1/*
2 *	fs/proc/vmcore.c Interface for accessing the crash
3 * 				 dump from the system's previous life.
4 * 	Heavily borrowed from fs/proc/kcore.c
5 *	Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
6 *	Copyright (C) IBM Corporation, 2004. All rights reserved
7 *
8 */
9
10#include <linux/mm.h>
11#include <linux/kcore.h>
12#include <linux/user.h>
13#include <linux/elf.h>
14#include <linux/elfcore.h>
15#include <linux/export.h>
16#include <linux/slab.h>
17#include <linux/highmem.h>
18#include <linux/printk.h>
19#include <linux/bootmem.h>
20#include <linux/init.h>
21#include <linux/crash_dump.h>
22#include <linux/list.h>
23#include <linux/vmalloc.h>
24#include <linux/pagemap.h>
25#include <asm/uaccess.h>
26#include <asm/io.h>
27#include "internal.h"
28
29/* List representing chunks of contiguous memory areas and their offsets in
30 * vmcore file.
31 */
32static LIST_HEAD(vmcore_list);
33
34/* Stores the pointer to the buffer containing kernel elf core headers. */
35static char *elfcorebuf;
36static size_t elfcorebuf_sz;
37static size_t elfcorebuf_sz_orig;
38
39static char *elfnotes_buf;
40static size_t elfnotes_sz;
41
42/* Total size of vmcore file. */
43static u64 vmcore_size;
44
45static struct proc_dir_entry *proc_vmcore;
46
47/*
48 * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
49 * The called function has to take care of module refcounting.
50 */
51static int (*oldmem_pfn_is_ram)(unsigned long pfn);
52
53int register_oldmem_pfn_is_ram(int (*fn)(unsigned long pfn))
54{
55	if (oldmem_pfn_is_ram)
56		return -EBUSY;
57	oldmem_pfn_is_ram = fn;
58	return 0;
59}
60EXPORT_SYMBOL_GPL(register_oldmem_pfn_is_ram);
61
62void unregister_oldmem_pfn_is_ram(void)
63{
64	oldmem_pfn_is_ram = NULL;
65	wmb();
66}
67EXPORT_SYMBOL_GPL(unregister_oldmem_pfn_is_ram);
68
69static int pfn_is_ram(unsigned long pfn)
70{
71	int (*fn)(unsigned long pfn);
72	/* pfn is ram unless fn() checks pagetype */
73	int ret = 1;
74
75	/*
76	 * Ask hypervisor if the pfn is really ram.
77	 * A ballooned page contains no data and reading from such a page
78	 * will cause high load in the hypervisor.
79	 */
80	fn = oldmem_pfn_is_ram;
81	if (fn)
82		ret = fn(pfn);
83
84	return ret;
85}
86
87/* Reads a page from the oldmem device from given offset. */
88static ssize_t read_from_oldmem(char *buf, size_t count,
89				u64 *ppos, int userbuf)
90{
91	unsigned long pfn, offset;
92	size_t nr_bytes;
93	ssize_t read = 0, tmp;
94
95	if (!count)
96		return 0;
97
98	offset = (unsigned long)(*ppos % PAGE_SIZE);
99	pfn = (unsigned long)(*ppos / PAGE_SIZE);
100
101	do {
102		if (count > (PAGE_SIZE - offset))
103			nr_bytes = PAGE_SIZE - offset;
104		else
105			nr_bytes = count;
106
107		/* If pfn is not ram, return zeros for sparse dump files */
108		if (pfn_is_ram(pfn) == 0)
109			memset(buf, 0, nr_bytes);
110		else {
111			tmp = copy_oldmem_page(pfn, buf, nr_bytes,
112						offset, userbuf);
113			if (tmp < 0)
114				return tmp;
115		}
116		*ppos += nr_bytes;
117		count -= nr_bytes;
118		buf += nr_bytes;
119		read += nr_bytes;
120		++pfn;
121		offset = 0;
122	} while (count);
123
124	return read;
125}
126
127/*
128 * Architectures may override this function to allocate ELF header in 2nd kernel
129 */
130int __weak elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
131{
132	return 0;
133}
134
135/*
136 * Architectures may override this function to free header
137 */
138void __weak elfcorehdr_free(unsigned long long addr)
139{}
140
141/*
142 * Architectures may override this function to read from ELF header
143 */
144ssize_t __weak elfcorehdr_read(char *buf, size_t count, u64 *ppos)
145{
146	return read_from_oldmem(buf, count, ppos, 0);
147}
148
149/*
150 * Architectures may override this function to read from notes sections
151 */
152ssize_t __weak elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
153{
154	return read_from_oldmem(buf, count, ppos, 0);
155}
156
157/*
158 * Architectures may override this function to map oldmem
159 */
160int __weak remap_oldmem_pfn_range(struct vm_area_struct *vma,
161				  unsigned long from, unsigned long pfn,
162				  unsigned long size, pgprot_t prot)
163{
164	return remap_pfn_range(vma, from, pfn, size, prot);
165}
166
167/*
168 * Copy to either kernel or user space
169 */
170static int copy_to(void *target, void *src, size_t size, int userbuf)
171{
172	if (userbuf) {
173		if (copy_to_user((char __user *) target, src, size))
174			return -EFAULT;
175	} else {
176		memcpy(target, src, size);
177	}
178	return 0;
179}
180
181/* Read from the ELF header and then the crash dump. On error, negative value is
182 * returned otherwise number of bytes read are returned.
