xref: /linux/fs/proc/vmcore.c (revision 30e48a75df9c6ead93866bdf1511ca6ecfe17fbe)
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 
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 
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 
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 
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. */
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  */
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  */
191 void __weak elfcorehdr_free(unsigned long long addr)
192 {}
193 
194 /*
195  * Architectures may override this function to read from ELF header
196  */
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  */
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  */
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  */
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
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
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  */
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 
409 	return acc;
410 }
411 
412 static ssize_t read_vmcore(struct kiocb *iocb, struct iov_iter *iter)
413 {
414 	return __read_vmcore(iter, &iocb->ki_pos);
415 }
416 
417 /*
418  * The vmcore fault handler uses the page cache and fills data using the
419  * standard __read_vmcore() function.
420  *
421  * On s390 the fault handler is used for memory regions that can't be mapped
422  * directly with remap_pfn_range().
423  */
424 static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
425 {
426 #ifdef CONFIG_S390
427 	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
428 	pgoff_t index = vmf->pgoff;
429 	struct iov_iter iter;
430 	struct kvec kvec;
431 	struct page *page;
432 	loff_t offset;
433 	int rc;
434 
435 	page = find_or_create_page(mapping, index, GFP_KERNEL);
436 	if (!page)
437 		return VM_FAULT_OOM;
438 	if (!PageUptodate(page)) {
439 		offset = (loff_t) index << PAGE_SHIFT;
440 		kvec.iov_base = page_address(page);
441 		kvec.iov_len = PAGE_SIZE;
442 		iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, PAGE_SIZE);
443 
444 		rc = __read_vmcore(&iter, &offset);
445 		if (rc < 0) {
446 			unlock_page(page);
447 			put_page(page);
448 			return vmf_error(rc);
449 		}
450 		SetPageUptodate(page);
451 	}
452 	unlock_page(page);
453 	vmf->page = page;
454 	return 0;
455 #else
456 	return VM_FAULT_SIGBUS;
457 #endif
458 }
459 
460 static const struct vm_operations_struct vmcore_mmap_ops = {
461 	.fault = mmap_vmcore_fault,
462 };
463 
464 /**
465  * vmcore_alloc_buf - allocate buffer in vmalloc memory
466  * @size: size of buffer
467  *
468  * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
469  * the buffer to user-space by means of remap_vmalloc_range().
470  *
471  * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
472  * disabled and there's no need to allow users to mmap the buffer.
473  */
474 static inline char *vmcore_alloc_buf(size_t size)
475 {
476 #ifdef CONFIG_MMU
477 	return vmalloc_user(size);
478 #else
479 	return vzalloc(size);
480 #endif
481 }
482 
483 /*
484  * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
485  * essential for mmap_vmcore() in order to map physically
486  * non-contiguous objects (ELF header, ELF note segment and memory
487  * regions in the 1st kernel pointed to by PT_LOAD entries) into
488  * virtually contiguous user-space in ELF layout.
489  */
490 #ifdef CONFIG_MMU
491 /*
492  * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
493  * reported as not being ram with the zero page.
494  *
495  * @vma: vm_area_struct describing requested mapping
496  * @from: start remapping from
497  * @pfn: page frame number to start remapping to
498  * @size: remapping size
499  * @prot: protection bits
500  *
501  * Returns zero on success, -EAGAIN on failure.
502  */
503 static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
504 				    unsigned long from, unsigned long pfn,
505 				    unsigned long size, pgprot_t prot)
506 {
507 	unsigned long map_size;
508 	unsigned long pos_start, pos_end, pos;
509 	unsigned long zeropage_pfn = my_zero_pfn(0);
510 	size_t len = 0;
511 
512 	pos_start = pfn;
513 	pos_end = pfn + (size >> PAGE_SHIFT);
514 
515 	for (pos = pos_start; pos < pos_end; ++pos) {
516 		if (!pfn_is_ram(pos)) {
517 			/*
518 			 * We hit a page which is not ram. Remap the continuous
519 			 * region between pos_start and pos-1 and replace
520 			 * the non-ram page at pos with the zero page.
521 			 */
522 			if (pos > pos_start) {
523 				/* Remap continuous region */
524 				map_size = (pos - pos_start) << PAGE_SHIFT;
525 				if (remap_oldmem_pfn_range(vma, from + len,
526 							   pos_start, map_size,
527 							   prot))
528 					goto fail;
529 				len += map_size;
530 			}
531 			/* Remap the zero page */
532 			if (remap_oldmem_pfn_range(vma, from + len,
533 						   zeropage_pfn,
534 						   PAGE_SIZE, prot))
535 				goto fail;
536 			len += PAGE_SIZE;
537 			pos_start = pos + 1;
538 		}
539 	}
540 	if (pos > pos_start) {
541 		/* Remap the rest */
542 		map_size = (pos - pos_start) << PAGE_SHIFT;
543 		if (remap_oldmem_pfn_range(vma, from + len, pos_start,
544 					   map_size, prot))
545 			goto fail;
546 	}
547 	return 0;
548 fail:
549 	do_munmap(vma->vm_mm, from, len, NULL);
550 	return -EAGAIN;
551 }
552 
553 static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
554 			    unsigned long from, unsigned long pfn,
555 			    unsigned long size, pgprot_t prot)
556 {
557 	int ret, idx;
558 
559 	/*
560 	 * Check if a callback was registered to avoid looping over all
561 	 * pages without a reason.
