Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * fs/proc/vmcore.c Interface for accessing the crash
4 * dump from the system's previous life.
5 * Heavily borrowed from fs/proc/kcore.c
6 * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
7 * Copyright (C) IBM Corporation, 2004. All rights reserved
8 *
9 */
10
11#include <linux/mm.h>
12#include <linux/kcore.h>
13#include <linux/user.h>
14#include <linux/elf.h>
15#include <linux/elfcore.h>
16#include <linux/export.h>
17#include <linux/slab.h>
18#include <linux/highmem.h>
19#include <linux/printk.h>
20#include <linux/memblock.h>
21#include <linux/init.h>
22#include <linux/crash_dump.h>
23#include <linux/list.h>
24#include <linux/moduleparam.h>
25#include <linux/mutex.h>
26#include <linux/vmalloc.h>
27#include <linux/pagemap.h>
28#include <linux/uio.h>
29#include <linux/cc_platform.h>
30#include <asm/io.h>
31#include "internal.h"
32
33/* List representing chunks of contiguous memory areas and their offsets in
34 * vmcore file.
35 */
36static LIST_HEAD(vmcore_list);
37
38/* Stores the pointer to the buffer containing kernel elf core headers. */
39static char *elfcorebuf;
40static size_t elfcorebuf_sz;
41static size_t elfcorebuf_sz_orig;
42
43static char *elfnotes_buf;
44static size_t elfnotes_sz;
45/* Size of all notes minus the device dump notes */
46static size_t elfnotes_orig_sz;
47
48/* Total size of vmcore file. */
49static u64 vmcore_size;
50
51static struct proc_dir_entry *proc_vmcore;
52
53#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
54/* Device Dump list and mutex to synchronize access to list */
55static LIST_HEAD(vmcoredd_list);
56static DEFINE_MUTEX(vmcoredd_mutex);
57
58static bool vmcoredd_disabled;
59core_param(novmcoredd, vmcoredd_disabled, bool, 0);
60#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
61
62/* Device Dump Size */
63static size_t vmcoredd_orig_sz;
64
65static DEFINE_SPINLOCK(vmcore_cb_lock);
66DEFINE_STATIC_SRCU(vmcore_cb_srcu);
67/* List of registered vmcore callbacks. */
68static LIST_HEAD(vmcore_cb_list);
69/* Whether the vmcore has been opened once. */
70static bool vmcore_opened;
71
72void register_vmcore_cb(struct vmcore_cb *cb)
73{
74 INIT_LIST_HEAD(&cb->next);
75 spin_lock(&vmcore_cb_lock);
76 list_add_tail(&cb->next, &vmcore_cb_list);
77 /*
78 * Registering a vmcore callback after the vmcore was opened is
79 * very unusual (e.g., manual driver loading).
80 */
81 if (vmcore_opened)
82 pr_warn_once("Unexpected vmcore callback registration\n");
83 spin_unlock(&vmcore_cb_lock);
84}
85EXPORT_SYMBOL_GPL(register_vmcore_cb);
86
87void unregister_vmcore_cb(struct vmcore_cb *cb)
88{
89 spin_lock(&vmcore_cb_lock);
90 list_del_rcu(&cb->next);
91 /*
92 * Unregistering a vmcore callback after the vmcore was opened is
93 * very unusual (e.g., forced driver removal), but we cannot stop
94 * unregistering.
95 */
96 if (vmcore_opened)
97 pr_warn_once("Unexpected vmcore callback unregistration\n");
98 spin_unlock(&vmcore_cb_lock);
99
100 synchronize_srcu(&vmcore_cb_srcu);
101}
102EXPORT_SYMBOL_GPL(unregister_vmcore_cb);
103
104static bool pfn_is_ram(unsigned long pfn)
105{
106 struct vmcore_cb *cb;
107 bool ret = true;
108
109 list_for_each_entry_srcu(cb, &vmcore_cb_list, next,
110 srcu_read_lock_held(&vmcore_cb_srcu)) {
111 if (unlikely(!cb->pfn_is_ram))
112 continue;
113 ret = cb->pfn_is_ram(cb, pfn);
114 if (!ret)
115 break;
116 }
117
118 return ret;
119}
120
121static int open_vmcore(struct inode *inode, struct file *file)
122{
123 spin_lock(&vmcore_cb_lock);
124 vmcore_opened = true;
125 spin_unlock(&vmcore_cb_lock);
126
127 return 0;
128}
129
130/* Reads a page from the oldmem device from given offset. */
131ssize_t read_from_oldmem(struct iov_iter *iter, size_t count,
132 u64 *ppos, bool encrypted)
133{
134 unsigned long pfn, offset;
135 ssize_t nr_bytes;
136 ssize_t read = 0, tmp;
137 int idx;
138
139 if (!count)
140 return 0;
141
142 offset = (unsigned long)(*ppos % PAGE_SIZE);
143 pfn = (unsigned long)(*ppos / PAGE_SIZE);
144
145 idx = srcu_read_lock(&vmcore_cb_srcu);
146 do {
147 if (count > (PAGE_SIZE - offset))
148 nr_bytes = PAGE_SIZE - offset;
149 else
150 nr_bytes = count;
151
152 /* If pfn is not ram, return zeros for sparse dump files */
153 if (!pfn_is_ram(pfn)) {
154 tmp = iov_iter_zero(nr_bytes, iter);
155 } else {
156 if (encrypted)
157 tmp = copy_oldmem_page_encrypted(iter, pfn,
158 nr_bytes,
159 offset);
160 else
161 tmp = copy_oldmem_page(iter, pfn, nr_bytes,
162 offset);
163 }
164 if (tmp < nr_bytes) {
165 srcu_read_unlock(&vmcore_cb_srcu, idx);
166 return -EFAULT;
167 }
168
169 *ppos += nr_bytes;
170 count -= nr_bytes;
171 read += nr_bytes;
172 ++pfn;
173 offset = 0;
174 } while (count);
175 srcu_read_unlock(&vmcore_cb_srcu, idx);
176
177 return read;
178}
179
180/*
181 * Architectures may override this function to allocate ELF header in 2nd kernel
182 */
183int __weak elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
184{
185 return 0;
186}
187
188/*
189 * Architectures may override this function to free header
190 */
191void __weak elfcorehdr_free(unsigned long long addr)
192{}
193
194/*
195 * Architectures may override this function to read from ELF header
196 */
197ssize_t __weak elfcorehdr_read(char *buf, size_t count, u64 *ppos)
198{
199 struct kvec kvec = { .iov_base = buf, .iov_len = count };
200 struct iov_iter iter;
201
202 iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, count);
203
204 return read_from_oldmem(&iter, count, ppos, false);
205}
206
207/*
208 * Architectures may override this function to read from notes sections
209 */
210ssize_t __weak elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
211{
212 struct kvec kvec = { .iov_base = buf, .iov_len = count };
213 struct iov_iter iter;
214
215 iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, count);
216
217 return read_from_oldmem(&iter, count, ppos,
218 cc_platform_has(CC_ATTR_MEM_ENCRYPT));
219}
220
221/*
222 * Architectures may override this function to map oldmem
223 */
224int __weak remap_oldmem_pfn_range(struct vm_area_struct *vma,
225 unsigned long from, unsigned long pfn,
226 unsigned long size, pgprot_t prot)
227{
228 prot = pgprot_encrypted(prot);
229 return remap_pfn_range(vma, from, pfn, size, prot);
230}
231
232/*
233 * Architectures which support memory encryption override this.
234 */
235ssize_t __weak copy_oldmem_page_encrypted(struct iov_iter *iter,
236 unsigned long pfn, size_t csize, unsigned long offset)
237{
238 return copy_oldmem_page(iter, pfn, csize, offset);
239}
240
241#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
242static int vmcoredd_copy_dumps(struct iov_iter *iter, u64 start, size_t size)
243{
244 struct vmcoredd_node *dump;
245 u64 offset = 0;
246 int ret = 0;
247 size_t tsz;
248 char *buf;
249
250 mutex_lock(&vmcoredd_mutex);
251 list_for_each_entry(dump, &vmcoredd_list, list) {
252 if (start < offset + dump->size) {
253 tsz = min(offset + (u64)dump->size - start, (u64)size);
254 buf = dump->buf + start - offset;
255 if (copy_to_iter(buf, tsz, iter) < tsz) {
256 ret = -EFAULT;
257 goto out_unlock;
258 }
259
260 size -= tsz;
261 start += tsz;
262
263 /* Leave now if buffer filled already */
264 if (!size)
265 goto out_unlock;
266 }
267 offset += dump->size;
268 }
269
270out_unlock:
271 mutex_unlock(&vmcoredd_mutex);
272 return ret;
273}
274
275#ifdef CONFIG_MMU
276static int vmcoredd_mmap_dumps(struct vm_area_struct *vma, unsigned long dst,
277 u64 start, size_t size)
278{
279 struct vmcoredd_node *dump;
280 u64 offset = 0;
281 int ret = 0;
282 size_t tsz;
283 char *buf;
284
285 mutex_lock(&vmcoredd_mutex);
286 list_for_each_entry(dump, &vmcoredd_list, list) {
287 if (start < offset + dump->size) {
288 tsz = min(offset + (u64)dump->size - start, (u64)size);
289 buf = dump->buf + start - offset;
290 if (remap_vmalloc_range_partial(vma, dst, buf, 0,
291 tsz)) {
292 ret = -EFAULT;
293 goto out_unlock;
294 }
295
296 size -= tsz;
297 start += tsz;
298 dst += tsz;
299
300 /* Leave now if buffer filled already */
301 if (!size)
302 goto out_unlock;
303 }
304 offset += dump->size;
305 }
306
307out_unlock:
308 mutex_unlock(&vmcoredd_mutex);
309 return ret;
310}
311#endif /* CONFIG_MMU */
312#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
313
314/* Read from the ELF header and then the crash dump. On error, negative value is
315 * returned otherwise number of bytes read are returned.
316 */
317static ssize_t __read_vmcore(struct iov_iter *iter, loff_t *fpos)
318{
319 ssize_t acc = 0, tmp;
320 size_t tsz;
321 u64 start;
322 struct vmcore *m = NULL;
323
324 if (!iov_iter_count(iter) || *fpos >= vmcore_size)
325 return 0;
326
327 iov_iter_truncate(iter, vmcore_size - *fpos);
328
329 /* Read ELF core header */
330 if (*fpos < elfcorebuf_sz) {
331 tsz = min(elfcorebuf_sz - (size_t)*fpos, iov_iter_count(iter));
332 if (copy_to_iter(elfcorebuf + *fpos, tsz, iter) < tsz)
333 return -EFAULT;
334 *fpos += tsz;
335 acc += tsz;
336
337 /* leave now if filled buffer already */
338 if (!iov_iter_count(iter))
339 return acc;
340 }
341
342 /* Read ELF note segment */
343 if (*fpos < elfcorebuf_sz + elfnotes_sz) {
344 void *kaddr;
345
346 /* We add device dumps before other elf notes because the
347 * other elf notes may not fill the elf notes buffer
348 * completely and we will end up with zero-filled data
349 * between the elf notes and the device dumps. Tools will
350 * then try to decode this zero-filled data as valid notes
351 * and we don't want that. Hence, adding device dumps before
352 * the other elf notes ensure that zero-filled data can be
353 * avoided.
354 */
355#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
356 /* Read device dumps */
357 if (*fpos < elfcorebuf_sz + vmcoredd_orig_sz) {
358 tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
359 (size_t)*fpos, iov_iter_count(iter));
360 start = *fpos - elfcorebuf_sz;
361 if (vmcoredd_copy_dumps(iter, start, tsz))
362 return -EFAULT;
363
364 *fpos += tsz;
365 acc += tsz;
366
367 /* leave now if filled buffer already */
368 if (!iov_iter_count(iter))
369 return acc;
370 }
371#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
372
373 /* Read remaining elf notes */
374 tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)*fpos,
375 iov_iter_count(iter));
376 kaddr = elfnotes_buf + *fpos - elfcorebuf_sz - vmcoredd_orig_sz;
377 if (copy_to_iter(kaddr, tsz, iter) < tsz)
378 return -EFAULT;
379
380 *fpos += tsz;
381 acc += tsz;
382
383 /* leave now if filled buffer already */
384 if (!iov_iter_count(iter))
385 return acc;
386
387 cond_resched();
388 }
389
390 list_for_each_entry(m, &vmcore_list, list) {
391 if (*fpos < m->offset + m->size) {
392 tsz = (size_t)min_t(unsigned long long,
393 m->offset + m->size - *fpos,
394 iov_iter_count(iter));
395 start = m->paddr + *fpos - m->offset;
396 tmp = read_from_oldmem(iter, tsz, &start,
397 cc_platform_has(CC_ATTR_MEM_ENCRYPT));
398 if (tmp < 0)
399 return tmp;
400 *fpos += tsz;
401 acc += tsz;
402
403 /* leave now if filled buffer already */
404 if (!iov_iter_count(iter))
405 return acc;
406 }
407
408 cond_resched();
409 }
410
411 return acc;
412}
413
414static ssize_t read_vmcore(struct kiocb *iocb, struct iov_iter *iter)
415{
416 return __read_vmcore(iter, &iocb->ki_pos);
417}
418
419/**
420 * vmcore_alloc_buf - allocate buffer in vmalloc memory
421 * @size: size of buffer
422 *
423 * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
424 * the buffer to user-space by means of remap_vmalloc_range().