183 */
184static ssize_t __read_vmcore(char *buffer, size_t buflen, loff_t *fpos,
185			     int userbuf)
186{
187	ssize_t acc = 0, tmp;
188	size_t tsz;
189	u64 start;
190	struct vmcore *m = NULL;
191
192	if (buflen == 0 || *fpos >= vmcore_size)
193		return 0;
194
195	/* trim buflen to not go beyond EOF */
196	if (buflen > vmcore_size - *fpos)
197		buflen = vmcore_size - *fpos;
198
199	/* Read ELF core header */
200	if (*fpos < elfcorebuf_sz) {
201		tsz = min(elfcorebuf_sz - (size_t)*fpos, buflen);
202		if (copy_to(buffer, elfcorebuf + *fpos, tsz, userbuf))
203			return -EFAULT;
204		buflen -= tsz;
205		*fpos += tsz;
206		buffer += tsz;
207		acc += tsz;
208
209		/* leave now if filled buffer already */
210		if (buflen == 0)
211			return acc;
212	}
213
214	/* Read Elf note segment */
215	if (*fpos < elfcorebuf_sz + elfnotes_sz) {
216		void *kaddr;
217
218		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)*fpos, buflen);
219		kaddr = elfnotes_buf + *fpos - elfcorebuf_sz;
220		if (copy_to(buffer, kaddr, tsz, userbuf))
221			return -EFAULT;
222		buflen -= tsz;
223		*fpos += tsz;
224		buffer += tsz;
225		acc += tsz;
226
227		/* leave now if filled buffer already */
228		if (buflen == 0)
229			return acc;
230	}
231
232	list_for_each_entry(m, &vmcore_list, list) {
233		if (*fpos < m->offset + m->size) {
234			tsz = min_t(size_t, m->offset + m->size - *fpos, buflen);
235			start = m->paddr + *fpos - m->offset;
236			tmp = read_from_oldmem(buffer, tsz, &start, userbuf);
237			if (tmp < 0)
238				return tmp;
239			buflen -= tsz;
240			*fpos += tsz;
241			buffer += tsz;
242			acc += tsz;
243
244			/* leave now if filled buffer already */
245			if (buflen == 0)
246				return acc;
247		}
248	}
249
250	return acc;
251}
252
253static ssize_t read_vmcore(struct file *file, char __user *buffer,
254			   size_t buflen, loff_t *fpos)
255{
256	return __read_vmcore((__force char *) buffer, buflen, fpos, 1);
257}
258
259/*
260 * The vmcore fault handler uses the page cache and fills data using the
261 * standard __vmcore_read() function.
262 *
263 * On s390 the fault handler is used for memory regions that can't be mapped
264 * directly with remap_pfn_range().
265 */
266static int mmap_vmcore_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
267{
268#ifdef CONFIG_S390
269	struct address_space *mapping = vma->vm_file->f_mapping;
270	pgoff_t index = vmf->pgoff;
271	struct page *page;
272	loff_t offset;
273	char *buf;
274	int rc;
275
276	page = find_or_create_page(mapping, index, GFP_KERNEL);
277	if (!page)
278		return VM_FAULT_OOM;
279	if (!PageUptodate(page)) {
280		offset = (loff_t) index << PAGE_CACHE_SHIFT;
281		buf = __va((page_to_pfn(page) << PAGE_SHIFT));
282		rc = __read_vmcore(buf, PAGE_SIZE, &offset, 0);
283		if (rc < 0) {
284			unlock_page(page);
285			page_cache_release(page);
286			return (rc == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
287		}
288		SetPageUptodate(page);
289	}
290	unlock_page(page);
291	vmf->page = page;
292	return 0;
293#else
294	return VM_FAULT_SIGBUS;
295#endif
296}
297
298static const struct vm_operations_struct vmcore_mmap_ops = {
299	.fault = mmap_vmcore_fault,
300};
301
302/**
303 * alloc_elfnotes_buf - allocate buffer for ELF note segment in
304 *                      vmalloc memory
305 *
306 * @notes_sz: size of buffer
307 *
308 * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
309 * the buffer to user-space by means of remap_vmalloc_range().