562 	 */
563 	idx = srcu_read_lock(&vmcore_cb_srcu);
564 	if (!list_empty(&vmcore_cb_list))
565 		ret = remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
566 	else
567 		ret = remap_oldmem_pfn_range(vma, from, pfn, size, prot);
568 	srcu_read_unlock(&vmcore_cb_srcu, idx);
569 	return ret;
570 }
571 
572 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
573 {
574 	size_t size = vma->vm_end - vma->vm_start;
575 	u64 start, end, len, tsz;
576 	struct vmcore *m;
577 
578 	start = (u64)vma->vm_pgoff << PAGE_SHIFT;
579 	end = start + size;
580 
581 	if (size > vmcore_size || end > vmcore_size)
582 		return -EINVAL;
583 
584 	if (vma->vm_flags & (VM_WRITE | VM_EXEC))
585 		return -EPERM;
586 
587 	vm_flags_mod(vma, VM_MIXEDMAP, VM_MAYWRITE | VM_MAYEXEC);
588 	vma->vm_ops = &vmcore_mmap_ops;
589 
590 	len = 0;
591 
592 	if (start < elfcorebuf_sz) {
593 		u64 pfn;
594 
595 		tsz = min(elfcorebuf_sz - (size_t)start, size);
596 		pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
597 		if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
598 				    vma->vm_page_prot))
599 			return -EAGAIN;
600 		size -= tsz;
601 		start += tsz;
602 		len += tsz;
603 
604 		if (size == 0)
605 			return 0;
606 	}
607 
608 	if (start < elfcorebuf_sz + elfnotes_sz) {
609 		void *kaddr;
610 
611 		/* We add device dumps before other elf notes because the
612 		 * other elf notes may not fill the elf notes buffer
613 		 * completely and we will end up with zero-filled data
614 		 * between the elf notes and the device dumps. Tools will
615 		 * then try to decode this zero-filled data as valid notes
616 		 * and we don't want that. Hence, adding device dumps before
617 		 * the other elf notes ensure that zero-filled data can be
618 		 * avoided. This also ensures that the device dumps and
619 		 * other elf notes can be properly mmaped at page aligned
620 		 * address.
621 		 */
622 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
623 		/* Read device dumps */
624 		if (start < elfcorebuf_sz + vmcoredd_orig_sz) {
625 			u64 start_off;
626 
627 			tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
628 				  (size_t)start, size);
629 			start_off = start - elfcorebuf_sz;
630 			if (vmcoredd_mmap_dumps(vma, vma->vm_start + len,
631 						start_off, tsz))
632 				goto fail;
633 
634 			size -= tsz;
635 			start += tsz;
636 			len += tsz;
637 
638 			/* leave now if filled buffer already */
639 			if (!size)
640 				return 0;
641 		}
642 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
643 
644 		/* Read remaining elf notes */
645 		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
646 		kaddr = elfnotes_buf + start - elfcorebuf_sz - vmcoredd_orig_sz;
647 		if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
648 						kaddr, 0, tsz))
649 			goto fail;
650 
651 		size -= tsz;
652 		start += tsz;
653 		len += tsz;
654 
655 		if (size == 0)
656 			return 0;
657 	}
658 
659 	list_for_each_entry(m, &vmcore_list, list) {
660 		if (start < m->offset + m->size) {
661 			u64 paddr = 0;
662 
663 			tsz = (size_t)min_t(unsigned long long,
664 					    m->offset + m->size - start, size);
665 			paddr = m->paddr + start - m->offset;
666 			if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
667 						    paddr >> PAGE_SHIFT, tsz,
668 						    vma->vm_page_prot))
669 				goto fail;
670 			size -= tsz;
671 			start += tsz;
672 			len += tsz;
673 
674 			if (size == 0)
675 				return 0;
676 		}
677 	}
678 
679 	return 0;
680 fail:
681 	do_munmap(vma->vm_mm, vma->vm_start, len, NULL);
682 	return -EAGAIN;
683 }
684 #else
685 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
686 {
687 	return -ENOSYS;
688 }
689 #endif
690 
691 static const struct proc_ops vmcore_proc_ops = {
692 	.proc_open	= open_vmcore,
693 	.proc_read_iter	= read_vmcore,
694 	.proc_lseek	= default_llseek,
695 	.proc_mmap	= mmap_vmcore,
696 };
697 
698 static struct vmcore* __init get_new_element(void)
699 {
700 	return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
701 }
702 
703 static u64 get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
704 			   struct list_head *vc_list)
705 {
706 	u64 size;
707 	struct vmcore *m;
708 
709 	size = elfsz + elfnotesegsz;
710 	list_for_each_entry(m, vc_list, list) {
711 		size += m->size;
712 	}
713 	return size;
714 }
715 
716 /**
717  * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
718  *
719  * @ehdr_ptr: ELF header
720  *
721  * This function updates p_memsz member of each PT_NOTE entry in the
722  * program header table pointed to by @ehdr_ptr to real size of ELF
723  * note segment.
