 
 
553}
554
555
556static int open_kcore(struct inode *inode, struct file *filp)
557{
 
 
558	if (!capable(CAP_SYS_RAWIO))
559		return -EPERM;
560
 
 
 
561	filp->private_data = kmalloc(PAGE_SIZE, GFP_KERNEL);
562	if (!filp->private_data)
563		return -ENOMEM;
564
565	if (kcore_need_update)
566		kcore_update_ram();
567	if (i_size_read(inode) != proc_root_kcore->size) {
568		inode_lock(inode);
569		i_size_write(inode, proc_root_kcore->size);
570		inode_unlock(inode);
571	}
572	return 0;
573}
574
575static int release_kcore(struct inode *inode, struct file *file)
576{
577	kfree(file->private_data);
578	return 0;
579}
580
581static const struct file_operations proc_kcore_operations = {
582	.read		= read_kcore,
583	.open		= open_kcore,
584	.release	= release_kcore,
585	.llseek		= default_llseek,
586};
587
588/* just remember that we have to update kcore */
589static int __meminit kcore_callback(struct notifier_block *self,
590				    unsigned long action, void *arg)
591{
592	switch (action) {
593	case MEM_ONLINE:
594	case MEM_OFFLINE:
595		write_lock(&kclist_lock);
596		kcore_need_update = 1;
597		write_unlock(&kclist_lock);
598	}
599	return NOTIFY_OK;
600}
601
602static struct notifier_block kcore_callback_nb __meminitdata = {
603	.notifier_call = kcore_callback,
604	.priority = 0,
605};
606
607static struct kcore_list kcore_vmalloc;
608
609#ifdef CONFIG_ARCH_PROC_KCORE_TEXT
610static struct kcore_list kcore_text;
611/*
612 * If defined, special segment is used for mapping kernel text instead of
613 * direct-map area. We need to create special TEXT section.
614 */
615static void __init proc_kcore_text_init(void)
616{
617	kclist_add(&kcore_text, _text, _end - _text, KCORE_TEXT);
618}
619#else
620static void __init proc_kcore_text_init(void)
621{
622}
623#endif
624
625#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
626/*
627 * MODULES_VADDR has no intersection with VMALLOC_ADDR.
628 */
629struct kcore_list kcore_modules;
630static void __init add_modules_range(void)
631{
632	if (MODULES_VADDR != VMALLOC_START && MODULES_END != VMALLOC_END) {
633		kclist_add(&kcore_modules, (void *)MODULES_VADDR,
634			MODULES_END - MODULES_VADDR, KCORE_VMALLOC);
635	}
636}
637#else
638static void __init add_modules_range(void)
639{
640}
641#endif
642
643static int __init proc_kcore_init(void)
644{
645	proc_root_kcore = proc_create("kcore", S_IRUSR, NULL,
646				      &proc_kcore_operations);
647	if (!proc_root_kcore) {
648		pr_err("couldn't create /proc/kcore\n");
649		return 0; /* Always returns 0. */
650	}
651	/* Store text area if it's special */
652	proc_kcore_text_init();
653	/* Store vmalloc area */
654	kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
655		VMALLOC_END - VMALLOC_START, KCORE_VMALLOC);
656	add_modules_range();
657	/* Store direct-map area from physical memory map */
658	kcore_update_ram();
659	register_hotmemory_notifier(&kcore_callback_nb);
660
661	return 0;
662}
663fs_initcall(proc_kcore_init);