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 */
36 static LIST_HEAD(vmcore_list);
37
38 /* Stores the pointer to the buffer containing kernel elf core headers. */
39 static char *elfcorebuf;
40 static size_t elfcorebuf_sz;
41 static size_t elfcorebuf_sz_orig;
42
43 static char *elfnotes_buf;
44 static size_t elfnotes_sz;
45 /* Size of all notes minus the device dump notes */
46 static size_t elfnotes_orig_sz;
47
48 /* Total size of vmcore file. */
49 static u64 vmcore_size;
50
51 static 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 */
55 static LIST_HEAD(vmcoredd_list);
56 static DEFINE_MUTEX(vmcoredd_mutex);
57
58 static bool vmcoredd_disabled;
59 core_param(novmcoredd, vmcoredd_disabled, bool, 0);
60 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
61
62 /* Device Dump Size */
63 static size_t vmcoredd_orig_sz;
64
65 static DEFINE_SPINLOCK(vmcore_cb_lock);
66 DEFINE_STATIC_SRCU(vmcore_cb_srcu);
67 /* List of registered vmcore callbacks. */
68 static LIST_HEAD(vmcore_cb_list);
69 /* Whether the vmcore has been opened once. */
70 static bool vmcore_opened;
71
register_vmcore_cb(struct vmcore_cb * cb)72 void 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 }
85 EXPORT_SYMBOL_GPL(register_vmcore_cb);
86
unregister_vmcore_cb(struct vmcore_cb * cb)87 void 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 }
102 EXPORT_SYMBOL_GPL(unregister_vmcore_cb);
103
pfn_is_ram(unsigned long pfn)104 static 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
open_vmcore(struct inode * inode,struct file * file)121 static 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. */
read_from_oldmem(struct iov_iter * iter,size_t count,u64 * ppos,bool encrypted)131 ssize_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 */
elfcorehdr_alloc(unsigned long long * addr,unsigned long long * size)183 int __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 */
elfcorehdr_free(unsigned long long addr)191 void __weak elfcorehdr_free(unsigned long long addr)
192 {}
193
194 /*
195 * Architectures may override this function to read from ELF header
196 */
elfcorehdr_read(char * buf,size_t count,u64 * ppos)197 ssize_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 */
elfcorehdr_read_notes(char * buf,size_t count,u64 * ppos)210 ssize_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 */
remap_oldmem_pfn_range(struct vm_area_struct * vma,unsigned long from,unsigned long pfn,unsigned long size,pgprot_t prot)224 int __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 */
copy_oldmem_page_encrypted(struct iov_iter * iter,unsigned long pfn,size_t csize,unsigned long offset)235 ssize_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
vmcoredd_copy_dumps(struct iov_iter * iter,u64 start,size_t size)242 static 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
270 out_unlock:
271 mutex_unlock(&vmcoredd_mutex);
272 return ret;
273 }
274
275 #ifdef CONFIG_MMU
vmcoredd_mmap_dumps(struct vm_area_struct * vma,unsigned long dst,u64 start,size_t size)276 static 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
307 out_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 */
__read_vmcore(struct iov_iter * iter,loff_t * fpos)317 static 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
read_vmcore(struct kiocb * iocb,struct iov_iter * iter)414 static ssize_t read_vmcore(struct kiocb *iocb, struct iov_iter *iter)
415 {
416 return __read_vmcore(iter, &iocb->ki_pos);
417 }
418
419 /*
420 * The vmcore fault handler uses the page cache and fills data using the
421 * standard __read_vmcore() function.
422 *
423 * On s390 the fault handler is used for memory regions that can't be mapped
424 * directly with remap_pfn_range().
425 */
mmap_vmcore_fault(struct vm_fault * vmf)426 static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
427 {
428 #ifdef CONFIG_S390
429 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
430 pgoff_t index = vmf->pgoff;
431 struct iov_iter iter;
432 struct kvec kvec;
433 struct page *page;
434 loff_t offset;
435 int rc;
436
437 page = find_or_create_page(mapping, index, GFP_KERNEL);
438 if (!page)
439 return VM_FAULT_OOM;
440 if (!PageUptodate(page)) {
441 offset = (loff_t) index << PAGE_SHIFT;
442 kvec.iov_base = page_address(page);
443 kvec.iov_len = PAGE_SIZE;
444 iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, PAGE_SIZE);
445
446 rc = __read_vmcore(&iter, &offset);
447 if (rc < 0) {
448 unlock_page(page);
449 put_page(page);
450 return vmf_error(rc);
451 }
452 SetPageUptodate(page);
453 }
454 unlock_page(page);
455 vmf->page = page;
456 return 0;
457 #else
458 return VM_FAULT_SIGBUS;
459 #endif
460 }
461
462 /**
463 * vmcore_alloc_buf - allocate buffer in vmalloc memory
464 * @size: size of buffer
465 *
466 * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
467 * the buffer to user-space by means of remap_vmalloc_range().