425 *
426 * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
427 * disabled and there's no need to allow users to mmap the buffer.
428 */
429static inline char *vmcore_alloc_buf(size_t size)
430{
431#ifdef CONFIG_MMU
432 return vmalloc_user(size);
433#else
434 return vzalloc(size);
435#endif
436}
437
438/*
439 * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
440 * essential for mmap_vmcore() in order to map physically
441 * non-contiguous objects (ELF header, ELF note segment and memory
442 * regions in the 1st kernel pointed to by PT_LOAD entries) into
443 * virtually contiguous user-space in ELF layout.
444 */
445#ifdef CONFIG_MMU
446
447/*
448 * The vmcore fault handler uses the page cache and fills data using the
449 * standard __read_vmcore() function.
450 *
451 * On s390 the fault handler is used for memory regions that can't be mapped
452 * directly with remap_pfn_range().
453 */
454static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
455{
456#ifdef CONFIG_S390
457 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
458 pgoff_t index = vmf->pgoff;
459 struct iov_iter iter;
460 struct kvec kvec;
461 struct page *page;
462 loff_t offset;
463 int rc;
464
465 page = find_or_create_page(mapping, index, GFP_KERNEL);
466 if (!page)
467 return VM_FAULT_OOM;
468 if (!PageUptodate(page)) {
469 offset = (loff_t) index << PAGE_SHIFT;
470 kvec.iov_base = page_address(page);
471 kvec.iov_len = PAGE_SIZE;
472 iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, PAGE_SIZE);
473
474 rc = __read_vmcore(&iter, &offset);
475 if (rc < 0) {
476 unlock_page(page);
477 put_page(page);
478 return vmf_error(rc);
479 }
480 SetPageUptodate(page);
481 }
482 unlock_page(page);
483 vmf->page = page;
484 return 0;
485#else
486 return VM_FAULT_SIGBUS;
487#endif
488}
489
490static const struct vm_operations_struct vmcore_mmap_ops = {
491 .fault = mmap_vmcore_fault,
492};
493
494/*
495 * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
496 * reported as not being ram with the zero page.
497 *
498 * @vma: vm_area_struct describing requested mapping
499 * @from: start remapping from
500 * @pfn: page frame number to start remapping to
501 * @size: remapping size
502 * @prot: protection bits
503 *
504 * Returns zero on success, -EAGAIN on failure.
505 */
506static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
507 unsigned long from, unsigned long pfn,
508 unsigned long size, pgprot_t prot)
509{
510 unsigned long map_size;
511 unsigned long pos_start, pos_end, pos;
512 unsigned long zeropage_pfn = my_zero_pfn(0);
513 size_t len = 0;
514
515 pos_start = pfn;
516 pos_end = pfn + (size >> PAGE_SHIFT);
517
518 for (pos = pos_start; pos < pos_end; ++pos) {
519 if (!pfn_is_ram(pos)) {
520 /*
521 * We hit a page which is not ram. Remap the continuous
522 * region between pos_start and pos-1 and replace
523 * the non-ram page at pos with the zero page.
524 */
525 if (pos > pos_start) {
526 /* Remap continuous region */
527 map_size = (pos - pos_start) << PAGE_SHIFT;
528 if (remap_oldmem_pfn_range(vma, from + len,
529 pos_start, map_size,
530 prot))
531 goto fail;
532 len += map_size;
533 }
534 /* Remap the zero page */
535 if (remap_oldmem_pfn_range(vma, from + len,
536 zeropage_pfn,
537 PAGE_SIZE, prot))
538 goto fail;
539 len += PAGE_SIZE;
540 pos_start = pos + 1;
541 }
542 }
543 if (pos > pos_start) {
544 /* Remap the rest */
545 map_size = (pos - pos_start) << PAGE_SHIFT;
546 if (remap_oldmem_pfn_range(vma, from + len, pos_start,
547 map_size, prot))
548 goto fail;
549 }
550 return 0;
551fail:
552 do_munmap(vma->vm_mm, from, len, NULL);
553 return -EAGAIN;
554}
555
556static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
557 unsigned long from, unsigned long pfn,
558 unsigned long size, pgprot_t prot)
559{
560 int ret, idx;
561
562 /*
563 * Check if a callback was registered to avoid looping over all
564 * pages without a reason.
565 */
566 idx = srcu_read_lock(&vmcore_cb_srcu);
567 if (!list_empty(&vmcore_cb_list))
568 ret = remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
569 else
570 ret = remap_oldmem_pfn_range(vma, from, pfn, size, prot);
571 srcu_read_unlock(&vmcore_cb_srcu, idx);
572 return ret;
573}
574
575static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
576{
577 size_t size = vma->vm_end - vma->vm_start;
578 u64 start, end, len, tsz;
579 struct vmcore *m;
580
581 start = (u64)vma->vm_pgoff << PAGE_SHIFT;
582 end = start + size;
583
584 if (size > vmcore_size || end > vmcore_size)
585 return -EINVAL;
586
587 if (vma->vm_flags & (VM_WRITE | VM_EXEC))
588 return -EPERM;
589
590 vm_flags_mod(vma, VM_MIXEDMAP, VM_MAYWRITE | VM_MAYEXEC);
591 vma->vm_ops = &vmcore_mmap_ops;
592
593 len = 0;
594
595 if (start < elfcorebuf_sz) {
596 u64 pfn;
597
598 tsz = min(elfcorebuf_sz - (size_t)start, size);
599 pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
600 if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
601 vma->vm_page_prot))
602 return -EAGAIN;
603 size -= tsz;
604 start += tsz;
605 len += tsz;
606
607 if (size == 0)
608 return 0;
609 }
610
611 if (start < elfcorebuf_sz + elfnotes_sz) {
612 void *kaddr;
613
614 /* We add device dumps before other elf notes because the
615 * other elf notes may not fill the elf notes buffer
616 * completely and we will end up with zero-filled data
617 * between the elf notes and the device dumps. Tools will
618 * then try to decode this zero-filled data as valid notes
619 * and we don't want that. Hence, adding device dumps before
620 * the other elf notes ensure that zero-filled data can be
621 * avoided. This also ensures that the device dumps and
622 * other elf notes can be properly mmaped at page aligned
623 * address.
624 */
625#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
626 /* Read device dumps */
627 if (start < elfcorebuf_sz + vmcoredd_orig_sz) {
628 u64 start_off;
629
630 tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
631 (size_t)start, size);
632 start_off = start - elfcorebuf_sz;
633 if (vmcoredd_mmap_dumps(vma, vma->vm_start + len,
634 start_off, tsz))
635 goto fail;
636
637 size -= tsz;
638 start += tsz;
639 len += tsz;
640
641 /* leave now if filled buffer already */
642 if (!size)
643 return 0;
644 }
645#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
646
647 /* Read remaining elf notes */
648 tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
649 kaddr = elfnotes_buf + start - elfcorebuf_sz - vmcoredd_orig_sz;
650 if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
651 kaddr, 0, tsz))
652 goto fail;
653
654 size -= tsz;
655 start += tsz;
656 len += tsz;
657
658 if (size == 0)
659 return 0;
660 }
661
662 list_for_each_entry(m, &vmcore_list, list) {
663 if (start < m->offset + m->size) {
664 u64 paddr = 0;
665
666 tsz = (size_t)min_t(unsigned long long,
667 m->offset + m->size - start, size);
668 paddr = m->paddr + start - m->offset;
669 if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
670 paddr >> PAGE_SHIFT, tsz,
671 vma->vm_page_prot))
672 goto fail;
673 size -= tsz;
674 start += tsz;
675 len += tsz;
676
677 if (size == 0)
678 return 0;
679 }
680 }
681
682 return 0;
683fail:
684 do_munmap(vma->vm_mm, vma->vm_start, len, NULL);
685 return -EAGAIN;
686}
687#else
688static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
689{
690 return -ENOSYS;
691}
692#endif
693
694static const struct proc_ops vmcore_proc_ops = {
695 .proc_open = open_vmcore,
696 .proc_read_iter = read_vmcore,
697 .proc_lseek = default_llseek,
698 .proc_mmap = mmap_vmcore,
699};
700
701static struct vmcore* __init get_new_element(void)
702{
703 return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
704}
705
706static u64 get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
707 struct list_head *vc_list)
708{
709 u64 size;
710 struct vmcore *m;
711
712 size = elfsz + elfnotesegsz;
713 list_for_each_entry(m, vc_list, list) {
714 size += m->size;
715 }
716 return size;
717}
718
719/**
720 * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
721 *
722 * @ehdr_ptr: ELF header
723 *
724 * This function updates p_memsz member of each PT_NOTE entry in the
725 * program header table pointed to by @ehdr_ptr to real size of ELF
726 * note segment.
727 */
728static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
729{
730 int i, rc=0;
731 Elf64_Phdr *phdr_ptr;
732 Elf64_Nhdr *nhdr_ptr;
733
734 phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
735 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
736 void *notes_section;
737 u64 offset, max_sz, sz, real_sz = 0;
738 if (phdr_ptr->p_type != PT_NOTE)
739 continue;
740 max_sz = phdr_ptr->p_memsz;
741 offset = phdr_ptr->p_offset;
742 notes_section = kmalloc(max_sz, GFP_KERNEL);
743 if (!notes_section)
744 return -ENOMEM;
745 rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
746 if (rc < 0) {
747 kfree(notes_section);
748 return rc;
749 }
750 nhdr_ptr = notes_section;
751 while (nhdr_ptr->n_namesz != 0) {
752 sz = sizeof(Elf64_Nhdr) +
753 (((u64)nhdr_ptr->n_namesz + 3) & ~3) +
754 (((u64)nhdr_ptr->n_descsz + 3) & ~3);
755 if ((real_sz + sz) > max_sz) {
756 pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
757 nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
758 break;
759 }
760 real_sz += sz;
761 nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
762 }
763 kfree(notes_section);
764 phdr_ptr->p_memsz = real_sz;
765 if (real_sz == 0) {
766 pr_warn("Warning: Zero PT_NOTE entries found\n");
767 }
768 }
769
770 return 0;
771}
772
773/**
774 * get_note_number_and_size_elf64 - get the number of PT_NOTE program
775 * headers and sum of real size of their ELF note segment headers and
776 * data.
777 *
778 * @ehdr_ptr: ELF header
779 * @nr_ptnote: buffer for the number of PT_NOTE program headers
780 * @sz_ptnote: buffer for size of unique PT_NOTE program header
781 *
782 * This function is used to merge multiple PT_NOTE program headers
783 * into a unique single one. The resulting unique entry will have
784 * @sz_ptnote in its phdr->p_mem.
785 *
786 * It is assumed that program headers with PT_NOTE type pointed to by
787 * @ehdr_ptr has already been updated by update_note_header_size_elf64
788 * and each of PT_NOTE program headers has actual ELF note segment
789 * size in its p_memsz member.
790 */
791static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
792 int *nr_ptnote, u64 *sz_ptnote)
793{
794 int i;
795 Elf64_Phdr *phdr_ptr;
796
797 *nr_ptnote = *sz_ptnote = 0;
798
799 phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
800 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
801 if (phdr_ptr->p_type != PT_NOTE)
802 continue;
803 *nr_ptnote += 1;
804 *sz_ptnote += phdr_ptr->p_memsz;
805 }
806
807 return 0;
808}
809
810/**
811 * copy_notes_elf64 - copy ELF note segments in a given buffer
812 *
813 * @ehdr_ptr: ELF header
814 * @notes_buf: buffer into which ELF note segments are copied
815 *
816 * This function is used to copy ELF note segment in the 1st kernel
817 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
818 * size of the buffer @notes_buf is equal to or larger than sum of the
819 * real ELF note segment headers and data.