724  */
725 static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
726 {
727 	int i, rc=0;
728 	Elf64_Phdr *phdr_ptr;
729 	Elf64_Nhdr *nhdr_ptr;
730 
731 	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
732 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
733 		void *notes_section;
734 		u64 offset, max_sz, sz, real_sz = 0;
735 		if (phdr_ptr->p_type != PT_NOTE)
736 			continue;
737 		max_sz = phdr_ptr->p_memsz;
738 		offset = phdr_ptr->p_offset;
739 		notes_section = kmalloc(max_sz, GFP_KERNEL);
740 		if (!notes_section)
741 			return -ENOMEM;
742 		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
743 		if (rc < 0) {
744 			kfree(notes_section);
745 			return rc;
746 		}
747 		nhdr_ptr = notes_section;
748 		while (nhdr_ptr->n_namesz != 0) {
749 			sz = sizeof(Elf64_Nhdr) +
750 				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
751 				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
752 			if ((real_sz + sz) > max_sz) {
753 				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
754 					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
755 				break;
756 			}
757 			real_sz += sz;
758 			nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
759 		}
760 		kfree(notes_section);
761 		phdr_ptr->p_memsz = real_sz;
762 		if (real_sz == 0) {
763 			pr_warn("Warning: Zero PT_NOTE entries found\n");
764 		}
765 	}
766 
767 	return 0;
768 }
769 
770 /**
771  * get_note_number_and_size_elf64 - get the number of PT_NOTE program
772  * headers and sum of real size of their ELF note segment headers and
773  * data.
774  *
775  * @ehdr_ptr: ELF header
776  * @nr_ptnote: buffer for the number of PT_NOTE program headers
777  * @sz_ptnote: buffer for size of unique PT_NOTE program header
778  *
779  * This function is used to merge multiple PT_NOTE program headers
780  * into a unique single one. The resulting unique entry will have
781  * @sz_ptnote in its phdr->p_mem.
782  *
783  * It is assumed that program headers with PT_NOTE type pointed to by
784  * @ehdr_ptr has already been updated by update_note_header_size_elf64
785  * and each of PT_NOTE program headers has actual ELF note segment
786  * size in its p_memsz member.
787  */
788 static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
789 						 int *nr_ptnote, u64 *sz_ptnote)
790 {
791 	int i;
792 	Elf64_Phdr *phdr_ptr;
793 
794 	*nr_ptnote = *sz_ptnote = 0;
795 
796 	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
797 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
798 		if (phdr_ptr->p_type != PT_NOTE)
799 			continue;
800 		*nr_ptnote += 1;
801 		*sz_ptnote += phdr_ptr->p_memsz;
802 	}
803 
804 	return 0;
805 }
806 
807 /**
808  * copy_notes_elf64 - copy ELF note segments in a given buffer
809  *
810  * @ehdr_ptr: ELF header
811  * @notes_buf: buffer into which ELF note segments are copied
812  *
813  * This function is used to copy ELF note segment in the 1st kernel
814  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
815  * size of the buffer @notes_buf is equal to or larger than sum of the
816  * real ELF note segment headers and data.
817  *
818  * It is assumed that program headers with PT_NOTE type pointed to by
819  * @ehdr_ptr has already been updated by update_note_header_size_elf64
820  * and each of PT_NOTE program headers has actual ELF note segment
821  * size in its p_memsz member.
822  */
823 static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
824 {
825 	int i, rc=0;
826 	Elf64_Phdr *phdr_ptr;
827 
828 	phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
829 
830 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
831 		u64 offset;
832 		if (phdr_ptr->p_type != PT_NOTE)
833 			continue;
834 		offset = phdr_ptr->p_offset;
835 		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
836 					   &offset);
837 		if (rc < 0)
838 			return rc;
839 		notes_buf += phdr_ptr->p_memsz;
840 	}
841 
842 	return 0;
843 }
844 
845 /* Merges all the PT_NOTE headers into one. */
846 static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
847 					   char **notes_buf, size_t *notes_sz)
848 {
849 	int i, nr_ptnote=0, rc=0;
850 	char *tmp;
851 	Elf64_Ehdr *ehdr_ptr;
852 	Elf64_Phdr phdr;
853 	u64 phdr_sz = 0, note_off;
854 
855 	ehdr_ptr = (Elf64_Ehdr *)elfptr;
856 
857 	rc = update_note_header_size_elf64(ehdr_ptr);
858 	if (rc < 0)
859 		return rc;
860 
861 	rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
862 	if (rc < 0)
863 		return rc;
864 
865 	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
866 	*notes_buf = vmcore_alloc_buf(*notes_sz);
867 	if (!*notes_buf)
868 		return -ENOMEM;
869 
870 	rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
871 	if (rc < 0)
872 		return rc;
873 
874 	/* Prepare merged PT_NOTE program header. */
875 	phdr.p_type    = PT_NOTE;
876 	phdr.p_flags   = 0;
877 	note_off = sizeof(Elf64_Ehdr) +
878 			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
879 	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
880 	phdr.p_vaddr   = phdr.p_paddr = 0;
881 	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
882 	phdr.p_align   = 4;
883 
884 	/* Add merged PT_NOTE program header*/
885 	tmp = elfptr + sizeof(Elf64_Ehdr);
886 	memcpy(tmp, &phdr, sizeof(phdr));
887 	tmp += sizeof(phdr);
888 
889 	/* Remove unwanted PT_NOTE program headers. */
890 	i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
891 	*elfsz = *elfsz - i;
892 	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
893 	memset(elfptr + *elfsz, 0, i);
894 	*elfsz = roundup(*elfsz, PAGE_SIZE);
895 
896 	/* Modify e_phnum to reflect merged headers. */
897 	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
898 
899 	/* Store the size of all notes.  We need this to update the note
900 	 * header when the device dumps will be added.