468 *
469 * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
470 * disabled and there's no need to allow users to mmap the buffer.
471 */
vmcore_alloc_buf(size_t size)472 static inline char *vmcore_alloc_buf(size_t size)
473 {
474 #ifdef CONFIG_MMU
475 return vmalloc_user(size);
476 #else
477 return vzalloc(size);
478 #endif
479 }
480
481 /*
482 * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
483 * essential for mmap_vmcore() in order to map physically
484 * non-contiguous objects (ELF header, ELF note segment and memory
485 * regions in the 1st kernel pointed to by PT_LOAD entries) into
486 * virtually contiguous user-space in ELF layout.
487 */
488 #ifdef CONFIG_MMU
489
490 static 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 */
remap_oldmem_pfn_checked(struct vm_area_struct * vma,unsigned long from,unsigned long pfn,unsigned long size,pgprot_t prot)506 static 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;
551 fail:
552 do_munmap(vma->vm_mm, from, len, NULL);
553 return -EAGAIN;
554 }
555
vmcore_remap_oldmem_pfn(struct vm_area_struct * vma,unsigned long from,unsigned long pfn,unsigned long size,pgprot_t prot)556 static 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
mmap_vmcore(struct file * file,struct vm_area_struct * vma)575 static 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;
683 fail:
684 do_munmap(vma->vm_mm, vma->vm_start, len, NULL);
685 return -EAGAIN;
686 }
687 #else
mmap_vmcore(struct file * file,struct vm_area_struct * vma)688 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
689 {
690 return -ENOSYS;
691 }
692 #endif
693
694 static 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
get_new_element(void)701 static struct vmcore* __init get_new_element(void)
702 {
703 return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
704 }
705
get_vmcore_size(size_t elfsz,size_t elfnotesegsz,struct list_head * vc_list)706 static 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 */
update_note_header_size_elf64(const Elf64_Ehdr * ehdr_ptr)728 static 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 */
get_note_number_and_size_elf64(const Elf64_Ehdr * ehdr_ptr,int * nr_ptnote,u64 * sz_ptnote)791 static 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 */
copy_notes_elf64(const Elf64_Ehdr * ehdr_ptr,char * notes_buf)826 static 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. */
merge_note_headers_elf64(char * elfptr,size_t * elfsz,char ** notes_buf,size_t * notes_sz)849 static 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 */
update_note_header_size_elf32(const Elf32_Ehdr * ehdr_ptr)919 static 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 */
get_note_number_and_size_elf32(const Elf32_Ehdr * ehdr_ptr,int * nr_ptnote,u64 * sz_ptnote)982 static 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 */
copy_notes_elf32(const Elf32_Ehdr * ehdr_ptr,char * notes_buf)1017 static 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. */
merge_note_headers_elf32(char * elfptr,size_t * elfsz,char ** notes_buf,size_t * notes_sz)1040 static 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. */
process_ptload_program_headers_elf64(char * elfptr,size_t elfsz,size_t elfnotes_sz,struct list_head * vc_list)1103 static 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
process_ptload_program_headers_elf32(char * elfptr,size_t elfsz,size_t elfnotes_sz,struct list_head * vc_list)1146 static 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. */
set_vmcore_list_offsets(size_t elfsz,size_t elfnotes_sz,struct list_head * vc_list)1190 static 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
free_elfcorebuf(void)1205 static 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
parse_crash_elf64_headers(void)1213 static 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;
1264 fail:
1265 free_elfcorebuf();
1266 return rc;
1267 }
1268
parse_crash_elf32_headers(void)1269 static 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;
1319 fail:
1320 free_elfcorebuf();
1321 return rc;
1322 }
1323
parse_crash_elf_headers(void)1324 static 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 */
vmcoredd_write_header(void * buf,struct vmcoredd_data * data,u32 size)1369 static 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 strncpy((char *)vdd_hdr->name, VMCOREDD_NOTE_NAME,
1379 sizeof(vdd_hdr->name));
1380 memcpy(vdd_hdr->dump_name, data->dump_name, sizeof(vdd_hdr->dump_name));
1381 }
1382
1383 /**
1384 * vmcoredd_update_program_headers - Update all ELF program headers
1385 * @elfptr: Pointer to elf header
1386 * @elfnotesz: Size of elf notes aligned to page size
1387 * @vmcoreddsz: Size of device dumps to be added to elf note header
1388 *
1389 * Determine type of ELF header (Elf64 or Elf32) and update the elf note size.