820 *
821 * It is assumed that program headers with PT_NOTE type pointed to by
822 * @ehdr_ptr has already been updated by update_note_header_size_elf64
823 * and each of PT_NOTE program headers has actual ELF note segment
824 * size in its p_memsz member.
825 */
826static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
827{
828 int i, rc=0;
829 Elf64_Phdr *phdr_ptr;
830
831 phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
832
833 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
834 u64 offset;
835 if (phdr_ptr->p_type != PT_NOTE)
836 continue;
837 offset = phdr_ptr->p_offset;
838 rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
839 &offset);
840 if (rc < 0)
841 return rc;
842 notes_buf += phdr_ptr->p_memsz;
843 }
844
845 return 0;
846}
847
848/* Merges all the PT_NOTE headers into one. */
849static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
850 char **notes_buf, size_t *notes_sz)
851{
852 int i, nr_ptnote=0, rc=0;
853 char *tmp;
854 Elf64_Ehdr *ehdr_ptr;
855 Elf64_Phdr phdr;
856 u64 phdr_sz = 0, note_off;
857
858 ehdr_ptr = (Elf64_Ehdr *)elfptr;
859
860 rc = update_note_header_size_elf64(ehdr_ptr);
861 if (rc < 0)
862 return rc;
863
864 rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
865 if (rc < 0)
866 return rc;
867
868 *notes_sz = roundup(phdr_sz, PAGE_SIZE);
869 *notes_buf = vmcore_alloc_buf(*notes_sz);
870 if (!*notes_buf)
871 return -ENOMEM;
872
873 rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
874 if (rc < 0)
875 return rc;
876
877 /* Prepare merged PT_NOTE program header. */
878 phdr.p_type = PT_NOTE;
879 phdr.p_flags = 0;
880 note_off = sizeof(Elf64_Ehdr) +
881 (ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
882 phdr.p_offset = roundup(note_off, PAGE_SIZE);
883 phdr.p_vaddr = phdr.p_paddr = 0;
884 phdr.p_filesz = phdr.p_memsz = phdr_sz;
885 phdr.p_align = 4;
886
887 /* Add merged PT_NOTE program header*/
888 tmp = elfptr + sizeof(Elf64_Ehdr);
889 memcpy(tmp, &phdr, sizeof(phdr));
890 tmp += sizeof(phdr);
891
892 /* Remove unwanted PT_NOTE program headers. */
893 i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
894 *elfsz = *elfsz - i;
895 memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
896 memset(elfptr + *elfsz, 0, i);
897 *elfsz = roundup(*elfsz, PAGE_SIZE);
898
899 /* Modify e_phnum to reflect merged headers. */
900 ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
901
902 /* Store the size of all notes. We need this to update the note
903 * header when the device dumps will be added.
904 */
905 elfnotes_orig_sz = phdr.p_memsz;
906
907 return 0;
908}
909
910/**
911 * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
912 *
913 * @ehdr_ptr: ELF header
914 *
915 * This function updates p_memsz member of each PT_NOTE entry in the
916 * program header table pointed to by @ehdr_ptr to real size of ELF
917 * note segment.
918 */
919static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
920{
921 int i, rc=0;
922 Elf32_Phdr *phdr_ptr;
923 Elf32_Nhdr *nhdr_ptr;
924
925 phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
926 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
927 void *notes_section;
928 u64 offset, max_sz, sz, real_sz = 0;
929 if (phdr_ptr->p_type != PT_NOTE)
930 continue;
931 max_sz = phdr_ptr->p_memsz;
932 offset = phdr_ptr->p_offset;
933 notes_section = kmalloc(max_sz, GFP_KERNEL);
934 if (!notes_section)
935 return -ENOMEM;
936 rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
937 if (rc < 0) {
938 kfree(notes_section);
939 return rc;
940 }
941 nhdr_ptr = notes_section;
942 while (nhdr_ptr->n_namesz != 0) {
943 sz = sizeof(Elf32_Nhdr) +
944 (((u64)nhdr_ptr->n_namesz + 3) & ~3) +
945 (((u64)nhdr_ptr->n_descsz + 3) & ~3);
946 if ((real_sz + sz) > max_sz) {
947 pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
948 nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
949 break;
950 }
951 real_sz += sz;
952 nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
953 }
954 kfree(notes_section);
955 phdr_ptr->p_memsz = real_sz;
956 if (real_sz == 0) {
957 pr_warn("Warning: Zero PT_NOTE entries found\n");
958 }
959 }
960
961 return 0;
962}
963
964/**
965 * get_note_number_and_size_elf32 - get the number of PT_NOTE program
966 * headers and sum of real size of their ELF note segment headers and
967 * data.
968 *
969 * @ehdr_ptr: ELF header
970 * @nr_ptnote: buffer for the number of PT_NOTE program headers
971 * @sz_ptnote: buffer for size of unique PT_NOTE program header
972 *
973 * This function is used to merge multiple PT_NOTE program headers
974 * into a unique single one. The resulting unique entry will have
975 * @sz_ptnote in its phdr->p_mem.
976 *
977 * It is assumed that program headers with PT_NOTE type pointed to by
978 * @ehdr_ptr has already been updated by update_note_header_size_elf32
979 * and each of PT_NOTE program headers has actual ELF note segment
980 * size in its p_memsz member.
981 */
982static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
983 int *nr_ptnote, u64 *sz_ptnote)
984{
985 int i;
986 Elf32_Phdr *phdr_ptr;
987
988 *nr_ptnote = *sz_ptnote = 0;
989
990 phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
991 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
992 if (phdr_ptr->p_type != PT_NOTE)
993 continue;
994 *nr_ptnote += 1;
995 *sz_ptnote += phdr_ptr->p_memsz;
996 }
997
998 return 0;
999}
1000
1001/**
1002 * copy_notes_elf32 - copy ELF note segments in a given buffer
1003 *
1004 * @ehdr_ptr: ELF header
1005 * @notes_buf: buffer into which ELF note segments are copied
1006 *
1007 * This function is used to copy ELF note segment in the 1st kernel
1008 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
1009 * size of the buffer @notes_buf is equal to or larger than sum of the
1010 * real ELF note segment headers and data.
1011 *
1012 * It is assumed that program headers with PT_NOTE type pointed to by
1013 * @ehdr_ptr has already been updated by update_note_header_size_elf32
1014 * and each of PT_NOTE program headers has actual ELF note segment
1015 * size in its p_memsz member.
1016 */
1017static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
1018{
1019 int i, rc=0;
1020 Elf32_Phdr *phdr_ptr;
1021
1022 phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
1023
1024 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1025 u64 offset;
1026 if (phdr_ptr->p_type != PT_NOTE)
1027 continue;
1028 offset = phdr_ptr->p_offset;
1029 rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
1030 &offset);
1031 if (rc < 0)
1032 return rc;
1033 notes_buf += phdr_ptr->p_memsz;
1034 }
1035
1036 return 0;
1037}
1038
1039/* Merges all the PT_NOTE headers into one. */
1040static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
1041 char **notes_buf, size_t *notes_sz)
1042{
1043 int i, nr_ptnote=0, rc=0;
1044 char *tmp;
1045 Elf32_Ehdr *ehdr_ptr;
1046 Elf32_Phdr phdr;
1047 u64 phdr_sz = 0, note_off;
1048
1049 ehdr_ptr = (Elf32_Ehdr *)elfptr;
1050
1051 rc = update_note_header_size_elf32(ehdr_ptr);
1052 if (rc < 0)
1053 return rc;
1054
1055 rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
1056 if (rc < 0)
1057 return rc;
1058
1059 *notes_sz = roundup(phdr_sz, PAGE_SIZE);
1060 *notes_buf = vmcore_alloc_buf(*notes_sz);
1061 if (!*notes_buf)
1062 return -ENOMEM;
1063
1064 rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
1065 if (rc < 0)
1066 return rc;
1067
1068 /* Prepare merged PT_NOTE program header. */
1069 phdr.p_type = PT_NOTE;
1070 phdr.p_flags = 0;
1071 note_off = sizeof(Elf32_Ehdr) +
1072 (ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
1073 phdr.p_offset = roundup(note_off, PAGE_SIZE);
1074 phdr.p_vaddr = phdr.p_paddr = 0;
1075 phdr.p_filesz = phdr.p_memsz = phdr_sz;
1076 phdr.p_align = 4;
1077
1078 /* Add merged PT_NOTE program header*/
1079 tmp = elfptr + sizeof(Elf32_Ehdr);
1080 memcpy(tmp, &phdr, sizeof(phdr));
1081 tmp += sizeof(phdr);
1082
1083 /* Remove unwanted PT_NOTE program headers. */
1084 i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
1085 *elfsz = *elfsz - i;
1086 memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
1087 memset(elfptr + *elfsz, 0, i);
1088 *elfsz = roundup(*elfsz, PAGE_SIZE);
1089
1090 /* Modify e_phnum to reflect merged headers. */
1091 ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
1092
1093 /* Store the size of all notes. We need this to update the note
1094 * header when the device dumps will be added.