310 *
311 * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
312 * disabled and there's no need to allow users to mmap the buffer.
313 */
314static inline char *alloc_elfnotes_buf(size_t notes_sz)
315{
316#ifdef CONFIG_MMU
317	return vmalloc_user(notes_sz);
318#else
319	return vzalloc(notes_sz);
320#endif
321}
322
323/*
324 * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
325 * essential for mmap_vmcore() in order to map physically
326 * non-contiguous objects (ELF header, ELF note segment and memory
327 * regions in the 1st kernel pointed to by PT_LOAD entries) into
328 * virtually contiguous user-space in ELF layout.
329 */
330#ifdef CONFIG_MMU
331/*
332 * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
333 * reported as not being ram with the zero page.
334 *
335 * @vma: vm_area_struct describing requested mapping
336 * @from: start remapping from
337 * @pfn: page frame number to start remapping to
338 * @size: remapping size
339 * @prot: protection bits
340 *
341 * Returns zero on success, -EAGAIN on failure.
342 */
343static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
344				    unsigned long from, unsigned long pfn,
345				    unsigned long size, pgprot_t prot)
346{
347	unsigned long map_size;
348	unsigned long pos_start, pos_end, pos;
349	unsigned long zeropage_pfn = my_zero_pfn(0);
350	size_t len = 0;
351
352	pos_start = pfn;
353	pos_end = pfn + (size >> PAGE_SHIFT);
354
355	for (pos = pos_start; pos < pos_end; ++pos) {
356		if (!pfn_is_ram(pos)) {
357			/*
358			 * We hit a page which is not ram. Remap the continuous
359			 * region between pos_start and pos-1 and replace
360			 * the non-ram page at pos with the zero page.
361			 */
362			if (pos > pos_start) {
363				/* Remap continuous region */
364				map_size = (pos - pos_start) << PAGE_SHIFT;
365				if (remap_oldmem_pfn_range(vma, from + len,
366							   pos_start, map_size,
367							   prot))
368					goto fail;
369				len += map_size;
370			}
371			/* Remap the zero page */
372			if (remap_oldmem_pfn_range(vma, from + len,
373						   zeropage_pfn,
374						   PAGE_SIZE, prot))
375				goto fail;
376			len += PAGE_SIZE;
377			pos_start = pos + 1;
378		}
379	}
380	if (pos > pos_start) {
381		/* Remap the rest */
382		map_size = (pos - pos_start) << PAGE_SHIFT;
383		if (remap_oldmem_pfn_range(vma, from + len, pos_start,
384					   map_size, prot))
385			goto fail;
386	}
387	return 0;
388fail:
389	do_munmap(vma->vm_mm, from, len);
390	return -EAGAIN;
391}
392
393static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
394			    unsigned long from, unsigned long pfn,
395			    unsigned long size, pgprot_t prot)
396{
397	/*
398	 * Check if oldmem_pfn_is_ram was registered to avoid
399	 * looping over all pages without a reason.