1390 * Also update the offsets of all the program headers after the elf note header.
1391 */
vmcoredd_update_program_headers(char * elfptr,size_t elfnotesz,size_t vmcoreddsz)1392 static void vmcoredd_update_program_headers(char *elfptr, size_t elfnotesz,
1393 size_t vmcoreddsz)
1394 {
1395 unsigned char *e_ident = (unsigned char *)elfptr;
1396 u64 start, end, size;
1397 loff_t vmcore_off;
1398 u32 i;
1399
1400 vmcore_off = elfcorebuf_sz + elfnotesz;
1401
1402 if (e_ident[EI_CLASS] == ELFCLASS64) {
1403 Elf64_Ehdr *ehdr = (Elf64_Ehdr *)elfptr;
1404 Elf64_Phdr *phdr = (Elf64_Phdr *)(elfptr + sizeof(Elf64_Ehdr));
1405
1406 /* Update all program headers */
1407 for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1408 if (phdr->p_type == PT_NOTE) {
1409 /* Update note size */
1410 phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1411 phdr->p_filesz = phdr->p_memsz;
1412 continue;
1413 }
1414
1415 start = rounddown(phdr->p_offset, PAGE_SIZE);
1416 end = roundup(phdr->p_offset + phdr->p_memsz,
1417 PAGE_SIZE);
1418 size = end - start;
1419 phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1420 vmcore_off += size;
1421 }
1422 } else {
1423 Elf32_Ehdr *ehdr = (Elf32_Ehdr *)elfptr;
1424 Elf32_Phdr *phdr = (Elf32_Phdr *)(elfptr + sizeof(Elf32_Ehdr));
1425
1426 /* Update all program headers */
1427 for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1428 if (phdr->p_type == PT_NOTE) {
1429 /* Update note size */
1430 phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1431 phdr->p_filesz = phdr->p_memsz;
1432 continue;
1433 }
1434
1435 start = rounddown(phdr->p_offset, PAGE_SIZE);
1436 end = roundup(phdr->p_offset + phdr->p_memsz,
1437 PAGE_SIZE);
1438 size = end - start;
1439 phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1440 vmcore_off += size;
1441 }
1442 }
1443 }
1444
1445 /**
1446 * vmcoredd_update_size - Update the total size of the device dumps and update
1447 * ELF header
1448 * @dump_size: Size of the current device dump to be added to total size
1449 *
1450 * Update the total size of all the device dumps and update the ELF program
1451 * headers. Calculate the new offsets for the vmcore list and update the
1452 * total vmcore size.
1453 */
vmcoredd_update_size(size_t dump_size)1454 static void vmcoredd_update_size(size_t dump_size)
1455 {
1456 vmcoredd_orig_sz += dump_size;
1457 elfnotes_sz = roundup(elfnotes_orig_sz, PAGE_SIZE) + vmcoredd_orig_sz;
1458 vmcoredd_update_program_headers(elfcorebuf, elfnotes_sz,
1459 vmcoredd_orig_sz);
1460
1461 /* Update vmcore list offsets */
1462 set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1463
1464 vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1465 &vmcore_list);
1466 proc_vmcore->size = vmcore_size;
1467 }
1468
1469 /**
1470 * vmcore_add_device_dump - Add a buffer containing device dump to vmcore
1471 * @data: dump info.
1472 *
1473 * Allocate a buffer and invoke the calling driver's dump collect routine.
1474 * Write ELF note at the beginning of the buffer to indicate vmcore device
1475 * dump and add the dump to global list.