1095 */
1096 elfnotes_orig_sz = phdr.p_memsz;
1097
1098 return 0;
1099}
1100
1101/* Add memory chunks represented by program headers to vmcore list. Also update
1102 * the new offset fields of exported program headers. */
1103static int __init process_ptload_program_headers_elf64(char *elfptr,
1104 size_t elfsz,
1105 size_t elfnotes_sz,
1106 struct list_head *vc_list)
1107{
1108 int i;
1109 Elf64_Ehdr *ehdr_ptr;
1110 Elf64_Phdr *phdr_ptr;
1111 loff_t vmcore_off;
1112 struct vmcore *new;
1113
1114 ehdr_ptr = (Elf64_Ehdr *)elfptr;
1115 phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
1116
1117 /* Skip ELF header, program headers and ELF note segment. */
1118 vmcore_off = elfsz + elfnotes_sz;
1119
1120 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1121 u64 paddr, start, end, size;
1122
1123 if (phdr_ptr->p_type != PT_LOAD)
1124 continue;
1125
1126 paddr = phdr_ptr->p_offset;
1127 start = rounddown(paddr, PAGE_SIZE);
1128 end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1129 size = end - start;
1130
1131 /* Add this contiguous chunk of memory to vmcore list.*/
1132 new = get_new_element();
1133 if (!new)
1134 return -ENOMEM;
1135 new->paddr = start;
1136 new->size = size;
1137 list_add_tail(&new->list, vc_list);
1138
1139 /* Update the program header offset. */
1140 phdr_ptr->p_offset = vmcore_off + (paddr - start);
1141 vmcore_off = vmcore_off + size;
1142 }
1143 return 0;
1144}
1145
1146static int __init process_ptload_program_headers_elf32(char *elfptr,
1147 size_t elfsz,
1148 size_t elfnotes_sz,
1149 struct list_head *vc_list)
1150{
1151 int i;
1152 Elf32_Ehdr *ehdr_ptr;
1153 Elf32_Phdr *phdr_ptr;
1154 loff_t vmcore_off;
1155 struct vmcore *new;
1156
1157 ehdr_ptr = (Elf32_Ehdr *)elfptr;
1158 phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
1159
1160 /* Skip ELF header, program headers and ELF note segment. */
1161 vmcore_off = elfsz + elfnotes_sz;
1162
1163 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1164 u64 paddr, start, end, size;
1165
1166 if (phdr_ptr->p_type != PT_LOAD)
1167 continue;
1168
1169 paddr = phdr_ptr->p_offset;
1170 start = rounddown(paddr, PAGE_SIZE);
1171 end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1172 size = end - start;
1173
1174 /* Add this contiguous chunk of memory to vmcore list.*/
1175 new = get_new_element();
1176 if (!new)
1177 return -ENOMEM;
1178 new->paddr = start;
1179 new->size = size;
1180 list_add_tail(&new->list, vc_list);
1181
1182 /* Update the program header offset */
1183 phdr_ptr->p_offset = vmcore_off + (paddr - start);
1184 vmcore_off = vmcore_off + size;
1185 }
1186 return 0;
1187}
1188
1189/* Sets offset fields of vmcore elements. */
1190static void set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
1191 struct list_head *vc_list)
1192{
1193 loff_t vmcore_off;
1194 struct vmcore *m;
1195
1196 /* Skip ELF header, program headers and ELF note segment. */
1197 vmcore_off = elfsz + elfnotes_sz;
1198
1199 list_for_each_entry(m, vc_list, list) {
1200 m->offset = vmcore_off;
1201 vmcore_off += m->size;
1202 }
1203}
1204
1205static void free_elfcorebuf(void)
1206{
1207 free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
1208 elfcorebuf = NULL;
1209 vfree(elfnotes_buf);
1210 elfnotes_buf = NULL;
1211}
1212
1213static int __init parse_crash_elf64_headers(void)
1214{
1215 int rc=0;
1216 Elf64_Ehdr ehdr;
1217 u64 addr;
1218
1219 addr = elfcorehdr_addr;
1220
1221 /* Read ELF header */
1222 rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
1223 if (rc < 0)
1224 return rc;
1225
1226 /* Do some basic Verification. */
1227 if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1228 (ehdr.e_type != ET_CORE) ||
1229 !vmcore_elf64_check_arch(&ehdr) ||
1230 ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
1231 ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1232 ehdr.e_version != EV_CURRENT ||
1233 ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
1234 ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
1235 ehdr.e_phnum == 0) {
1236 pr_warn("Warning: Core image elf header is not sane\n");
1237 return -EINVAL;
1238 }
1239
1240 /* Read in all elf headers. */
1241 elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
1242 ehdr.e_phnum * sizeof(Elf64_Phdr);
1243 elfcorebuf_sz = elfcorebuf_sz_orig;
1244 elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1245 get_order(elfcorebuf_sz_orig));
1246 if (!elfcorebuf)
1247 return -ENOMEM;
1248 addr = elfcorehdr_addr;
1249 rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1250 if (rc < 0)
1251 goto fail;
1252
1253 /* Merge all PT_NOTE headers into one. */
1254 rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
1255 &elfnotes_buf, &elfnotes_sz);
1256 if (rc)
1257 goto fail;
1258 rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
1259 elfnotes_sz, &vmcore_list);
1260 if (rc)
1261 goto fail;
1262 set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1263 return 0;
1264fail:
1265 free_elfcorebuf();
1266 return rc;
1267}
1268
1269static int __init parse_crash_elf32_headers(void)
1270{
1271 int rc=0;
1272 Elf32_Ehdr ehdr;
1273 u64 addr;
1274
1275 addr = elfcorehdr_addr;
1276
1277 /* Read ELF header */
1278 rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
1279 if (rc < 0)
1280 return rc;
1281
1282 /* Do some basic Verification. */
1283 if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1284 (ehdr.e_type != ET_CORE) ||
1285 !vmcore_elf32_check_arch(&ehdr) ||
1286 ehdr.e_ident[EI_CLASS] != ELFCLASS32||
1287 ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1288 ehdr.e_version != EV_CURRENT ||
1289 ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1290 ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1291 ehdr.e_phnum == 0) {
1292 pr_warn("Warning: Core image elf header is not sane\n");
1293 return -EINVAL;
1294 }
1295
1296 /* Read in all elf headers. */
1297 elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1298 elfcorebuf_sz = elfcorebuf_sz_orig;
1299 elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1300 get_order(elfcorebuf_sz_orig));
1301 if (!elfcorebuf)
1302 return -ENOMEM;
1303 addr = elfcorehdr_addr;
1304 rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1305 if (rc < 0)
1306 goto fail;
1307
1308 /* Merge all PT_NOTE headers into one. */
1309 rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1310 &elfnotes_buf, &elfnotes_sz);
1311 if (rc)
1312 goto fail;
1313 rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1314 elfnotes_sz, &vmcore_list);
1315 if (rc)
1316 goto fail;
1317 set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1318 return 0;
1319fail:
1320 free_elfcorebuf();
1321 return rc;
1322}
1323
1324static int __init parse_crash_elf_headers(void)
1325{
1326 unsigned char e_ident[EI_NIDENT];
1327 u64 addr;
1328 int rc=0;
1329
1330 addr = elfcorehdr_addr;
1331 rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1332 if (rc < 0)
1333 return rc;
1334 if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1335 pr_warn("Warning: Core image elf header not found\n");
1336 return -EINVAL;
1337 }
1338
1339 if (e_ident[EI_CLASS] == ELFCLASS64) {
1340 rc = parse_crash_elf64_headers();
1341 if (rc)
1342 return rc;
1343 } else if (e_ident[EI_CLASS] == ELFCLASS32) {
1344 rc = parse_crash_elf32_headers();
1345 if (rc)
1346 return rc;
1347 } else {
1348 pr_warn("Warning: Core image elf header is not sane\n");
1349 return -EINVAL;
1350 }
1351
1352 /* Determine vmcore size. */
1353 vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1354 &vmcore_list);
1355
1356 return 0;
1357}
1358
1359#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1360/**
1361 * vmcoredd_write_header - Write vmcore device dump header at the
1362 * beginning of the dump's buffer.
1363 * @buf: Output buffer where the note is written
1364 * @data: Dump info
1365 * @size: Size of the dump
1366 *
1367 * Fills beginning of the dump's buffer with vmcore device dump header.
1368 */
1369static void vmcoredd_write_header(void *buf, struct vmcoredd_data *data,
1370 u32 size)
1371{
1372 struct vmcoredd_header *vdd_hdr = (struct vmcoredd_header *)buf;
1373
1374 vdd_hdr->n_namesz = sizeof(vdd_hdr->name);
1375 vdd_hdr->n_descsz = size + sizeof(vdd_hdr->dump_name);
1376 vdd_hdr->n_type = NT_VMCOREDD;
1377
1378 strscpy_pad(vdd_hdr->name, VMCOREDD_NOTE_NAME);
1379 strscpy_pad(vdd_hdr->dump_name, data->dump_name);
1380}
1381
1382/**
1383 * vmcoredd_update_program_headers - Update all ELF program headers
1384 * @elfptr: Pointer to elf header
1385 * @elfnotesz: Size of elf notes aligned to page size
1386 * @vmcoreddsz: Size of device dumps to be added to elf note header
1387 *
1388 * Determine type of ELF header (Elf64 or Elf32) and update the elf note size.
1389 * Also update the offsets of all the program headers after the elf note header.
1390 */
1391static void vmcoredd_update_program_headers(char *elfptr, size_t elfnotesz,
1392 size_t vmcoreddsz)
1393{
1394 unsigned char *e_ident = (unsigned char *)elfptr;
1395 u64 start, end, size;
1396 loff_t vmcore_off;
1397 u32 i;
1398
1399 vmcore_off = elfcorebuf_sz + elfnotesz;
1400
1401 if (e_ident[EI_CLASS] == ELFCLASS64) {
1402 Elf64_Ehdr *ehdr = (Elf64_Ehdr *)elfptr;
1403 Elf64_Phdr *phdr = (Elf64_Phdr *)(elfptr + sizeof(Elf64_Ehdr));
1404
1405 /* Update all program headers */
1406 for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1407 if (phdr->p_type == PT_NOTE) {
1408 /* Update note size */
1409 phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1410 phdr->p_filesz = phdr->p_memsz;
1411 continue;
1412 }
1413
1414 start = rounddown(phdr->p_offset, PAGE_SIZE);
1415 end = roundup(phdr->p_offset + phdr->p_memsz,
1416 PAGE_SIZE);
1417 size = end - start;
1418 phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1419 vmcore_off += size;
1420 }
1421 } else {
1422 Elf32_Ehdr *ehdr = (Elf32_Ehdr *)elfptr;
1423 Elf32_Phdr *phdr = (Elf32_Phdr *)(elfptr + sizeof(Elf32_Ehdr));
1424
1425 /* Update all program headers */
1426 for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1427 if (phdr->p_type == PT_NOTE) {
1428 /* Update note size */
1429 phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1430 phdr->p_filesz = phdr->p_memsz;
1431 continue;
1432 }
1433
1434 start = rounddown(phdr->p_offset, PAGE_SIZE);
1435 end = roundup(phdr->p_offset + phdr->p_memsz,
1436 PAGE_SIZE);
1437 size = end - start;
1438 phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1439 vmcore_off += size;
1440 }
1441 }
1442}
1443
1444/**
1445 * vmcoredd_update_size - Update the total size of the device dumps and update
1446 * ELF header
1447 * @dump_size: Size of the current device dump to be added to total size
1448 *
1449 * Update the total size of all the device dumps and update the ELF program
1450 * headers. Calculate the new offsets for the vmcore list and update the
1451 * total vmcore size.
1452 */
1453static void vmcoredd_update_size(size_t dump_size)
1454{
1455 vmcoredd_orig_sz += dump_size;
1456 elfnotes_sz = roundup(elfnotes_orig_sz, PAGE_SIZE) + vmcoredd_orig_sz;
1457 vmcoredd_update_program_headers(elfcorebuf, elfnotes_sz,
1458 vmcoredd_orig_sz);
1459
1460 /* Update vmcore list offsets */
1461 set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1462
1463 vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1464 &vmcore_list);
1465 proc_vmcore->size = vmcore_size;
1466}
1467
1468/**
1469 * vmcore_add_device_dump - Add a buffer containing device dump to vmcore
1470 * @data: dump info.
1471 *
1472 * Allocate a buffer and invoke the calling driver's dump collect routine.
1473 * Write ELF note at the beginning of the buffer to indicate vmcore device
1474 * dump and add the dump to global list.
1475 */
1476int vmcore_add_device_dump(struct vmcoredd_data *data)
1477{
1478 struct vmcoredd_node *dump;
1479 void *buf = NULL;
1480 size_t data_size;
1481 int ret;
1482
1483 if (vmcoredd_disabled) {
1484 pr_err_once("Device dump is disabled\n");
1485 return -EINVAL;
1486 }
1487
1488 if (!data || !strlen(data->dump_name) ||
1489 !data->vmcoredd_callback || !data->size)
1490 return -EINVAL;
1491
1492 dump = vzalloc(sizeof(*dump));
1493 if (!dump) {
1494 ret = -ENOMEM;
1495 goto out_err;
1496 }
1497
1498 /* Keep size of the buffer page aligned so that it can be mmaped */
1499 data_size = roundup(sizeof(struct vmcoredd_header) + data->size,
1500 PAGE_SIZE);
1501
1502 /* Allocate buffer for driver's to write their dumps */
1503 buf = vmcore_alloc_buf(data_size);
1504 if (!buf) {
1505 ret = -ENOMEM;
1506 goto out_err;
1507 }
1508
1509 vmcoredd_write_header(buf, data, data_size -
1510 sizeof(struct vmcoredd_header));
1511
1512 /* Invoke the driver's dump collection routing */
1513 ret = data->vmcoredd_callback(data, buf +
1514 sizeof(struct vmcoredd_header));
1515 if (ret)
1516 goto out_err;
1517
1518 dump->buf = buf;
1519 dump->size = data_size;
1520
1521 /* Add the dump to driver sysfs list */
1522 mutex_lock(&vmcoredd_mutex);
1523 list_add_tail(&dump->list, &vmcoredd_list);
1524 mutex_unlock(&vmcoredd_mutex);
1525
1526 vmcoredd_update_size(data_size);
1527 return 0;
1528
1529out_err:
1530 vfree(buf);
1531 vfree(dump);
1532
1533 return ret;
1534}
1535EXPORT_SYMBOL(vmcore_add_device_dump);
1536#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1537
1538/* Free all dumps in vmcore device dump list */
1539static void vmcore_free_device_dumps(void)
1540{
1541#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1542 mutex_lock(&vmcoredd_mutex);
1543 while (!list_empty(&vmcoredd_list)) {
1544 struct vmcoredd_node *dump;
1545
1546 dump = list_first_entry(&vmcoredd_list, struct vmcoredd_node,
1547 list);
1548 list_del(&dump->list);
1549 vfree(dump->buf);
1550 vfree(dump);
1551 }
1552 mutex_unlock(&vmcoredd_mutex);
1553#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1554}
1555
1556/* Init function for vmcore module. */
1557static int __init vmcore_init(void)
1558{
1559 int rc = 0;
1560
1561 /* Allow architectures to allocate ELF header in 2nd kernel */
1562 rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1563 if (rc)
1564 return rc;
1565 /*
1566 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1567 * then capture the dump.