901 	 */
902 	elfnotes_orig_sz = phdr.p_memsz;
903 
904 	return 0;
905 }
906 
907 /**
908  * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
909  *
910  * @ehdr_ptr: ELF header
911  *
912  * This function updates p_memsz member of each PT_NOTE entry in the
913  * program header table pointed to by @ehdr_ptr to real size of ELF
914  * note segment.
915  */
916 static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
917 {
918 	int i, rc=0;
919 	Elf32_Phdr *phdr_ptr;
920 	Elf32_Nhdr *nhdr_ptr;
921 
922 	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
923 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
924 		void *notes_section;
925 		u64 offset, max_sz, sz, real_sz = 0;
926 		if (phdr_ptr->p_type != PT_NOTE)
927 			continue;
928 		max_sz = phdr_ptr->p_memsz;
929 		offset = phdr_ptr->p_offset;
930 		notes_section = kmalloc(max_sz, GFP_KERNEL);
931 		if (!notes_section)
932 			return -ENOMEM;
933 		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
934 		if (rc < 0) {
935 			kfree(notes_section);
936 			return rc;
937 		}
938 		nhdr_ptr = notes_section;
939 		while (nhdr_ptr->n_namesz != 0) {
940 			sz = sizeof(Elf32_Nhdr) +
941 				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
942 				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
943 			if ((real_sz + sz) > max_sz) {
944 				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
945 					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
946 				break;
947 			}
948 			real_sz += sz;
949 			nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
950 		}
951 		kfree(notes_section);
952 		phdr_ptr->p_memsz = real_sz;
953 		if (real_sz == 0) {
954 			pr_warn("Warning: Zero PT_NOTE entries found\n");
955 		}
956 	}
957 
958 	return 0;
959 }
960 
961 /**
962  * get_note_number_and_size_elf32 - get the number of PT_NOTE program
963  * headers and sum of real size of their ELF note segment headers and
964  * data.
965  *
966  * @ehdr_ptr: ELF header
967  * @nr_ptnote: buffer for the number of PT_NOTE program headers
968  * @sz_ptnote: buffer for size of unique PT_NOTE program header
969  *
970  * This function is used to merge multiple PT_NOTE program headers
971  * into a unique single one. The resulting unique entry will have
972  * @sz_ptnote in its phdr->p_mem.
973  *
974  * It is assumed that program headers with PT_NOTE type pointed to by
975  * @ehdr_ptr has already been updated by update_note_header_size_elf32
976  * and each of PT_NOTE program headers has actual ELF note segment
977  * size in its p_memsz member.
978  */
979 static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
980 						 int *nr_ptnote, u64 *sz_ptnote)
981 {
982 	int i;
983 	Elf32_Phdr *phdr_ptr;
984 
985 	*nr_ptnote = *sz_ptnote = 0;
986 
987 	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
988 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
989 		if (phdr_ptr->p_type != PT_NOTE)
990 			continue;
991 		*nr_ptnote += 1;
992 		*sz_ptnote += phdr_ptr->p_memsz;
993 	}
994 
995 	return 0;
996 }
997 
998 /**
999  * copy_notes_elf32 - copy ELF note segments in a given buffer
1000  *
1001  * @ehdr_ptr: ELF header
1002  * @notes_buf: buffer into which ELF note segments are copied
1003  *
1004  * This function is used to copy ELF note segment in the 1st kernel
1005  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
1006  * size of the buffer @notes_buf is equal to or larger than sum of the
1007  * real ELF note segment headers and data.
1008  *
1009  * It is assumed that program headers with PT_NOTE type pointed to by
1010  * @ehdr_ptr has already been updated by update_note_header_size_elf32
1011  * and each of PT_NOTE program headers has actual ELF note segment
1012  * size in its p_memsz member.