400	 */
401	if (oldmem_pfn_is_ram)
402		return remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
403	else
404		return remap_oldmem_pfn_range(vma, from, pfn, size, prot);
405}
406
407static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
408{
409	size_t size = vma->vm_end - vma->vm_start;
410	u64 start, end, len, tsz;
411	struct vmcore *m;
412
413	start = (u64)vma->vm_pgoff << PAGE_SHIFT;
414	end = start + size;
415
416	if (size > vmcore_size || end > vmcore_size)
417		return -EINVAL;
418
419	if (vma->vm_flags & (VM_WRITE | VM_EXEC))
420		return -EPERM;
421
422	vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
423	vma->vm_flags |= VM_MIXEDMAP;
424	vma->vm_ops = &vmcore_mmap_ops;
425
426	len = 0;
427
428	if (start < elfcorebuf_sz) {
429		u64 pfn;
430
431		tsz = min(elfcorebuf_sz - (size_t)start, size);
432		pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
433		if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
434				    vma->vm_page_prot))
435			return -EAGAIN;
436		size -= tsz;
437		start += tsz;
438		len += tsz;
439
440		if (size == 0)
441			return 0;
442	}
443
444	if (start < elfcorebuf_sz + elfnotes_sz) {
445		void *kaddr;
446
447		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
448		kaddr = elfnotes_buf + start - elfcorebuf_sz;
449		if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
450						kaddr, tsz))
451			goto fail;
452		size -= tsz;
453		start += tsz;
454		len += tsz;
455
456		if (size == 0)
457			return 0;
458	}
459
460	list_for_each_entry(m, &vmcore_list, list) {
461		if (start < m->offset + m->size) {
462			u64 paddr = 0;
463
464			tsz = min_t(size_t, m->offset + m->size - start, size);
465			paddr = m->paddr + start - m->offset;
466			if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
467						    paddr >> PAGE_SHIFT, tsz,
468						    vma->vm_page_prot))
469				goto fail;
470			size -= tsz;
471			start += tsz;
472			len += tsz;
473
474			if (size == 0)
475				return 0;
476		}
477	}
478
479	return 0;
480fail:
481	do_munmap(vma->vm_mm, vma->vm_start, len);
482	return -EAGAIN;
483}
484#else
485static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
486{
487	return -ENOSYS;
488}
489#endif
490
491static const struct file_operations proc_vmcore_operations = {
492	.read		= read_vmcore,
493	.llseek		= default_llseek,
494	.mmap		= mmap_vmcore,
495};
496
497static struct vmcore* __init get_new_element(void)
498{
499	return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
500}
501
502static u64 __init get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
503				  struct list_head *vc_list)
504{
505	u64 size;
506	struct vmcore *m;
507
508	size = elfsz + elfnotesegsz;
509	list_for_each_entry(m, vc_list, list) {
510		size += m->size;
511	}
512	return size;
513}
514
515/**
516 * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
517 *
518 * @ehdr_ptr: ELF header
519 *
520 * This function updates p_memsz member of each PT_NOTE entry in the
521 * program header table pointed to by @ehdr_ptr to real size of ELF
522 * note segment.
523 */
524static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
525{
526	int i, rc=0;
527	Elf64_Phdr *phdr_ptr;
528	Elf64_Nhdr *nhdr_ptr;
529
530	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
531	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
532		void *notes_section;
533		u64 offset, max_sz, sz, real_sz = 0;
534		if (phdr_ptr->p_type != PT_NOTE)
535			continue;
536		max_sz = phdr_ptr->p_memsz;
537		offset = phdr_ptr->p_offset;
538		notes_section = kmalloc(max_sz, GFP_KERNEL);
539		if (!notes_section)
540			return -ENOMEM;
541		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
542		if (rc < 0) {
543			kfree(notes_section);
544			return rc;
545		}
546		nhdr_ptr = notes_section;
547		while (nhdr_ptr->n_namesz != 0) {
548			sz = sizeof(Elf64_Nhdr) +
549				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
550				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
551			if ((real_sz + sz) > max_sz) {
552				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
553					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
554				break;
555			}
556			real_sz += sz;
557			nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
558		}
559		kfree(notes_section);
560		phdr_ptr->p_memsz = real_sz;
561		if (real_sz == 0) {
562			pr_warn("Warning: Zero PT_NOTE entries found\n");
563		}
564	}
565
566	return 0;
567}
568
569/**
570 * get_note_number_and_size_elf64 - get the number of PT_NOTE program
571 * headers and sum of real size of their ELF note segment headers and
572 * data.
573 *
574 * @ehdr_ptr: ELF header
575 * @nr_ptnote: buffer for the number of PT_NOTE program headers
576 * @sz_ptnote: buffer for size of unique PT_NOTE program header
577 *
578 * This function is used to merge multiple PT_NOTE program headers
579 * into a unique single one. The resulting unique entry will have
580 * @sz_ptnote in its phdr->p_mem.
581 *
582 * It is assumed that program headers with PT_NOTE type pointed to by
583 * @ehdr_ptr has already been updated by update_note_header_size_elf64
584 * and each of PT_NOTE program headers has actual ELF note segment
585 * size in its p_memsz member.