1568 */
1569 if (!(is_vmcore_usable()))
1570 return rc;
1571 rc = parse_crash_elf_headers();
1572 if (rc) {
1573 elfcorehdr_free(elfcorehdr_addr);
1574 pr_warn("Kdump: vmcore not initialized\n");
1575 return rc;
1576 }
1577 elfcorehdr_free(elfcorehdr_addr);
1578 elfcorehdr_addr = ELFCORE_ADDR_ERR;
1579
1580 proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &vmcore_proc_ops);
1581 if (proc_vmcore)
1582 proc_vmcore->size = vmcore_size;
1583 return 0;
1584}
1585fs_initcall(vmcore_init);
1586
1587/* Cleanup function for vmcore module. */
1588void vmcore_cleanup(void)
1589{
1590 if (proc_vmcore) {
1591 proc_remove(proc_vmcore);
1592 proc_vmcore = NULL;
1593 }
1594
1595 /* clear the vmcore list. */
1596 while (!list_empty(&vmcore_list)) {
1597 struct vmcore *m;
1598
1599 m = list_first_entry(&vmcore_list, struct vmcore, list);
1600 list_del(&m->list);
1601 kfree(m);
1602 }
1603 free_elfcorebuf();
1604
1605 /* clear vmcore device dump list */
1606 vmcore_free_device_dumps();
1607}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * fs/proc/vmcore.c Interface for accessing the crash
4 * dump from the system's previous life.
5 * Heavily borrowed from fs/proc/kcore.c
6 * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
7 * Copyright (C) IBM Corporation, 2004. All rights reserved
8 *
9 */
10
11#include <linux/mm.h>
12#include <linux/kcore.h>
13#include <linux/user.h>
14#include <linux/elf.h>
15#include <linux/elfcore.h>
16#include <linux/export.h>
17#include <linux/slab.h>
18#include <linux/highmem.h>
19#include <linux/printk.h>
20#include <linux/memblock.h>
21#include <linux/init.h>
22#include <linux/crash_dump.h>
23#include <linux/list.h>
24#include <linux/moduleparam.h>
25#include <linux/mutex.h>
26#include <linux/vmalloc.h>
27#include <linux/pagemap.h>
28#include <linux/uaccess.h>
29#include <linux/mem_encrypt.h>
30#include <asm/io.h>
31#include "internal.h"
32
33/* List representing chunks of contiguous memory areas and their offsets in
34 * vmcore file.
35 */
36static LIST_HEAD(vmcore_list);
37
38/* Stores the pointer to the buffer containing kernel elf core headers. */
39static char *elfcorebuf;
40static size_t elfcorebuf_sz;
41static size_t elfcorebuf_sz_orig;
42
43static char *elfnotes_buf;
44static size_t elfnotes_sz;
45/* Size of all notes minus the device dump notes */
46static size_t elfnotes_orig_sz;
47
48/* Total size of vmcore file. */
49static u64 vmcore_size;
50
51static struct proc_dir_entry *proc_vmcore;
52
53#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
54/* Device Dump list and mutex to synchronize access to list */
55static LIST_HEAD(vmcoredd_list);
56static DEFINE_MUTEX(vmcoredd_mutex);
57
58static bool vmcoredd_disabled;
59core_param(novmcoredd, vmcoredd_disabled, bool, 0);
60#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
61
62/* Device Dump Size */
63static size_t vmcoredd_orig_sz;
64
65/*
66 * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
67 * The called function has to take care of module refcounting.
68 */
69static int (*oldmem_pfn_is_ram)(unsigned long pfn);
70
71int register_oldmem_pfn_is_ram(int (*fn)(unsigned long pfn))
72{
73 if (oldmem_pfn_is_ram)
74 return -EBUSY;
75 oldmem_pfn_is_ram = fn;
76 return 0;
77}
78EXPORT_SYMBOL_GPL(register_oldmem_pfn_is_ram);
79
80void unregister_oldmem_pfn_is_ram(void)
81{
82 oldmem_pfn_is_ram = NULL;
83 wmb();
84}
85EXPORT_SYMBOL_GPL(unregister_oldmem_pfn_is_ram);
86
87static int pfn_is_ram(unsigned long pfn)
88{
89 int (*fn)(unsigned long pfn);
90 /* pfn is ram unless fn() checks pagetype */
91 int ret = 1;
92
93 /*
94 * Ask hypervisor if the pfn is really ram.
95 * A ballooned page contains no data and reading from such a page
96 * will cause high load in the hypervisor.
97 */
98 fn = oldmem_pfn_is_ram;
99 if (fn)
100 ret = fn(pfn);
101
102 return ret;
103}
104
105/* Reads a page from the oldmem device from given offset. */
106ssize_t read_from_oldmem(char *buf, size_t count,
107 u64 *ppos, int userbuf,
108 bool encrypted)
109{
110 unsigned long pfn, offset;
111 size_t nr_bytes;
112 ssize_t read = 0, tmp;
113
114 if (!count)
115 return 0;
116
117 offset = (unsigned long)(*ppos % PAGE_SIZE);
118 pfn = (unsigned long)(*ppos / PAGE_SIZE);
119
120 do {
121 if (count > (PAGE_SIZE - offset))
122 nr_bytes = PAGE_SIZE - offset;
123 else
124 nr_bytes = count;
125
126 /* If pfn is not ram, return zeros for sparse dump files */
127 if (pfn_is_ram(pfn) == 0)
128 memset(buf, 0, nr_bytes);
129 else {
130 if (encrypted)
131 tmp = copy_oldmem_page_encrypted(pfn, buf,
132 nr_bytes,
133 offset,
134 userbuf);
135 else
136 tmp = copy_oldmem_page(pfn, buf, nr_bytes,
137 offset, userbuf);
138
139 if (tmp < 0)
140 return tmp;
141 }
142 *ppos += nr_bytes;
143 count -= nr_bytes;
144 buf += nr_bytes;
145 read += nr_bytes;
146 ++pfn;
147 offset = 0;
148 } while (count);
149
150 return read;
151}
152
153/*
154 * Architectures may override this function to allocate ELF header in 2nd kernel
155 */
156int __weak elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
157{
158 return 0;
159}
160
161/*
162 * Architectures may override this function to free header
163 */
164void __weak elfcorehdr_free(unsigned long long addr)
165{}
166
167/*
168 * Architectures may override this function to read from ELF header
169 */
170ssize_t __weak elfcorehdr_read(char *buf, size_t count, u64 *ppos)
171{
172 return read_from_oldmem(buf, count, ppos, 0, false);
173}
174
175/*
176 * Architectures may override this function to read from notes sections
177 */
178ssize_t __weak elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
179{
180 return read_from_oldmem(buf, count, ppos, 0, mem_encrypt_active());
181}
182
183/*
184 * Architectures may override this function to map oldmem
185 */
186int __weak remap_oldmem_pfn_range(struct vm_area_struct *vma,
187 unsigned long from, unsigned long pfn,
188 unsigned long size, pgprot_t prot)
189{
190 prot = pgprot_encrypted(prot);
191 return remap_pfn_range(vma, from, pfn, size, prot);
192}
193
194/*
195 * Architectures which support memory encryption override this.
196 */
197ssize_t __weak
198copy_oldmem_page_encrypted(unsigned long pfn, char *buf, size_t csize,
199 unsigned long offset, int userbuf)
200{
201 return copy_oldmem_page(pfn, buf, csize, offset, userbuf);
202}
203
204/*
205 * Copy to either kernel or user space
206 */
207static int copy_to(void *target, void *src, size_t size, int userbuf)
208{
209 if (userbuf) {
210 if (copy_to_user((char __user *) target, src, size))
211 return -EFAULT;
212 } else {
213 memcpy(target, src, size);
214 }
215 return 0;
216}
217
218#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
219static int vmcoredd_copy_dumps(void *dst, u64 start, size_t size, int userbuf)
220{
221 struct vmcoredd_node *dump;
222 u64 offset = 0;
223 int ret = 0;
224 size_t tsz;
225 char *buf;
226
227 mutex_lock(&vmcoredd_mutex);
228 list_for_each_entry(dump, &vmcoredd_list, list) {
229 if (start < offset + dump->size) {
230 tsz = min(offset + (u64)dump->size - start, (u64)size);
231 buf = dump->buf + start - offset;
232 if (copy_to(dst, buf, tsz, userbuf)) {
233 ret = -EFAULT;
234 goto out_unlock;
235 }
236
237 size -= tsz;
238 start += tsz;
239 dst += tsz;
240
241 /* Leave now if buffer filled already */
242 if (!size)
243 goto out_unlock;
244 }
245 offset += dump->size;
246 }
247
248out_unlock:
249 mutex_unlock(&vmcoredd_mutex);
250 return ret;
251}
252
253#ifdef CONFIG_MMU
254static int vmcoredd_mmap_dumps(struct vm_area_struct *vma, unsigned long dst,
255 u64 start, size_t size)
256{
257 struct vmcoredd_node *dump;
258 u64 offset = 0;
259 int ret = 0;
260 size_t tsz;
261 char *buf;
262
263 mutex_lock(&vmcoredd_mutex);
264 list_for_each_entry(dump, &vmcoredd_list, list) {
265 if (start < offset + dump->size) {
266 tsz = min(offset + (u64)dump->size - start, (u64)size);
267 buf = dump->buf + start - offset;
268 if (remap_vmalloc_range_partial(vma, dst, buf, 0,
269 tsz)) {
270 ret = -EFAULT;
271 goto out_unlock;
272 }
273
274 size -= tsz;
275 start += tsz;
276 dst += tsz;
277
278 /* Leave now if buffer filled already */
279 if (!size)
280 goto out_unlock;
281 }
282 offset += dump->size;
283 }
284
285out_unlock:
286 mutex_unlock(&vmcoredd_mutex);
287 return ret;
288}
289#endif /* CONFIG_MMU */
290#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
291
292/* Read from the ELF header and then the crash dump. On error, negative value is
293 * returned otherwise number of bytes read are returned.
294 */
295static ssize_t __read_vmcore(char *buffer, size_t buflen, loff_t *fpos,
296 int userbuf)
297{
298 ssize_t acc = 0, tmp;
299 size_t tsz;
300 u64 start;
301 struct vmcore *m = NULL;
302
303 if (buflen == 0 || *fpos >= vmcore_size)
304 return 0;
305
306 /* trim buflen to not go beyond EOF */
307 if (buflen > vmcore_size - *fpos)
308 buflen = vmcore_size - *fpos;
309
310 /* Read ELF core header */
311 if (*fpos < elfcorebuf_sz) {
312 tsz = min(elfcorebuf_sz - (size_t)*fpos, buflen);
313 if (copy_to(buffer, elfcorebuf + *fpos, tsz, userbuf))
314 return -EFAULT;
315 buflen -= tsz;
316 *fpos += tsz;
317 buffer += tsz;
318 acc += tsz;
319
320 /* leave now if filled buffer already */
321 if (buflen == 0)
322 return acc;
323 }
324
325 /* Read Elf note segment */
326 if (*fpos < elfcorebuf_sz + elfnotes_sz) {
327 void *kaddr;
328
329 /* We add device dumps before other elf notes because the
330 * other elf notes may not fill the elf notes buffer
331 * completely and we will end up with zero-filled data
332 * between the elf notes and the device dumps. Tools will
333 * then try to decode this zero-filled data as valid notes
334 * and we don't want that. Hence, adding device dumps before
335 * the other elf notes ensure that zero-filled data can be
336 * avoided.
337 */
338#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
339 /* Read device dumps */
340 if (*fpos < elfcorebuf_sz + vmcoredd_orig_sz) {
341 tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
342 (size_t)*fpos, buflen);
343 start = *fpos - elfcorebuf_sz;
344 if (vmcoredd_copy_dumps(buffer, start, tsz, userbuf))
345 return -EFAULT;
346
347 buflen -= tsz;
348 *fpos += tsz;
349 buffer += tsz;
350 acc += tsz;
351
352 /* leave now if filled buffer already */
353 if (!buflen)
354 return acc;
355 }
356#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
357
358 /* Read remaining elf notes */
359 tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)*fpos, buflen);
360 kaddr = elfnotes_buf + *fpos - elfcorebuf_sz - vmcoredd_orig_sz;
361 if (copy_to(buffer, kaddr, tsz, userbuf))
362 return -EFAULT;
363
364 buflen -= tsz;
365 *fpos += tsz;
366 buffer += tsz;
367 acc += tsz;
368
369 /* leave now if filled buffer already */
370 if (buflen == 0)
371 return acc;
372 }
373
374 list_for_each_entry(m, &vmcore_list, list) {
375 if (*fpos < m->offset + m->size) {
376 tsz = (size_t)min_t(unsigned long long,
377 m->offset + m->size - *fpos,
378 buflen);
379 start = m->paddr + *fpos - m->offset;
380 tmp = read_from_oldmem(buffer, tsz, &start,
381 userbuf, mem_encrypt_active());
382 if (tmp < 0)
383 return tmp;
384 buflen -= tsz;
385 *fpos += tsz;
386 buffer += tsz;
387 acc += tsz;
388
389 /* leave now if filled buffer already */
390 if (buflen == 0)
391 return acc;
392 }
393 }
394
395 return acc;
396}
397
398static ssize_t read_vmcore(struct file *file, char __user *buffer,
399 size_t buflen, loff_t *fpos)
400{
401 return __read_vmcore((__force char *) buffer, buflen, fpos, 1);
402}
403
404/*
405 * The vmcore fault handler uses the page cache and fills data using the
406 * standard __vmcore_read() function.