1013  */
1014 static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
1015 {
1016 	int i, rc=0;
1017 	Elf32_Phdr *phdr_ptr;
1018 
1019 	phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
1020 
1021 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1022 		u64 offset;
1023 		if (phdr_ptr->p_type != PT_NOTE)
1024 			continue;
1025 		offset = phdr_ptr->p_offset;
1026 		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
1027 					   &offset);
1028 		if (rc < 0)
1029 			return rc;
1030 		notes_buf += phdr_ptr->p_memsz;
1031 	}
1032 
1033 	return 0;
1034 }
1035 
1036 /* Merges all the PT_NOTE headers into one. */
1037 static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
1038 					   char **notes_buf, size_t *notes_sz)
1039 {
1040 	int i, nr_ptnote=0, rc=0;
1041 	char *tmp;
1042 	Elf32_Ehdr *ehdr_ptr;
1043 	Elf32_Phdr phdr;
1044 	u64 phdr_sz = 0, note_off;
1045 
1046 	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1047 
1048 	rc = update_note_header_size_elf32(ehdr_ptr);
1049 	if (rc < 0)
1050 		return rc;
1051 
1052 	rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
1053 	if (rc < 0)
1054 		return rc;
1055 
1056 	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
1057 	*notes_buf = vmcore_alloc_buf(*notes_sz);
1058 	if (!*notes_buf)
1059 		return -ENOMEM;
1060 
1061 	rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
1062 	if (rc < 0)
1063 		return rc;
1064 
1065 	/* Prepare merged PT_NOTE program header. */
1066 	phdr.p_type    = PT_NOTE;
1067 	phdr.p_flags   = 0;
1068 	note_off = sizeof(Elf32_Ehdr) +
1069 			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
1070 	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
1071 	phdr.p_vaddr   = phdr.p_paddr = 0;
1072 	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
1073 	phdr.p_align   = 4;
1074 
1075 	/* Add merged PT_NOTE program header*/
1076 	tmp = elfptr + sizeof(Elf32_Ehdr);
1077 	memcpy(tmp, &phdr, sizeof(phdr));
1078 	tmp += sizeof(phdr);
1079 
1080 	/* Remove unwanted PT_NOTE program headers. */
1081 	i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
1082 	*elfsz = *elfsz - i;
1083 	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
1084 	memset(elfptr + *elfsz, 0, i);
1085 	*elfsz = roundup(*elfsz, PAGE_SIZE);
1086 
1087 	/* Modify e_phnum to reflect merged headers. */
1088 	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
1089 
1090 	/* Store the size of all notes.  We need this to update the note
1091 	 * header when the device dumps will be added.
1092 	 */
1093 	elfnotes_orig_sz = phdr.p_memsz;
1094 
1095 	return 0;
1096 }
1097 
1098 /* Add memory chunks represented by program headers to vmcore list. Also update
1099  * the new offset fields of exported program headers. */
1100 static int __init process_ptload_program_headers_elf64(char *elfptr,
1101 						size_t elfsz,
1102 						size_t elfnotes_sz,
1103 						struct list_head *vc_list)
1104 {
1105 	int i;
1106 	Elf64_Ehdr *ehdr_ptr;
1107 	Elf64_Phdr *phdr_ptr;
1108 	loff_t vmcore_off;
1109 	struct vmcore *new;
1110 
1111 	ehdr_ptr = (Elf64_Ehdr *)elfptr;
1112 	phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
1113 
1114 	/* Skip ELF header, program headers and ELF note segment. */
1115 	vmcore_off = elfsz + elfnotes_sz;
1116 
1117 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1118 		u64 paddr, start, end, size;
1119 
1120 		if (phdr_ptr->p_type != PT_LOAD)
1121 			continue;
1122 
1123 		paddr = phdr_ptr->p_offset;
1124 		start = rounddown(paddr, PAGE_SIZE);
1125 		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1126 		size = end - start;
1127 
1128 		/* Add this contiguous chunk of memory to vmcore list.*/
1129 		new = get_new_element();
1130 		if (!new)
1131 			return -ENOMEM;
1132 		new->paddr = start;
1133 		new->size = size;
1134 		list_add_tail(&new->list, vc_list);
1135 
1136 		/* Update the program header offset. */
1137 		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1138 		vmcore_off = vmcore_off + size;
1139 	}
1140 	return 0;
1141 }
1142 
1143 static int __init process_ptload_program_headers_elf32(char *elfptr,
1144 						size_t elfsz,
1145 						size_t elfnotes_sz,
1146 						struct list_head *vc_list)
1147 {
1148 	int i;
1149 	Elf32_Ehdr *ehdr_ptr;
1150 	Elf32_Phdr *phdr_ptr;
1151 	loff_t vmcore_off;
1152 	struct vmcore *new;
1153 
1154 	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1155 	phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
1156 
1157 	/* Skip ELF header, program headers and ELF note segment. */
1158 	vmcore_off = elfsz + elfnotes_sz;
1159 
1160 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1161 		u64 paddr, start, end, size;
1162 
1163 		if (phdr_ptr->p_type != PT_LOAD)
1164 			continue;
1165 
1166 		paddr = phdr_ptr->p_offset;
1167 		start = rounddown(paddr, PAGE_SIZE);
1168 		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1169 		size = end - start;
1170 
1171 		/* Add this contiguous chunk of memory to vmcore list.*/
1172 		new = get_new_element();
1173 		if (!new)
1174 			return -ENOMEM;
1175 		new->paddr = start;
1176 		new->size = size;
1177 		list_add_tail(&new->list, vc_list);
1178 
1179 		/* Update the program header offset */
1180 		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1181 		vmcore_off = vmcore_off + size;
1182 	}
1183 	return 0;
1184 }
1185 