586 */
587static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
588						 int *nr_ptnote, u64 *sz_ptnote)
589{
590	int i;
591	Elf64_Phdr *phdr_ptr;
592
593	*nr_ptnote = *sz_ptnote = 0;
594
595	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
596	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
597		if (phdr_ptr->p_type != PT_NOTE)
598			continue;
599		*nr_ptnote += 1;
600		*sz_ptnote += phdr_ptr->p_memsz;
601	}
602
603	return 0;
604}
605
606/**
607 * copy_notes_elf64 - copy ELF note segments in a given buffer
608 *
609 * @ehdr_ptr: ELF header
610 * @notes_buf: buffer into which ELF note segments are copied
611 *
612 * This function is used to copy ELF note segment in the 1st kernel
613 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
614 * size of the buffer @notes_buf is equal to or larger than sum of the
615 * real ELF note segment headers and data.
616 *
617 * It is assumed that program headers with PT_NOTE type pointed to by
618 * @ehdr_ptr has already been updated by update_note_header_size_elf64
619 * and each of PT_NOTE program headers has actual ELF note segment
620 * size in its p_memsz member.
621 */
622static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
623{
624	int i, rc=0;
625	Elf64_Phdr *phdr_ptr;
626
627	phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
628
629	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
630		u64 offset;
631		if (phdr_ptr->p_type != PT_NOTE)
632			continue;
633		offset = phdr_ptr->p_offset;
634		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
635					   &offset);
636		if (rc < 0)
637			return rc;
638		notes_buf += phdr_ptr->p_memsz;
639	}
640
641	return 0;
642}
643
644/* Merges all the PT_NOTE headers into one. */
645static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
646					   char **notes_buf, size_t *notes_sz)
647{
648	int i, nr_ptnote=0, rc=0;
649	char *tmp;
650	Elf64_Ehdr *ehdr_ptr;
651	Elf64_Phdr phdr;
652	u64 phdr_sz = 0, note_off;
653
654	ehdr_ptr = (Elf64_Ehdr *)elfptr;
655
656	rc = update_note_header_size_elf64(ehdr_ptr);
657	if (rc < 0)
658		return rc;
659
660	rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
661	if (rc < 0)
662		return rc;
663
664	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
665	*notes_buf = alloc_elfnotes_buf(*notes_sz);
666	if (!*notes_buf)
667		return -ENOMEM;
668
669	rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
670	if (rc < 0)
671		return rc;
672
673	/* Prepare merged PT_NOTE program header. */
674	phdr.p_type    = PT_NOTE;
675	phdr.p_flags   = 0;
676	note_off = sizeof(Elf64_Ehdr) +
677			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
678	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
679	phdr.p_vaddr   = phdr.p_paddr = 0;
680	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
681	phdr.p_align   = 0;
682
683	/* Add merged PT_NOTE program header*/
684	tmp = elfptr + sizeof(Elf64_Ehdr);
685	memcpy(tmp, &phdr, sizeof(phdr));
686	tmp += sizeof(phdr);
687
688	/* Remove unwanted PT_NOTE program headers. */
689	i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
690	*elfsz = *elfsz - i;
691	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
692	memset(elfptr + *elfsz, 0, i);
693	*elfsz = roundup(*elfsz, PAGE_SIZE);
694
695	/* Modify e_phnum to reflect merged headers. */
696	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
697
698	return 0;
699}
700
701/**
702 * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
703 *
704 * @ehdr_ptr: ELF header
705 *
706 * This function updates p_memsz member of each PT_NOTE entry in the
707 * program header table pointed to by @ehdr_ptr to real size of ELF
708 * note segment.
709 */
710static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
711{
712	int i, rc=0;
713	Elf32_Phdr *phdr_ptr;
714	Elf32_Nhdr *nhdr_ptr;
715
716	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
717	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
718		void *notes_section;
719		u64 offset, max_sz, sz, real_sz = 0;
720		if (phdr_ptr->p_type != PT_NOTE)
721			continue;
722		max_sz = phdr_ptr->p_memsz;
723		offset = phdr_ptr->p_offset;
724		notes_section = kmalloc(max_sz, GFP_KERNEL);
725		if (!notes_section)
726			return -ENOMEM;
727		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
728		if (rc < 0) {
729			kfree(notes_section);
730			return rc;
731		}
732		nhdr_ptr = notes_section;
733		while (nhdr_ptr->n_namesz != 0) {
734			sz = sizeof(Elf32_Nhdr) +
735				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
736				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
737			if ((real_sz + sz) > max_sz) {
738				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
739					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
740				break;
741			}
742			real_sz += sz;
743			nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
744		}
745		kfree(notes_section);
746		phdr_ptr->p_memsz = real_sz;
747		if (real_sz == 0) {
748			pr_warn("Warning: Zero PT_NOTE entries found\n");
749		}
750	}
751
752	return 0;
753}
754
755/**
756 * get_note_number_and_size_elf32 - get the number of PT_NOTE program
757 * headers and sum of real size of their ELF note segment headers and
758 * data.