407 *
408 * On s390 the fault handler is used for memory regions that can't be mapped
409 * directly with remap_pfn_range().
410 */
411static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
412{
413#ifdef CONFIG_S390
414 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
415 pgoff_t index = vmf->pgoff;
416 struct page *page;
417 loff_t offset;
418 char *buf;
419 int rc;
420
421 page = find_or_create_page(mapping, index, GFP_KERNEL);
422 if (!page)
423 return VM_FAULT_OOM;
424 if (!PageUptodate(page)) {
425 offset = (loff_t) index << PAGE_SHIFT;
426 buf = __va((page_to_pfn(page) << PAGE_SHIFT));
427 rc = __read_vmcore(buf, PAGE_SIZE, &offset, 0);
428 if (rc < 0) {
429 unlock_page(page);
430 put_page(page);
431 return vmf_error(rc);
432 }
433 SetPageUptodate(page);
434 }
435 unlock_page(page);
436 vmf->page = page;
437 return 0;
438#else
439 return VM_FAULT_SIGBUS;
440#endif
441}
442
443static const struct vm_operations_struct vmcore_mmap_ops = {
444 .fault = mmap_vmcore_fault,
445};
446
447/**
448 * vmcore_alloc_buf - allocate buffer in vmalloc memory
449 * @sizez: size of buffer
450 *
451 * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
452 * the buffer to user-space by means of remap_vmalloc_range().
453 *
454 * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
455 * disabled and there's no need to allow users to mmap the buffer.
456 */
457static inline char *vmcore_alloc_buf(size_t size)
458{
459#ifdef CONFIG_MMU
460 return vmalloc_user(size);
461#else
462 return vzalloc(size);
463#endif
464}
465
466/*
467 * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
468 * essential for mmap_vmcore() in order to map physically
469 * non-contiguous objects (ELF header, ELF note segment and memory
470 * regions in the 1st kernel pointed to by PT_LOAD entries) into
471 * virtually contiguous user-space in ELF layout.
472 */
473#ifdef CONFIG_MMU
474/*
475 * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
476 * reported as not being ram with the zero page.
477 *
478 * @vma: vm_area_struct describing requested mapping
479 * @from: start remapping from
480 * @pfn: page frame number to start remapping to
481 * @size: remapping size
482 * @prot: protection bits
483 *
484 * Returns zero on success, -EAGAIN on failure.
485 */
486static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
487 unsigned long from, unsigned long pfn,
488 unsigned long size, pgprot_t prot)
489{
490 unsigned long map_size;
491 unsigned long pos_start, pos_end, pos;
492 unsigned long zeropage_pfn = my_zero_pfn(0);
493 size_t len = 0;
494
495 pos_start = pfn;
496 pos_end = pfn + (size >> PAGE_SHIFT);
497
498 for (pos = pos_start; pos < pos_end; ++pos) {
499 if (!pfn_is_ram(pos)) {
500 /*
501 * We hit a page which is not ram. Remap the continuous
502 * region between pos_start and pos-1 and replace
503 * the non-ram page at pos with the zero page.
504 */
505 if (pos > pos_start) {
506 /* Remap continuous region */
507 map_size = (pos - pos_start) << PAGE_SHIFT;
508 if (remap_oldmem_pfn_range(vma, from + len,
509 pos_start, map_size,
510 prot))
511 goto fail;
512 len += map_size;
513 }
514 /* Remap the zero page */
515 if (remap_oldmem_pfn_range(vma, from + len,
516 zeropage_pfn,
517 PAGE_SIZE, prot))
518 goto fail;
519 len += PAGE_SIZE;
520 pos_start = pos + 1;
521 }
522 }
523 if (pos > pos_start) {
524 /* Remap the rest */
525 map_size = (pos - pos_start) << PAGE_SHIFT;
526 if (remap_oldmem_pfn_range(vma, from + len, pos_start,
527 map_size, prot))
528 goto fail;
529 }
530 return 0;
531fail:
532 do_munmap(vma->vm_mm, from, len, NULL);
533 return -EAGAIN;
534}
535
536static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
537 unsigned long from, unsigned long pfn,
538 unsigned long size, pgprot_t prot)
539{
540 /*
541 * Check if oldmem_pfn_is_ram was registered to avoid
542 * looping over all pages without a reason.
543 */
544 if (oldmem_pfn_is_ram)
545 return remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
546 else
547 return remap_oldmem_pfn_range(vma, from, pfn, size, prot);
548}
549
550static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
551{
552 size_t size = vma->vm_end - vma->vm_start;
553 u64 start, end, len, tsz;
554 struct vmcore *m;
555
556 start = (u64)vma->vm_pgoff << PAGE_SHIFT;
557 end = start + size;
558
559 if (size > vmcore_size || end > vmcore_size)
560 return -EINVAL;
561
562 if (vma->vm_flags & (VM_WRITE | VM_EXEC))
563 return -EPERM;
564
565 vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
566 vma->vm_flags |= VM_MIXEDMAP;
567 vma->vm_ops = &vmcore_mmap_ops;
568
569 len = 0;
570
571 if (start < elfcorebuf_sz) {
572 u64 pfn;
573
574 tsz = min(elfcorebuf_sz - (size_t)start, size);
575 pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
576 if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
577 vma->vm_page_prot))
578 return -EAGAIN;
579 size -= tsz;
580 start += tsz;
581 len += tsz;
582
583 if (size == 0)
584 return 0;
585 }
586
587 if (start < elfcorebuf_sz + elfnotes_sz) {
588 void *kaddr;
589
590 /* We add device dumps before other elf notes because the
591 * other elf notes may not fill the elf notes buffer
592 * completely and we will end up with zero-filled data
593 * between the elf notes and the device dumps. Tools will
594 * then try to decode this zero-filled data as valid notes
595 * and we don't want that. Hence, adding device dumps before
596 * the other elf notes ensure that zero-filled data can be
597 * avoided. This also ensures that the device dumps and
598 * other elf notes can be properly mmaped at page aligned
599 * address.
600 */
601#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
602 /* Read device dumps */
603 if (start < elfcorebuf_sz + vmcoredd_orig_sz) {
604 u64 start_off;
605
606 tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
607 (size_t)start, size);
608 start_off = start - elfcorebuf_sz;
609 if (vmcoredd_mmap_dumps(vma, vma->vm_start + len,
610 start_off, tsz))
611 goto fail;
612
613 size -= tsz;
614 start += tsz;
615 len += tsz;
616
617 /* leave now if filled buffer already */
618 if (!size)
619 return 0;
620 }
621#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
622
623 /* Read remaining elf notes */
624 tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
625 kaddr = elfnotes_buf + start - elfcorebuf_sz - vmcoredd_orig_sz;
626 if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
627 kaddr, 0, tsz))
628 goto fail;
629
630 size -= tsz;
631 start += tsz;
632 len += tsz;
633
634 if (size == 0)
635 return 0;
636 }
637
638 list_for_each_entry(m, &vmcore_list, list) {
639 if (start < m->offset + m->size) {
640 u64 paddr = 0;
641
642 tsz = (size_t)min_t(unsigned long long,
643 m->offset + m->size - start, size);
644 paddr = m->paddr + start - m->offset;
645 if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
646 paddr >> PAGE_SHIFT, tsz,
647 vma->vm_page_prot))
648 goto fail;
649 size -= tsz;
650 start += tsz;
651 len += tsz;
652
653 if (size == 0)
654 return 0;
655 }
656 }
657
658 return 0;
659fail:
660 do_munmap(vma->vm_mm, vma->vm_start, len, NULL);
661 return -EAGAIN;
662}
663#else
664static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
665{
666 return -ENOSYS;
667}
668#endif
669
670static const struct proc_ops vmcore_proc_ops = {
671 .proc_read = read_vmcore,
672 .proc_lseek = default_llseek,
673 .proc_mmap = mmap_vmcore,
674};
675
676static struct vmcore* __init get_new_element(void)
677{
678 return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
679}
680
681static u64 get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
682 struct list_head *vc_list)
683{
684 u64 size;
685 struct vmcore *m;
686
687 size = elfsz + elfnotesegsz;
688 list_for_each_entry(m, vc_list, list) {
689 size += m->size;
690 }
691 return size;
692}
693
694/**
695 * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
696 *
697 * @ehdr_ptr: ELF header
698 *
699 * This function updates p_memsz member of each PT_NOTE entry in the
700 * program header table pointed to by @ehdr_ptr to real size of ELF
701 * note segment.
702 */
703static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
704{
705 int i, rc=0;
706 Elf64_Phdr *phdr_ptr;
707 Elf64_Nhdr *nhdr_ptr;
708
709 phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
710 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
711 void *notes_section;
712 u64 offset, max_sz, sz, real_sz = 0;
713 if (phdr_ptr->p_type != PT_NOTE)
714 continue;
715 max_sz = phdr_ptr->p_memsz;
716 offset = phdr_ptr->p_offset;
717 notes_section = kmalloc(max_sz, GFP_KERNEL);
718 if (!notes_section)
719 return -ENOMEM;
720 rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
721 if (rc < 0) {
722 kfree(notes_section);
723 return rc;
724 }
725 nhdr_ptr = notes_section;
726 while (nhdr_ptr->n_namesz != 0) {
727 sz = sizeof(Elf64_Nhdr) +
728 (((u64)nhdr_ptr->n_namesz + 3) & ~3) +
729 (((u64)nhdr_ptr->n_descsz + 3) & ~3);
730 if ((real_sz + sz) > max_sz) {
731 pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
732 nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
733 break;
734 }
735 real_sz += sz;
736 nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
737 }
738 kfree(notes_section);
739 phdr_ptr->p_memsz = real_sz;
740 if (real_sz == 0) {
741 pr_warn("Warning: Zero PT_NOTE entries found\n");
742 }
743 }
744
745 return 0;
746}
747
748/**
749 * get_note_number_and_size_elf64 - get the number of PT_NOTE program
750 * headers and sum of real size of their ELF note segment headers and
751 * data.
752 *
753 * @ehdr_ptr: ELF header
754 * @nr_ptnote: buffer for the number of PT_NOTE program headers
755 * @sz_ptnote: buffer for size of unique PT_NOTE program header
756 *
757 * This function is used to merge multiple PT_NOTE program headers
758 * into a unique single one. The resulting unique entry will have
759 * @sz_ptnote in its phdr->p_mem.
760 *
761 * It is assumed that program headers with PT_NOTE type pointed to by
762 * @ehdr_ptr has already been updated by update_note_header_size_elf64
763 * and each of PT_NOTE program headers has actual ELF note segment
764 * size in its p_memsz member.
765 */
766static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
767 int *nr_ptnote, u64 *sz_ptnote)
768{
769 int i;
770 Elf64_Phdr *phdr_ptr;
771
772 *nr_ptnote = *sz_ptnote = 0;
773
774 phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
775 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
776 if (phdr_ptr->p_type != PT_NOTE)
777 continue;
778 *nr_ptnote += 1;
779 *sz_ptnote += phdr_ptr->p_memsz;
780 }
781
782 return 0;
783}
784
785/**
786 * copy_notes_elf64 - copy ELF note segments in a given buffer
787 *
788 * @ehdr_ptr: ELF header
789 * @notes_buf: buffer into which ELF note segments are copied
790 *
791 * This function is used to copy ELF note segment in the 1st kernel
792 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
793 * size of the buffer @notes_buf is equal to or larger than sum of the
794 * real ELF note segment headers and data.
795 *
796 * It is assumed that program headers with PT_NOTE type pointed to by
797 * @ehdr_ptr has already been updated by update_note_header_size_elf64
798 * and each of PT_NOTE program headers has actual ELF note segment
799 * size in its p_memsz member.