1186 /* Sets offset fields of vmcore elements. */
1187 static void set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
1188 				    struct list_head *vc_list)
1189 {
1190 	loff_t vmcore_off;
1191 	struct vmcore *m;
1192 
1193 	/* Skip ELF header, program headers and ELF note segment. */
1194 	vmcore_off = elfsz + elfnotes_sz;
1195 
1196 	list_for_each_entry(m, vc_list, list) {
1197 		m->offset = vmcore_off;
1198 		vmcore_off += m->size;
1199 	}
1200 }
1201 
1202 static void free_elfcorebuf(void)
1203 {
1204 	free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
1205 	elfcorebuf = NULL;
1206 	vfree(elfnotes_buf);
1207 	elfnotes_buf = NULL;
1208 }
1209 
1210 static int __init parse_crash_elf64_headers(void)
1211 {
1212 	int rc=0;
1213 	Elf64_Ehdr ehdr;
1214 	u64 addr;
1215 
1216 	addr = elfcorehdr_addr;
1217 
1218 	/* Read ELF header */
1219 	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
1220 	if (rc < 0)
1221 		return rc;
1222 
1223 	/* Do some basic Verification. */
1224 	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1225 		(ehdr.e_type != ET_CORE) ||
1226 		!vmcore_elf64_check_arch(&ehdr) ||
1227 		ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
1228 		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1229 		ehdr.e_version != EV_CURRENT ||
1230 		ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
1231 		ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
1232 		ehdr.e_phnum == 0) {
1233 		pr_warn("Warning: Core image elf header is not sane\n");
1234 		return -EINVAL;
1235 	}
1236 
1237 	/* Read in all elf headers. */
1238 	elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
1239 				ehdr.e_phnum * sizeof(Elf64_Phdr);
1240 	elfcorebuf_sz = elfcorebuf_sz_orig;
1241 	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1242 					      get_order(elfcorebuf_sz_orig));
1243 	if (!elfcorebuf)
1244 		return -ENOMEM;
1245 	addr = elfcorehdr_addr;
1246 	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1247 	if (rc < 0)
1248 		goto fail;
1249 
1250 	/* Merge all PT_NOTE headers into one. */
1251 	rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
1252 				      &elfnotes_buf, &elfnotes_sz);
1253 	if (rc)
1254 		goto fail;
1255 	rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
1256 						  elfnotes_sz, &vmcore_list);
1257 	if (rc)
1258 		goto fail;
1259 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1260 	return 0;
1261 fail:
1262 	free_elfcorebuf();
1263 	return rc;
1264 }
1265 
1266 static int __init parse_crash_elf32_headers(void)
1267 {
1268 	int rc=0;
1269 	Elf32_Ehdr ehdr;
1270 	u64 addr;
1271 
1272 	addr = elfcorehdr_addr;
1273 
1274 	/* Read ELF header */
1275 	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
1276 	if (rc < 0)
1277 		return rc;
1278 
1279 	/* Do some basic Verification. */
1280 	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1281 		(ehdr.e_type != ET_CORE) ||
1282 		!vmcore_elf32_check_arch(&ehdr) ||
1283 		ehdr.e_ident[EI_CLASS] != ELFCLASS32||
1284 		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1285 		ehdr.e_version != EV_CURRENT ||
1286 		ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1287 		ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1288 		ehdr.e_phnum == 0) {
1289 		pr_warn("Warning: Core image elf header is not sane\n");
1290 		return -EINVAL;
1291 	}
1292 
1293 	/* Read in all elf headers. */
1294 	elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1295 	elfcorebuf_sz = elfcorebuf_sz_orig;
1296 	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1297 					      get_order(elfcorebuf_sz_orig));
1298 	if (!elfcorebuf)
1299 		return -ENOMEM;
1300 	addr = elfcorehdr_addr;
1301 	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1302 	if (rc < 0)
1303 		goto fail;
1304 
1305 	/* Merge all PT_NOTE headers into one. */
1306 	rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1307 				      &elfnotes_buf, &elfnotes_sz);
1308 	if (rc)
1309 		goto fail;
1310 	rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1311 						  elfnotes_sz, &vmcore_list);
1312 	if (rc)
1313 		goto fail;
1314 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1315 	return 0;
1316 fail:
1317 	free_elfcorebuf();
1318 	return rc;
1319 }
1320 
1321 static int __init parse_crash_elf_headers(void)
1322 {
1323 	unsigned char e_ident[EI_NIDENT];
1324 	u64 addr;
1325 	int rc=0;
1326 
1327 	addr = elfcorehdr_addr;
1328 	rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1329 	if (rc < 0)
1330 		return rc;
1331 	if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1332 		pr_warn("Warning: Core image elf header not found\n");
1333 		return -EINVAL;
1334 	}
1335 
1336 	if (e_ident[EI_CLASS] == ELFCLASS64) {
1337 		rc = parse_crash_elf64_headers();
1338 		if (rc)
1339 			return rc;
1340 	} else if (e_ident[EI_CLASS] == ELFCLASS32) {
1341 		rc = parse_crash_elf32_headers();
1342 		if (rc)
1343 			return rc;
1344 	} else {
1345 		pr_warn("Warning: Core image elf header is not sane\n");
1346 		return -EINVAL;
1347 	}
1348 
1349 	/* Determine vmcore size. */
1350 	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1351 				      &vmcore_list);
1352 
1353 	return 0;
1354 }
1355 
1356 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1357 /**
1358  * vmcoredd_write_header - Write vmcore device dump header at the
1359  * beginning of the dump's buffer.