759 *
760 * @ehdr_ptr: ELF header
761 * @nr_ptnote: buffer for the number of PT_NOTE program headers
762 * @sz_ptnote: buffer for size of unique PT_NOTE program header
763 *
764 * This function is used to merge multiple PT_NOTE program headers
765 * into a unique single one. The resulting unique entry will have
766 * @sz_ptnote in its phdr->p_mem.
767 *
768 * It is assumed that program headers with PT_NOTE type pointed to by
769 * @ehdr_ptr has already been updated by update_note_header_size_elf32
770 * and each of PT_NOTE program headers has actual ELF note segment
771 * size in its p_memsz member.
772 */
773static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
774						 int *nr_ptnote, u64 *sz_ptnote)
775{
776	int i;
777	Elf32_Phdr *phdr_ptr;
778
779	*nr_ptnote = *sz_ptnote = 0;
780
781	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
782	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
783		if (phdr_ptr->p_type != PT_NOTE)
784			continue;
785		*nr_ptnote += 1;
786		*sz_ptnote += phdr_ptr->p_memsz;
787	}
788
789	return 0;
790}
791
792/**
793 * copy_notes_elf32 - copy ELF note segments in a given buffer
794 *
795 * @ehdr_ptr: ELF header
796 * @notes_buf: buffer into which ELF note segments are copied
797 *
798 * This function is used to copy ELF note segment in the 1st kernel
799 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
800 * size of the buffer @notes_buf is equal to or larger than sum of the
801 * real ELF note segment headers and data.
802 *
803 * It is assumed that program headers with PT_NOTE type pointed to by
804 * @ehdr_ptr has already been updated by update_note_header_size_elf32
805 * and each of PT_NOTE program headers has actual ELF note segment
806 * size in its p_memsz member.
807 */
808static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
809{
810	int i, rc=0;
811	Elf32_Phdr *phdr_ptr;
812
813	phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
814
815	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
816		u64 offset;
817		if (phdr_ptr->p_type != PT_NOTE)
818			continue;
819		offset = phdr_ptr->p_offset;
820		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
821					   &offset);
822		if (rc < 0)
823			return rc;
824		notes_buf += phdr_ptr->p_memsz;
825	}
826
827	return 0;
828}
829
830/* Merges all the PT_NOTE headers into one. */
831static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
832					   char **notes_buf, size_t *notes_sz)
833{
834	int i, nr_ptnote=0, rc=0;
835	char *tmp;
836	Elf32_Ehdr *ehdr_ptr;
837	Elf32_Phdr phdr;
838	u64 phdr_sz = 0, note_off;
839
840	ehdr_ptr = (Elf32_Ehdr *)elfptr;
841
842	rc = update_note_header_size_elf32(ehdr_ptr);
843	if (rc < 0)
844		return rc;
845
846	rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
847	if (rc < 0)
848		return rc;
849
850	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
851	*notes_buf = alloc_elfnotes_buf(*notes_sz);
852	if (!*notes_buf)
853		return -ENOMEM;
854
855	rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
856	if (rc < 0)
857		return rc;
858
859	/* Prepare merged PT_NOTE program header. */
860	phdr.p_type    = PT_NOTE;
861	phdr.p_flags   = 0;
862	note_off = sizeof(Elf32_Ehdr) +
863			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
864	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
865	phdr.p_vaddr   = phdr.p_paddr = 0;
866	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
867	phdr.p_align   = 0;
868
869	/* Add merged PT_NOTE program header*/
870	tmp = elfptr + sizeof(Elf32_Ehdr);
871	memcpy(tmp, &phdr, sizeof(phdr));
872	tmp += sizeof(phdr);
873
874	/* Remove unwanted PT_NOTE program headers. */
875	i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
876	*elfsz = *elfsz - i;
877	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
878	memset(elfptr + *elfsz, 0, i);
879	*elfsz = roundup(*elfsz, PAGE_SIZE);
880
881	/* Modify e_phnum to reflect merged headers. */
882	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
883
884	return 0;
885}
886
887/* Add memory chunks represented by program headers to vmcore list. Also update
888 * the new offset fields of exported program headers. */
889static int __init process_ptload_program_headers_elf64(char *elfptr,
890						size_t elfsz,
891						size_t elfnotes_sz,