800 */
801static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
802{
803 int i, rc=0;
804 Elf64_Phdr *phdr_ptr;
805
806 phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
807
808 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
809 u64 offset;
810 if (phdr_ptr->p_type != PT_NOTE)
811 continue;
812 offset = phdr_ptr->p_offset;
813 rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
814 &offset);
815 if (rc < 0)
816 return rc;
817 notes_buf += phdr_ptr->p_memsz;
818 }
819
820 return 0;
821}
822
823/* Merges all the PT_NOTE headers into one. */
824static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
825 char **notes_buf, size_t *notes_sz)
826{
827 int i, nr_ptnote=0, rc=0;
828 char *tmp;
829 Elf64_Ehdr *ehdr_ptr;
830 Elf64_Phdr phdr;
831 u64 phdr_sz = 0, note_off;
832
833 ehdr_ptr = (Elf64_Ehdr *)elfptr;
834
835 rc = update_note_header_size_elf64(ehdr_ptr);
836 if (rc < 0)
837 return rc;
838
839 rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
840 if (rc < 0)
841 return rc;
842
843 *notes_sz = roundup(phdr_sz, PAGE_SIZE);
844 *notes_buf = vmcore_alloc_buf(*notes_sz);
845 if (!*notes_buf)
846 return -ENOMEM;
847
848 rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
849 if (rc < 0)
850 return rc;
851
852 /* Prepare merged PT_NOTE program header. */
853 phdr.p_type = PT_NOTE;
854 phdr.p_flags = 0;
855 note_off = sizeof(Elf64_Ehdr) +
856 (ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
857 phdr.p_offset = roundup(note_off, PAGE_SIZE);
858 phdr.p_vaddr = phdr.p_paddr = 0;
859 phdr.p_filesz = phdr.p_memsz = phdr_sz;
860 phdr.p_align = 0;
861
862 /* Add merged PT_NOTE program header*/
863 tmp = elfptr + sizeof(Elf64_Ehdr);
864 memcpy(tmp, &phdr, sizeof(phdr));
865 tmp += sizeof(phdr);
866
867 /* Remove unwanted PT_NOTE program headers. */
868 i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
869 *elfsz = *elfsz - i;
870 memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
871 memset(elfptr + *elfsz, 0, i);
872 *elfsz = roundup(*elfsz, PAGE_SIZE);
873
874 /* Modify e_phnum to reflect merged headers. */
875 ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
876
877 /* Store the size of all notes. We need this to update the note
878 * header when the device dumps will be added.
879 */
880 elfnotes_orig_sz = phdr.p_memsz;
881
882 return 0;
883}
884
885/**
886 * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
887 *
888 * @ehdr_ptr: ELF header
889 *
890 * This function updates p_memsz member of each PT_NOTE entry in the
891 * program header table pointed to by @ehdr_ptr to real size of ELF
892 * note segment.
893 */
894static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
895{
896 int i, rc=0;
897 Elf32_Phdr *phdr_ptr;
898 Elf32_Nhdr *nhdr_ptr;
899
900 phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
901 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
902 void *notes_section;
903 u64 offset, max_sz, sz, real_sz = 0;
904 if (phdr_ptr->p_type != PT_NOTE)
905 continue;
906 max_sz = phdr_ptr->p_memsz;
907 offset = phdr_ptr->p_offset;
908 notes_section = kmalloc(max_sz, GFP_KERNEL);
909 if (!notes_section)
910 return -ENOMEM;
911 rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
912 if (rc < 0) {
913 kfree(notes_section);
914 return rc;
915 }
916 nhdr_ptr = notes_section;
917 while (nhdr_ptr->n_namesz != 0) {
918 sz = sizeof(Elf32_Nhdr) +
919 (((u64)nhdr_ptr->n_namesz + 3) & ~3) +
920 (((u64)nhdr_ptr->n_descsz + 3) & ~3);
921 if ((real_sz + sz) > max_sz) {
922 pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
923 nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
924 break;
925 }
926 real_sz += sz;
927 nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
928 }
929 kfree(notes_section);
930 phdr_ptr->p_memsz = real_sz;
931 if (real_sz == 0) {
932 pr_warn("Warning: Zero PT_NOTE entries found\n");
933 }
934 }
935
936 return 0;
937}
938
939/**
940 * get_note_number_and_size_elf32 - get the number of PT_NOTE program
941 * headers and sum of real size of their ELF note segment headers and
942 * data.
943 *
944 * @ehdr_ptr: ELF header
945 * @nr_ptnote: buffer for the number of PT_NOTE program headers
946 * @sz_ptnote: buffer for size of unique PT_NOTE program header
947 *
948 * This function is used to merge multiple PT_NOTE program headers
949 * into a unique single one. The resulting unique entry will have
950 * @sz_ptnote in its phdr->p_mem.
951 *
952 * It is assumed that program headers with PT_NOTE type pointed to by
953 * @ehdr_ptr has already been updated by update_note_header_size_elf32
954 * and each of PT_NOTE program headers has actual ELF note segment
955 * size in its p_memsz member.
956 */
957static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
958 int *nr_ptnote, u64 *sz_ptnote)
959{
960 int i;
961 Elf32_Phdr *phdr_ptr;
962
963 *nr_ptnote = *sz_ptnote = 0;
964
965 phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
966 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
967 if (phdr_ptr->p_type != PT_NOTE)
968 continue;
969 *nr_ptnote += 1;
970 *sz_ptnote += phdr_ptr->p_memsz;
971 }
972
973 return 0;
974}
975
976/**
977 * copy_notes_elf32 - copy ELF note segments in a given buffer
978 *
979 * @ehdr_ptr: ELF header
980 * @notes_buf: buffer into which ELF note segments are copied
981 *
982 * This function is used to copy ELF note segment in the 1st kernel
983 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
984 * size of the buffer @notes_buf is equal to or larger than sum of the
985 * real ELF note segment headers and data.
986 *
987 * It is assumed that program headers with PT_NOTE type pointed to by
988 * @ehdr_ptr has already been updated by update_note_header_size_elf32
989 * and each of PT_NOTE program headers has actual ELF note segment
990 * size in its p_memsz member.
991 */
992static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
993{
994 int i, rc=0;
995 Elf32_Phdr *phdr_ptr;
996
997 phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
998
999 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1000 u64 offset;
1001 if (phdr_ptr->p_type != PT_NOTE)
1002 continue;
1003 offset = phdr_ptr->p_offset;
1004 rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
1005 &offset);
1006 if (rc < 0)
1007 return rc;
1008 notes_buf += phdr_ptr->p_memsz;
1009 }
1010
1011 return 0;
1012}
1013
1014/* Merges all the PT_NOTE headers into one. */
1015static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
1016 char **notes_buf, size_t *notes_sz)
1017{
1018 int i, nr_ptnote=0, rc=0;
1019 char *tmp;
1020 Elf32_Ehdr *ehdr_ptr;
1021 Elf32_Phdr phdr;
1022 u64 phdr_sz = 0, note_off;
1023
1024 ehdr_ptr = (Elf32_Ehdr *)elfptr;
1025
1026 rc = update_note_header_size_elf32(ehdr_ptr);
1027 if (rc < 0)
1028 return rc;
1029
1030 rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
1031 if (rc < 0)
1032 return rc;
1033
1034 *notes_sz = roundup(phdr_sz, PAGE_SIZE);
1035 *notes_buf = vmcore_alloc_buf(*notes_sz);
1036 if (!*notes_buf)
1037 return -ENOMEM;
1038
1039 rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
1040 if (rc < 0)
1041 return rc;
1042
1043 /* Prepare merged PT_NOTE program header. */
1044 phdr.p_type = PT_NOTE;
1045 phdr.p_flags = 0;
1046 note_off = sizeof(Elf32_Ehdr) +
1047 (ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
1048 phdr.p_offset = roundup(note_off, PAGE_SIZE);
1049 phdr.p_vaddr = phdr.p_paddr = 0;
1050 phdr.p_filesz = phdr.p_memsz = phdr_sz;
1051 phdr.p_align = 0;
1052
1053 /* Add merged PT_NOTE program header*/
1054 tmp = elfptr + sizeof(Elf32_Ehdr);
1055 memcpy(tmp, &phdr, sizeof(phdr));
1056 tmp += sizeof(phdr);
1057
1058 /* Remove unwanted PT_NOTE program headers. */
1059 i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
1060 *elfsz = *elfsz - i;
1061 memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
1062 memset(elfptr + *elfsz, 0, i);
1063 *elfsz = roundup(*elfsz, PAGE_SIZE);
1064
1065 /* Modify e_phnum to reflect merged headers. */
1066 ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
1067
1068 /* Store the size of all notes. We need this to update the note
1069 * header when the device dumps will be added.
1070 */
1071 elfnotes_orig_sz = phdr.p_memsz;
1072
1073 return 0;
1074}
1075
1076/* Add memory chunks represented by program headers to vmcore list. Also update
1077 * the new offset fields of exported program headers. */
1078static int __init process_ptload_program_headers_elf64(char *elfptr,
1079 size_t elfsz,
1080 size_t elfnotes_sz,
1081 struct list_head *vc_list)
1082{
1083 int i;
1084 Elf64_Ehdr *ehdr_ptr;
1085 Elf64_Phdr *phdr_ptr;
1086 loff_t vmcore_off;
1087 struct vmcore *new;
1088
1089 ehdr_ptr = (Elf64_Ehdr *)elfptr;
1090 phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
1091
1092 /* Skip Elf header, program headers and Elf note segment. */
1093 vmcore_off = elfsz + elfnotes_sz;
1094
1095 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1096 u64 paddr, start, end, size;
1097
1098 if (phdr_ptr->p_type != PT_LOAD)
1099 continue;
1100
1101 paddr = phdr_ptr->p_offset;
1102 start = rounddown(paddr, PAGE_SIZE);
1103 end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1104 size = end - start;
1105
1106 /* Add this contiguous chunk of memory to vmcore list.*/
1107 new = get_new_element();
1108 if (!new)
1109 return -ENOMEM;
1110 new->paddr = start;
1111 new->size = size;
1112 list_add_tail(&new->list, vc_list);
1113
1114 /* Update the program header offset. */
1115 phdr_ptr->p_offset = vmcore_off + (paddr - start);
1116 vmcore_off = vmcore_off + size;
1117 }
1118 return 0;
1119}
1120
1121static int __init process_ptload_program_headers_elf32(char *elfptr,
1122 size_t elfsz,
1123 size_t elfnotes_sz,
1124 struct list_head *vc_list)
1125{
1126 int i;
1127 Elf32_Ehdr *ehdr_ptr;
1128 Elf32_Phdr *phdr_ptr;
1129 loff_t vmcore_off;
1130 struct vmcore *new;
1131
1132 ehdr_ptr = (Elf32_Ehdr *)elfptr;
1133 phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
1134
1135 /* Skip Elf header, program headers and Elf note segment. */
1136 vmcore_off = elfsz + elfnotes_sz;
1137
1138 for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1139 u64 paddr, start, end, size;
1140
1141 if (phdr_ptr->p_type != PT_LOAD)
1142 continue;
1143
1144 paddr = phdr_ptr->p_offset;
1145 start = rounddown(paddr, PAGE_SIZE);
1146 end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1147 size = end - start;
1148
1149 /* Add this contiguous chunk of memory to vmcore list.*/
1150 new = get_new_element();
1151 if (!new)
1152 return -ENOMEM;
1153 new->paddr = start;
1154 new->size = size;
1155 list_add_tail(&new->list, vc_list);
1156
1157 /* Update the program header offset */
1158 phdr_ptr->p_offset = vmcore_off + (paddr - start);
1159 vmcore_off = vmcore_off + size;
1160 }
1161 return 0;
1162}
1163
1164/* Sets offset fields of vmcore elements. */
1165static void set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
1166 struct list_head *vc_list)
1167{
1168 loff_t vmcore_off;
1169 struct vmcore *m;
1170
1171 /* Skip Elf header, program headers and Elf note segment. */
1172 vmcore_off = elfsz + elfnotes_sz;
1173
1174 list_for_each_entry(m, vc_list, list) {
1175 m->offset = vmcore_off;
1176 vmcore_off += m->size;
1177 }
1178}
1179
1180static void free_elfcorebuf(void)
1181{
1182 free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
1183 elfcorebuf = NULL;
1184 vfree(elfnotes_buf);
1185 elfnotes_buf = NULL;
1186}
1187
1188static int __init parse_crash_elf64_headers(void)
1189{
1190 int rc=0;
1191 Elf64_Ehdr ehdr;
1192 u64 addr;
1193
1194 addr = elfcorehdr_addr;
1195
1196 /* Read Elf header */
1197 rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