1360  * @buf: Output buffer where the note is written
1361  * @data: Dump info
1362  * @size: Size of the dump
1363  *
1364  * Fills beginning of the dump's buffer with vmcore device dump header.
1365  */
1366 static void vmcoredd_write_header(void *buf, struct vmcoredd_data *data,
1367 				  u32 size)
1368 {
1369 	struct vmcoredd_header *vdd_hdr = (struct vmcoredd_header *)buf;
1370 
1371 	vdd_hdr->n_namesz = sizeof(vdd_hdr->name);
1372 	vdd_hdr->n_descsz = size + sizeof(vdd_hdr->dump_name);
1373 	vdd_hdr->n_type = NT_VMCOREDD;
1374 
1375 	strscpy_pad(vdd_hdr->name, VMCOREDD_NOTE_NAME);
1376 	strscpy_pad(vdd_hdr->dump_name, data->dump_name);
1377 }
1378 
1379 /**
1380  * vmcoredd_update_program_headers - Update all ELF program headers
1381  * @elfptr: Pointer to elf header
1382  * @elfnotesz: Size of elf notes aligned to page size
1383  * @vmcoreddsz: Size of device dumps to be added to elf note header
1384  *
1385  * Determine type of ELF header (Elf64 or Elf32) and update the elf note size.
1386  * Also update the offsets of all the program headers after the elf note header.
1387  */
1388 static void vmcoredd_update_program_headers(char *elfptr, size_t elfnotesz,
1389 					    size_t vmcoreddsz)
1390 {
1391 	unsigned char *e_ident = (unsigned char *)elfptr;
1392 	u64 start, end, size;
1393 	loff_t vmcore_off;
1394 	u32 i;
1395 
1396 	vmcore_off = elfcorebuf_sz + elfnotesz;
1397 
1398 	if (e_ident[EI_CLASS] == ELFCLASS64) {
1399 		Elf64_Ehdr *ehdr = (Elf64_Ehdr *)elfptr;
1400 		Elf64_Phdr *phdr = (Elf64_Phdr *)(elfptr + sizeof(Elf64_Ehdr));
1401 
1402 		/* Update all program headers */
1403 		for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1404 			if (phdr->p_type == PT_NOTE) {
1405 				/* Update note size */
1406 				phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1407 				phdr->p_filesz = phdr->p_memsz;
1408 				continue;
1409 			}
1410 
1411 			start = rounddown(phdr->p_offset, PAGE_SIZE);
1412 			end = roundup(phdr->p_offset + phdr->p_memsz,
1413 				      PAGE_SIZE);
1414 			size = end - start;
1415 			phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1416 			vmcore_off += size;
1417 		}
1418 	} else {
1419 		Elf32_Ehdr *ehdr = (Elf32_Ehdr *)elfptr;
1420 		Elf32_Phdr *phdr = (Elf32_Phdr *)(elfptr + sizeof(Elf32_Ehdr));
1421 
1422 		/* Update all program headers */
1423 		for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1424 			if (phdr->p_type == PT_NOTE) {
1425 				/* Update note size */
1426 				phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1427 				phdr->p_filesz = phdr->p_memsz;
1428 				continue;
1429 			}
1430 
1431 			start = rounddown(phdr->p_offset, PAGE_SIZE);
1432 			end = roundup(phdr->p_offset + phdr->p_memsz,
1433 				      PAGE_SIZE);
1434 			size = end - start;
1435 			phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1436 			vmcore_off += size;
1437 		}
1438 	}
1439 }
1440 
1441 /**
1442  * vmcoredd_update_size - Update the total size of the device dumps and update
1443  * ELF header
1444  * @dump_size: Size of the current device dump to be added to total size
1445  *
1446  * Update the total size of all the device dumps and update the ELF program
1447  * headers. Calculate the new offsets for the vmcore list and update the
1448  * total vmcore size.
1449  */
1450 static void vmcoredd_update_size(size_t dump_size)
1451 {
1452 	vmcoredd_orig_sz += dump_size;
1453 	elfnotes_sz = roundup(elfnotes_orig_sz, PAGE_SIZE) + vmcoredd_orig_sz;
1454 	vmcoredd_update_program_headers(elfcorebuf, elfnotes_sz,
1455 					vmcoredd_orig_sz);
1456 
1457 	/* Update vmcore list offsets */
1458 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1459 
1460 	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1461 				      &vmcore_list);
1462 	proc_vmcore->size = vmcore_size;
1463 }
1464 
1465 /**
1466  * vmcore_add_device_dump - Add a buffer containing device dump to vmcore
1467  * @data: dump info.