892						struct list_head *vc_list)
893{
894	int i;
895	Elf64_Ehdr *ehdr_ptr;
896	Elf64_Phdr *phdr_ptr;
897	loff_t vmcore_off;
898	struct vmcore *new;
899
900	ehdr_ptr = (Elf64_Ehdr *)elfptr;
901	phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
902
903	/* Skip Elf header, program headers and Elf note segment. */
904	vmcore_off = elfsz + elfnotes_sz;
905
906	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
907		u64 paddr, start, end, size;
908
909		if (phdr_ptr->p_type != PT_LOAD)
910			continue;
911
912		paddr = phdr_ptr->p_offset;
913		start = rounddown(paddr, PAGE_SIZE);
914		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
915		size = end - start;
916
917		/* Add this contiguous chunk of memory to vmcore list.*/
918		new = get_new_element();
919		if (!new)
920			return -ENOMEM;
921		new->paddr = start;
922		new->size = size;
923		list_add_tail(&new->list, vc_list);
924
925		/* Update the program header offset. */
926		phdr_ptr->p_offset = vmcore_off + (paddr - start);
927		vmcore_off = vmcore_off + size;
928	}
929	return 0;
930}
931
932static int __init process_ptload_program_headers_elf32(char *elfptr,
933						size_t elfsz,
934						size_t elfnotes_sz,
935						struct list_head *vc_list)
936{
937	int i;
938	Elf32_Ehdr *ehdr_ptr;
939	Elf32_Phdr *phdr_ptr;
940	loff_t vmcore_off;
941	struct vmcore *new;
942
943	ehdr_ptr = (Elf32_Ehdr *)elfptr;
944	phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
945
946	/* Skip Elf header, program headers and Elf note segment. */
947	vmcore_off = elfsz + elfnotes_sz;
948
949	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
950		u64 paddr, start, end, size;
951
952		if (phdr_ptr->p_type != PT_LOAD)
953			continue;
954
955		paddr = phdr_ptr->p_offset;
956		start = rounddown(paddr, PAGE_SIZE);
957		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
958		size = end - start;
959
960		/* Add this contiguous chunk of memory to vmcore list.*/
961		new = get_new_element();
962		if (!new)
963			return -ENOMEM;
964		new->paddr = start;
965		new->size = size;
966		list_add_tail(&new->list, vc_list);
967
968		/* Update the program header offset */
969		phdr_ptr->p_offset = vmcore_off + (paddr - start);
970		vmcore_off = vmcore_off + size;
971	}
972	return 0;
973}
974
975/* Sets offset fields of vmcore elements. */
976static void __init set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
977					   struct list_head *vc_list)
978{
979	loff_t vmcore_off;
980	struct vmcore *m;
981
982	/* Skip Elf header, program headers and Elf note segment. */
983	vmcore_off = elfsz + elfnotes_sz;
984
985	list_for_each_entry(m, vc_list, list) {
986		m->offset = vmcore_off;
987		vmcore_off += m->size;
988	}
989}
990
991static void free_elfcorebuf(void)
992{
993	free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
994	elfcorebuf = NULL;
995	vfree(elfnotes_buf);
996	elfnotes_buf = NULL;
997}
998
999static int __init parse_crash_elf64_headers(void)
1000{
1001	int rc=0;
1002	Elf64_Ehdr ehdr;
1003	u64 addr;
1004
1005	addr = elfcorehdr_addr;
1006
1007	/* Read Elf header */
1008	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
1009	if (rc < 0)
1010		return rc;
1011
1012	/* Do some basic Verification. */
1013	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1014		(ehdr.e_type != ET_CORE) ||
1015		!vmcore_elf64_check_arch(&ehdr) ||
1016		ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
1017		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1018		ehdr.e_version != EV_CURRENT ||
1019		ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
1020		ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
1021		ehdr.e_phnum == 0) {
1022		pr_warn("Warning: Core image elf header is not sane\n");
1023		return -EINVAL;
1024	}
1025
1026	/* Read in all elf headers. */
1027	elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
1028				ehdr.e_phnum * sizeof(Elf64_Phdr);
1029	elfcorebuf_sz = elfcorebuf_sz_orig;
1030	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1031					      get_order(elfcorebuf_sz_orig));
1032	if (!elfcorebuf)
1033		return -ENOMEM;
1034	addr = elfcorehdr_addr;
1035	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1036	if (rc < 0)
1037		goto fail;
1038
1039	/* Merge all PT_NOTE headers into one. */