1198 if (rc < 0)
1199 return rc;
1200
1201 /* Do some basic Verification. */
1202 if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1203 (ehdr.e_type != ET_CORE) ||
1204 !vmcore_elf64_check_arch(&ehdr) ||
1205 ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
1206 ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1207 ehdr.e_version != EV_CURRENT ||
1208 ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
1209 ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
1210 ehdr.e_phnum == 0) {
1211 pr_warn("Warning: Core image elf header is not sane\n");
1212 return -EINVAL;
1213 }
1214
1215 /* Read in all elf headers. */
1216 elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
1217 ehdr.e_phnum * sizeof(Elf64_Phdr);
1218 elfcorebuf_sz = elfcorebuf_sz_orig;
1219 elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1220 get_order(elfcorebuf_sz_orig));
1221 if (!elfcorebuf)
1222 return -ENOMEM;
1223 addr = elfcorehdr_addr;
1224 rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1225 if (rc < 0)
1226 goto fail;
1227
1228 /* Merge all PT_NOTE headers into one. */
1229 rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
1230 &elfnotes_buf, &elfnotes_sz);
1231 if (rc)
1232 goto fail;
1233 rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
1234 elfnotes_sz, &vmcore_list);
1235 if (rc)
1236 goto fail;
1237 set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1238 return 0;
1239fail:
1240 free_elfcorebuf();
1241 return rc;
1242}
1243
1244static int __init parse_crash_elf32_headers(void)
1245{
1246 int rc=0;
1247 Elf32_Ehdr ehdr;
1248 u64 addr;
1249
1250 addr = elfcorehdr_addr;
1251
1252 /* Read Elf header */
1253 rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
1254 if (rc < 0)
1255 return rc;
1256
1257 /* Do some basic Verification. */
1258 if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1259 (ehdr.e_type != ET_CORE) ||
1260 !vmcore_elf32_check_arch(&ehdr) ||
1261 ehdr.e_ident[EI_CLASS] != ELFCLASS32||
1262 ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1263 ehdr.e_version != EV_CURRENT ||
1264 ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1265 ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1266 ehdr.e_phnum == 0) {
1267 pr_warn("Warning: Core image elf header is not sane\n");
1268 return -EINVAL;
1269 }
1270
1271 /* Read in all elf headers. */
1272 elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1273 elfcorebuf_sz = elfcorebuf_sz_orig;
1274 elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1275 get_order(elfcorebuf_sz_orig));
1276 if (!elfcorebuf)
1277 return -ENOMEM;
1278 addr = elfcorehdr_addr;
1279 rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1280 if (rc < 0)
1281 goto fail;
1282
1283 /* Merge all PT_NOTE headers into one. */
1284 rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1285 &elfnotes_buf, &elfnotes_sz);
1286 if (rc)
1287 goto fail;
1288 rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1289 elfnotes_sz, &vmcore_list);
1290 if (rc)
1291 goto fail;
1292 set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1293 return 0;
1294fail:
1295 free_elfcorebuf();
1296 return rc;
1297}
1298
1299static int __init parse_crash_elf_headers(void)
1300{
1301 unsigned char e_ident[EI_NIDENT];
1302 u64 addr;
1303 int rc=0;
1304
1305 addr = elfcorehdr_addr;
1306 rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1307 if (rc < 0)
1308 return rc;
1309 if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1310 pr_warn("Warning: Core image elf header not found\n");
1311 return -EINVAL;
1312 }
1313
1314 if (e_ident[EI_CLASS] == ELFCLASS64) {
1315 rc = parse_crash_elf64_headers();
1316 if (rc)
1317 return rc;
1318 } else if (e_ident[EI_CLASS] == ELFCLASS32) {
1319 rc = parse_crash_elf32_headers();
1320 if (rc)
1321 return rc;
1322 } else {
1323 pr_warn("Warning: Core image elf header is not sane\n");
1324 return -EINVAL;
1325 }
1326
1327 /* Determine vmcore size. */
1328 vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1329 &vmcore_list);
1330
1331 return 0;
1332}
1333
1334#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1335/**
1336 * vmcoredd_write_header - Write vmcore device dump header at the
1337 * beginning of the dump's buffer.
1338 * @buf: Output buffer where the note is written
1339 * @data: Dump info
1340 * @size: Size of the dump
1341 *
1342 * Fills beginning of the dump's buffer with vmcore device dump header.
1343 */
1344static void vmcoredd_write_header(void *buf, struct vmcoredd_data *data,
1345 u32 size)
1346{
1347 struct vmcoredd_header *vdd_hdr = (struct vmcoredd_header *)buf;
1348
1349 vdd_hdr->n_namesz = sizeof(vdd_hdr->name);
1350 vdd_hdr->n_descsz = size + sizeof(vdd_hdr->dump_name);
1351 vdd_hdr->n_type = NT_VMCOREDD;
1352
1353 strncpy((char *)vdd_hdr->name, VMCOREDD_NOTE_NAME,
1354 sizeof(vdd_hdr->name));
1355 memcpy(vdd_hdr->dump_name, data->dump_name, sizeof(vdd_hdr->dump_name));
1356}
1357
1358/**
1359 * vmcoredd_update_program_headers - Update all Elf program headers
1360 * @elfptr: Pointer to elf header
1361 * @elfnotesz: Size of elf notes aligned to page size
1362 * @vmcoreddsz: Size of device dumps to be added to elf note header
1363 *
1364 * Determine type of Elf header (Elf64 or Elf32) and update the elf note size.
1365 * Also update the offsets of all the program headers after the elf note header.
1366 */
1367static void vmcoredd_update_program_headers(char *elfptr, size_t elfnotesz,
1368 size_t vmcoreddsz)
1369{
1370 unsigned char *e_ident = (unsigned char *)elfptr;
1371 u64 start, end, size;
1372 loff_t vmcore_off;
1373 u32 i;
1374
1375 vmcore_off = elfcorebuf_sz + elfnotesz;
1376
1377 if (e_ident[EI_CLASS] == ELFCLASS64) {
1378 Elf64_Ehdr *ehdr = (Elf64_Ehdr *)elfptr;
1379 Elf64_Phdr *phdr = (Elf64_Phdr *)(elfptr + sizeof(Elf64_Ehdr));
1380
1381 /* Update all program headers */
1382 for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1383 if (phdr->p_type == PT_NOTE) {
1384 /* Update note size */
1385 phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1386 phdr->p_filesz = phdr->p_memsz;
1387 continue;
1388 }
1389
1390 start = rounddown(phdr->p_offset, PAGE_SIZE);
1391 end = roundup(phdr->p_offset + phdr->p_memsz,
1392 PAGE_SIZE);
1393 size = end - start;
1394 phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1395 vmcore_off += size;
1396 }
1397 } else {
1398 Elf32_Ehdr *ehdr = (Elf32_Ehdr *)elfptr;
1399 Elf32_Phdr *phdr = (Elf32_Phdr *)(elfptr + sizeof(Elf32_Ehdr));
1400
1401 /* Update all program headers */
1402 for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1403 if (phdr->p_type == PT_NOTE) {
1404 /* Update note size */
1405 phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1406 phdr->p_filesz = phdr->p_memsz;
1407 continue;
1408 }
1409
1410 start = rounddown(phdr->p_offset, PAGE_SIZE);
1411 end = roundup(phdr->p_offset + phdr->p_memsz,
1412 PAGE_SIZE);
1413 size = end - start;
1414 phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1415 vmcore_off += size;
1416 }
1417 }
1418}
1419
1420/**
1421 * vmcoredd_update_size - Update the total size of the device dumps and update
1422 * Elf header
1423 * @dump_size: Size of the current device dump to be added to total size
1424 *
1425 * Update the total size of all the device dumps and update the Elf program
1426 * headers. Calculate the new offsets for the vmcore list and update the
1427 * total vmcore size.
1428 */
1429static void vmcoredd_update_size(size_t dump_size)
1430{
1431 vmcoredd_orig_sz += dump_size;
1432 elfnotes_sz = roundup(elfnotes_orig_sz, PAGE_SIZE) + vmcoredd_orig_sz;
1433 vmcoredd_update_program_headers(elfcorebuf, elfnotes_sz,
1434 vmcoredd_orig_sz);
1435
1436 /* Update vmcore list offsets */
1437 set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1438
1439 vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1440 &vmcore_list);
1441 proc_vmcore->size = vmcore_size;
1442}
1443
1444/**
1445 * vmcore_add_device_dump - Add a buffer containing device dump to vmcore
1446 * @data: dump info.
1447 *
1448 * Allocate a buffer and invoke the calling driver's dump collect routine.
1449 * Write Elf note at the beginning of the buffer to indicate vmcore device
1450 * dump and add the dump to global list.
1451 */
1452int vmcore_add_device_dump(struct vmcoredd_data *data)
1453{
1454 struct vmcoredd_node *dump;
1455 void *buf = NULL;
1456 size_t data_size;
1457 int ret;
1458
1459 if (vmcoredd_disabled) {
1460 pr_err_once("Device dump is disabled\n");
1461 return -EINVAL;
1462 }
1463
1464 if (!data || !strlen(data->dump_name) ||
1465 !data->vmcoredd_callback || !data->size)
1466 return -EINVAL;
1467
1468 dump = vzalloc(sizeof(*dump));
1469 if (!dump) {
1470 ret = -ENOMEM;
1471 goto out_err;
1472 }
1473
1474 /* Keep size of the buffer page aligned so that it can be mmaped */
1475 data_size = roundup(sizeof(struct vmcoredd_header) + data->size,
1476 PAGE_SIZE);
1477
1478 /* Allocate buffer for driver's to write their dumps */
1479 buf = vmcore_alloc_buf(data_size);
1480 if (!buf) {
1481 ret = -ENOMEM;
1482 goto out_err;
1483 }
1484
1485 vmcoredd_write_header(buf, data, data_size -
1486 sizeof(struct vmcoredd_header));
1487
1488 /* Invoke the driver's dump collection routing */
1489 ret = data->vmcoredd_callback(data, buf +
1490 sizeof(struct vmcoredd_header));
1491 if (ret)
1492 goto out_err;
1493
1494 dump->buf = buf;
1495 dump->size = data_size;
1496
1497 /* Add the dump to driver sysfs list */
1498 mutex_lock(&vmcoredd_mutex);
1499 list_add_tail(&dump->list, &vmcoredd_list);
1500 mutex_unlock(&vmcoredd_mutex);
1501
1502 vmcoredd_update_size(data_size);
1503 return 0;
1504
1505out_err:
1506 vfree(buf);
1507 vfree(dump);
1508
1509 return ret;
1510}
1511EXPORT_SYMBOL(vmcore_add_device_dump);
1512#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1513
1514/* Free all dumps in vmcore device dump list */
1515static void vmcore_free_device_dumps(void)
1516{
1517#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1518 mutex_lock(&vmcoredd_mutex);
1519 while (!list_empty(&vmcoredd_list)) {
1520 struct vmcoredd_node *dump;
1521
1522 dump = list_first_entry(&vmcoredd_list, struct vmcoredd_node,
1523 list);
1524 list_del(&dump->list);
1525 vfree(dump->buf);
1526 vfree(dump);
1527 }
1528 mutex_unlock(&vmcoredd_mutex);
1529#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1530}
1531
1532/* Init function for vmcore module. */
1533static int __init vmcore_init(void)
1534{
1535 int rc = 0;
1536
1537 /* Allow architectures to allocate ELF header in 2nd kernel */
1538 rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1539 if (rc)
1540 return rc;
1541 /*
1542 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1543 * then capture the dump.
1544 */
1545 if (!(is_vmcore_usable()))
1546 return rc;
1547 rc = parse_crash_elf_headers();
1548 if (rc) {
1549 pr_warn("Kdump: vmcore not initialized\n");
1550 return rc;
1551 }
1552 elfcorehdr_free(elfcorehdr_addr);
1553 elfcorehdr_addr = ELFCORE_ADDR_ERR;
1554
1555 proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &vmcore_proc_ops);
1556 if (proc_vmcore)
1557 proc_vmcore->size = vmcore_size;
1558 return 0;
1559}
1560fs_initcall(vmcore_init);
1561
1562/* Cleanup function for vmcore module. */
1563void vmcore_cleanup(void)
1564{
1565 if (proc_vmcore) {
1566 proc_remove(proc_vmcore);
1567 proc_vmcore = NULL;
1568 }
1569
1570 /* clear the vmcore list. */
1571 while (!list_empty(&vmcore_list)) {
1572 struct vmcore *m;
1573
1574 m = list_first_entry(&vmcore_list, struct vmcore, list);
1575 list_del(&m->list);
1576 kfree(m);
1577 }
1578 free_elfcorebuf();
1579
1580 /* clear vmcore device dump list */
1581 vmcore_free_device_dumps();
1582}