1476 */
vmcore_add_device_dump(struct vmcoredd_data * data)1477 int vmcore_add_device_dump(struct vmcoredd_data *data)
1478 {
1479 struct vmcoredd_node *dump;
1480 void *buf = NULL;
1481 size_t data_size;
1482 int ret;
1483
1484 if (vmcoredd_disabled) {
1485 pr_err_once("Device dump is disabled\n");
1486 return -EINVAL;
1487 }
1488
1489 if (!data || !strlen(data->dump_name) ||
1490 !data->vmcoredd_callback || !data->size)
1491 return -EINVAL;
1492
1493 dump = vzalloc(sizeof(*dump));
1494 if (!dump) {
1495 ret = -ENOMEM;
1496 goto out_err;
1497 }
1498
1499 /* Keep size of the buffer page aligned so that it can be mmaped */
1500 data_size = roundup(sizeof(struct vmcoredd_header) + data->size,
1501 PAGE_SIZE);
1502
1503 /* Allocate buffer for driver's to write their dumps */
1504 buf = vmcore_alloc_buf(data_size);
1505 if (!buf) {
1506 ret = -ENOMEM;
1507 goto out_err;
1508 }
1509
1510 vmcoredd_write_header(buf, data, data_size -
1511 sizeof(struct vmcoredd_header));
1512
1513 /* Invoke the driver's dump collection routing */
1514 ret = data->vmcoredd_callback(data, buf +
1515 sizeof(struct vmcoredd_header));
1516 if (ret)
1517 goto out_err;
1518
1519 dump->buf = buf;
1520 dump->size = data_size;
1521
1522 /* Add the dump to driver sysfs list */
1523 mutex_lock(&vmcoredd_mutex);
1524 list_add_tail(&dump->list, &vmcoredd_list);
1525 mutex_unlock(&vmcoredd_mutex);
1526
1527 vmcoredd_update_size(data_size);
1528 return 0;
1529
1530 out_err:
1531 vfree(buf);
1532 vfree(dump);
1533
1534 return ret;
1535 }
1536 EXPORT_SYMBOL(vmcore_add_device_dump);
1537 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1538
1539 /* Free all dumps in vmcore device dump list */
vmcore_free_device_dumps(void)1540 static void vmcore_free_device_dumps(void)
1541 {
1542 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1543 mutex_lock(&vmcoredd_mutex);
1544 while (!list_empty(&vmcoredd_list)) {
1545 struct vmcoredd_node *dump;
1546
1547 dump = list_first_entry(&vmcoredd_list, struct vmcoredd_node,
1548 list);
1549 list_del(&dump->list);
1550 vfree(dump->buf);
1551 vfree(dump);
1552 }
1553 mutex_unlock(&vmcoredd_mutex);
1554 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1555 }
1556
1557 /* Init function for vmcore module. */
vmcore_init(void)1558 static int __init vmcore_init(void)
1559 {
1560 int rc = 0;
1561
1562 /* Allow architectures to allocate ELF header in 2nd kernel */
1563 rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1564 if (rc)
1565 return rc;
1566 /*
1567 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1568 * then capture the dump.
1569 */
1570 if (!(is_vmcore_usable()))
1571 return rc;
1572 rc = parse_crash_elf_headers();
1573 if (rc) {
1574 elfcorehdr_free(elfcorehdr_addr);
1575 pr_warn("Kdump: vmcore not initialized\n");
1576 return rc;
1577 }
1578 elfcorehdr_free(elfcorehdr_addr);
1579 elfcorehdr_addr = ELFCORE_ADDR_ERR;
1580
1581 proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &vmcore_proc_ops);
1582 if (proc_vmcore)
1583 proc_vmcore->size = vmcore_size;
1584 return 0;
1585 }
1586 fs_initcall(vmcore_init);
1587
1588 /* Cleanup function for vmcore module. */
vmcore_cleanup(void)1589 void vmcore_cleanup(void)
1590 {
1591 if (proc_vmcore) {
1592 proc_remove(proc_vmcore);
1593 proc_vmcore = NULL;
1594 }
1595
1596 /* clear the vmcore list. */
1597 while (!list_empty(&vmcore_list)) {
1598 struct vmcore *m;
1599
1600 m = list_first_entry(&vmcore_list, struct vmcore, list);
1601 list_del(&m->list);
1602 kfree(m);
1603 }
1604 free_elfcorebuf();
1605
1606 /* clear vmcore device dump list */
1607 vmcore_free_device_dumps();
1608 }
1609