1040	rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
1041				      &elfnotes_buf, &elfnotes_sz);
1042	if (rc)
1043		goto fail;
1044	rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
1045						  elfnotes_sz, &vmcore_list);
1046	if (rc)
1047		goto fail;
1048	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1049	return 0;
1050fail:
1051	free_elfcorebuf();
1052	return rc;
1053}
1054
1055static int __init parse_crash_elf32_headers(void)
1056{
1057	int rc=0;
1058	Elf32_Ehdr ehdr;
1059	u64 addr;
1060
1061	addr = elfcorehdr_addr;
1062
1063	/* Read Elf header */
1064	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
1065	if (rc < 0)
1066		return rc;
1067
1068	/* Do some basic Verification. */
1069	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1070		(ehdr.e_type != ET_CORE) ||
1071		!elf_check_arch(&ehdr) ||
1072		ehdr.e_ident[EI_CLASS] != ELFCLASS32||
1073		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1074		ehdr.e_version != EV_CURRENT ||
1075		ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1076		ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1077		ehdr.e_phnum == 0) {
1078		pr_warn("Warning: Core image elf header is not sane\n");
1079		return -EINVAL;
1080	}
1081
1082	/* Read in all elf headers. */
1083	elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1084	elfcorebuf_sz = elfcorebuf_sz_orig;
1085	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1086					      get_order(elfcorebuf_sz_orig));
1087	if (!elfcorebuf)
1088		return -ENOMEM;
1089	addr = elfcorehdr_addr;
1090	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1091	if (rc < 0)
1092		goto fail;
1093
1094	/* Merge all PT_NOTE headers into one. */
1095	rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1096				      &elfnotes_buf, &elfnotes_sz);
1097	if (rc)
1098		goto fail;
1099	rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1100						  elfnotes_sz, &vmcore_list);
1101	if (rc)
1102		goto fail;
1103	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1104	return 0;
1105fail:
1106	free_elfcorebuf();
1107	return rc;
1108}
1109
1110static int __init parse_crash_elf_headers(void)
1111{
1112	unsigned char e_ident[EI_NIDENT];
1113	u64 addr;
1114	int rc=0;
1115
1116	addr = elfcorehdr_addr;
1117	rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1118	if (rc < 0)
1119		return rc;
1120	if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1121		pr_warn("Warning: Core image elf header not found\n");
1122		return -EINVAL;
1123	}
1124
1125	if (e_ident[EI_CLASS] == ELFCLASS64) {
1126		rc = parse_crash_elf64_headers();
1127		if (rc)
1128			return rc;
1129	} else if (e_ident[EI_CLASS] == ELFCLASS32) {
1130		rc = parse_crash_elf32_headers();
1131		if (rc)
1132			return rc;
1133	} else {
1134		pr_warn("Warning: Core image elf header is not sane\n");
1135		return -EINVAL;
1136	}
1137
1138	/* Determine vmcore size. */
1139	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1140				      &vmcore_list);
1141
1142	return 0;
1143}
1144
1145/* Init function for vmcore module. */
1146static int __init vmcore_init(void)
1147{
1148	int rc = 0;
1149
1150	/* Allow architectures to allocate ELF header in 2nd kernel */
1151	rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1152	if (rc)
1153		return rc;
1154	/*
1155	 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1156	 * then capture the dump.
1157	 */
1158	if (!(is_vmcore_usable()))
1159		return rc;
1160	rc = parse_crash_elf_headers();
1161	if (rc) {
1162		pr_warn("Kdump: vmcore not initialized\n");
1163		return rc;
1164	}
1165	elfcorehdr_free(elfcorehdr_addr);
1166	elfcorehdr_addr = ELFCORE_ADDR_ERR;
1167
1168	proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &proc_vmcore_operations);
1169	if (proc_vmcore)
1170		proc_vmcore->size = vmcore_size;
1171	return 0;
1172}
1173fs_initcall(vmcore_init);
1174
1175/* Cleanup function for vmcore module. */
1176void vmcore_cleanup(void)
1177{
1178	struct list_head *pos, *next;
1179
1180	if (proc_vmcore) {
1181		proc_remove(proc_vmcore);
1182		proc_vmcore = NULL;
1183	}
1184
1185	/* clear the vmcore list. */
1186	list_for_each_safe(pos, next, &vmcore_list) {
1187		struct vmcore *m;
1188
1189		m = list_entry(pos, struct vmcore, list);
1190		list_del(&m->list);
1191		kfree(m);
1192	}
1193	free_elfcorebuf();
1194}
1195