1468  *
1469  * Allocate a buffer and invoke the calling driver's dump collect routine.
1470  * Write ELF note at the beginning of the buffer to indicate vmcore device
1471  * dump and add the dump to global list.
1472  */
1473 int vmcore_add_device_dump(struct vmcoredd_data *data)
1474 {
1475 	struct vmcoredd_node *dump;
1476 	void *buf = NULL;
1477 	size_t data_size;
1478 	int ret;
1479 
1480 	if (vmcoredd_disabled) {
1481 		pr_err_once("Device dump is disabled\n");
1482 		return -EINVAL;
1483 	}
1484 
1485 	if (!data || !strlen(data->dump_name) ||
1486 	    !data->vmcoredd_callback || !data->size)
1487 		return -EINVAL;
1488 
1489 	dump = vzalloc(sizeof(*dump));
1490 	if (!dump) {
1491 		ret = -ENOMEM;
1492 		goto out_err;
1493 	}
1494 
1495 	/* Keep size of the buffer page aligned so that it can be mmaped */
1496 	data_size = roundup(sizeof(struct vmcoredd_header) + data->size,
1497 			    PAGE_SIZE);
1498 
1499 	/* Allocate buffer for driver's to write their dumps */
1500 	buf = vmcore_alloc_buf(data_size);
1501 	if (!buf) {
1502 		ret = -ENOMEM;
1503 		goto out_err;
1504 	}
1505 
1506 	vmcoredd_write_header(buf, data, data_size -
1507 			      sizeof(struct vmcoredd_header));
1508 
1509 	/* Invoke the driver's dump collection routing */
1510 	ret = data->vmcoredd_callback(data, buf +
1511 				      sizeof(struct vmcoredd_header));
1512 	if (ret)
1513 		goto out_err;
1514 
1515 	dump->buf = buf;
1516 	dump->size = data_size;
1517 
1518 	/* Add the dump to driver sysfs list */
1519 	mutex_lock(&vmcoredd_mutex);
1520 	list_add_tail(&dump->list, &vmcoredd_list);
1521 	mutex_unlock(&vmcoredd_mutex);
1522 
1523 	vmcoredd_update_size(data_size);
1524 	return 0;
1525 
1526 out_err:
1527 	vfree(buf);
1528 	vfree(dump);
1529 
1530 	return ret;
1531 }
1532 EXPORT_SYMBOL(vmcore_add_device_dump);
1533 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1534 
1535 /* Free all dumps in vmcore device dump list */
1536 static void vmcore_free_device_dumps(void)
1537 {
1538 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1539 	mutex_lock(&vmcoredd_mutex);
1540 	while (!list_empty(&vmcoredd_list)) {
1541 		struct vmcoredd_node *dump;
1542 
1543 		dump = list_first_entry(&vmcoredd_list, struct vmcoredd_node,
1544 					list);
1545 		list_del(&dump->list);
1546 		vfree(dump->buf);
1547 		vfree(dump);
1548 	}
1549 	mutex_unlock(&vmcoredd_mutex);
1550 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1551 }
1552 
1553 /* Init function for vmcore module. */
1554 static int __init vmcore_init(void)
1555 {
1556 	int rc = 0;
1557 
1558 	/* Allow architectures to allocate ELF header in 2nd kernel */
1559 	rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1560 	if (rc)
1561 		return rc;
1562 	/*
1563 	 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1564 	 * then capture the dump.
1565 	 */
1566 	if (!(is_vmcore_usable()))
1567 		return rc;
1568 	rc = parse_crash_elf_headers();
1569 	if (rc) {
1570 		elfcorehdr_free(elfcorehdr_addr);
1571 		pr_warn("Kdump: vmcore not initialized\n");
1572 		return rc;
1573 	}
1574 	elfcorehdr_free(elfcorehdr_addr);
1575 	elfcorehdr_addr = ELFCORE_ADDR_ERR;
1576 
1577 	proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &vmcore_proc_ops);
1578 	if (proc_vmcore)
1579 		proc_vmcore->size = vmcore_size;
1580 	return 0;
1581 }
1582 fs_initcall(vmcore_init);
1583 
1584 /* Cleanup function for vmcore module. */
1585 void vmcore_cleanup(void)
1586 {
1587 	if (proc_vmcore) {
1588 		proc_remove(proc_vmcore);
1589 		proc_vmcore = NULL;
1590 	}
1591 
1592 	/* clear the vmcore list. */
1593 	while (!list_empty(&vmcore_list)) {
1594 		struct vmcore *m;
1595 
1596 		m = list_first_entry(&vmcore_list, struct vmcore, list);
1597 		list_del(&m->list);
1598 		kfree(m);
1599 	}
1600 	free_elfcorebuf();
1601 
1602 	/* clear vmcore device dump list */
1603 	vmcore_free_device_dumps();
1604 }
1605