xref: /openbmc/linux/fs/proc/vmcore.c (revision 78bb17f7)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *	fs/proc/vmcore.c Interface for accessing the crash
4  * 				 dump from the system's previous life.
5  * 	Heavily borrowed from fs/proc/kcore.c
6  *	Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
7  *	Copyright (C) IBM Corporation, 2004. All rights reserved
8  *
9  */
10 
11 #include <linux/mm.h>
12 #include <linux/kcore.h>
13 #include <linux/user.h>
14 #include <linux/elf.h>
15 #include <linux/elfcore.h>
16 #include <linux/export.h>
17 #include <linux/slab.h>
18 #include <linux/highmem.h>
19 #include <linux/printk.h>
20 #include <linux/memblock.h>
21 #include <linux/init.h>
22 #include <linux/crash_dump.h>
23 #include <linux/list.h>
24 #include <linux/moduleparam.h>
25 #include <linux/mutex.h>
26 #include <linux/vmalloc.h>
27 #include <linux/pagemap.h>
28 #include <linux/uaccess.h>
29 #include <linux/mem_encrypt.h>
30 #include <asm/pgtable.h>
31 #include <asm/io.h>
32 #include "internal.h"
33 
34 /* List representing chunks of contiguous memory areas and their offsets in
35  * vmcore file.
36  */
37 static LIST_HEAD(vmcore_list);
38 
39 /* Stores the pointer to the buffer containing kernel elf core headers. */
40 static char *elfcorebuf;
41 static size_t elfcorebuf_sz;
42 static size_t elfcorebuf_sz_orig;
43 
44 static char *elfnotes_buf;
45 static size_t elfnotes_sz;
46 /* Size of all notes minus the device dump notes */
47 static size_t elfnotes_orig_sz;
48 
49 /* Total size of vmcore file. */
50 static u64 vmcore_size;
51 
52 static struct proc_dir_entry *proc_vmcore;
53 
54 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
55 /* Device Dump list and mutex to synchronize access to list */
56 static LIST_HEAD(vmcoredd_list);
57 static DEFINE_MUTEX(vmcoredd_mutex);
58 
59 static bool vmcoredd_disabled;
60 core_param(novmcoredd, vmcoredd_disabled, bool, 0);
61 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
62 
63 /* Device Dump Size */
64 static size_t vmcoredd_orig_sz;
65 
66 /*
67  * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
68  * The called function has to take care of module refcounting.
69  */
70 static int (*oldmem_pfn_is_ram)(unsigned long pfn);
71 
72 int register_oldmem_pfn_is_ram(int (*fn)(unsigned long pfn))
73 {
74 	if (oldmem_pfn_is_ram)
75 		return -EBUSY;
76 	oldmem_pfn_is_ram = fn;
77 	return 0;
78 }
79 EXPORT_SYMBOL_GPL(register_oldmem_pfn_is_ram);
80 
81 void unregister_oldmem_pfn_is_ram(void)
82 {
83 	oldmem_pfn_is_ram = NULL;
84 	wmb();
85 }
86 EXPORT_SYMBOL_GPL(unregister_oldmem_pfn_is_ram);
87 
88 static int pfn_is_ram(unsigned long pfn)
89 {
90 	int (*fn)(unsigned long pfn);
91 	/* pfn is ram unless fn() checks pagetype */
92 	int ret = 1;
93 
94 	/*
95 	 * Ask hypervisor if the pfn is really ram.
96 	 * A ballooned page contains no data and reading from such a page
97 	 * will cause high load in the hypervisor.
98 	 */
99 	fn = oldmem_pfn_is_ram;
100 	if (fn)
101 		ret = fn(pfn);
102 
103 	return ret;
104 }
105 
106 /* Reads a page from the oldmem device from given offset. */
107 ssize_t read_from_oldmem(char *buf, size_t count,
108 			 u64 *ppos, int userbuf,
109 			 bool encrypted)
110 {
111 	unsigned long pfn, offset;
112 	size_t nr_bytes;
113 	ssize_t read = 0, tmp;
114 
115 	if (!count)
116 		return 0;
117 
118 	offset = (unsigned long)(*ppos % PAGE_SIZE);
119 	pfn = (unsigned long)(*ppos / PAGE_SIZE);
120 
121 	do {
122 		if (count > (PAGE_SIZE - offset))
123 			nr_bytes = PAGE_SIZE - offset;
124 		else
125 			nr_bytes = count;
126 
127 		/* If pfn is not ram, return zeros for sparse dump files */
128 		if (pfn_is_ram(pfn) == 0)
129 			memset(buf, 0, nr_bytes);
130 		else {
131 			if (encrypted)
132 				tmp = copy_oldmem_page_encrypted(pfn, buf,
133 								 nr_bytes,
134 								 offset,
135 								 userbuf);
136 			else
137 				tmp = copy_oldmem_page(pfn, buf, nr_bytes,
138 						       offset, userbuf);
139 
140 			if (tmp < 0)
141 				return tmp;
142 		}
143 		*ppos += nr_bytes;
144 		count -= nr_bytes;
145 		buf += nr_bytes;
146 		read += nr_bytes;
147 		++pfn;
148 		offset = 0;
149 	} while (count);
150 
151 	return read;
152 }
153 
154 /*
155  * Architectures may override this function to allocate ELF header in 2nd kernel
156  */
157 int __weak elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
158 {
159 	return 0;
160 }
161 
162 /*
163  * Architectures may override this function to free header
164  */
165 void __weak elfcorehdr_free(unsigned long long addr)
166 {}
167 
168 /*
169  * Architectures may override this function to read from ELF header
170  */
171 ssize_t __weak elfcorehdr_read(char *buf, size_t count, u64 *ppos)
172 {
173 	return read_from_oldmem(buf, count, ppos, 0, false);
174 }
175 
176 /*
177  * Architectures may override this function to read from notes sections
178  */
179 ssize_t __weak elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
180 {
181 	return read_from_oldmem(buf, count, ppos, 0, mem_encrypt_active());
182 }
183 
184 /*
185  * Architectures may override this function to map oldmem
186  */
187 int __weak remap_oldmem_pfn_range(struct vm_area_struct *vma,
188 				  unsigned long from, unsigned long pfn,
189 				  unsigned long size, pgprot_t prot)
190 {
191 	prot = pgprot_encrypted(prot);
192 	return remap_pfn_range(vma, from, pfn, size, prot);
193 }
194 
195 /*
196  * Architectures which support memory encryption override this.
197  */
198 ssize_t __weak
199 copy_oldmem_page_encrypted(unsigned long pfn, char *buf, size_t csize,
200 			   unsigned long offset, int userbuf)
201 {
202 	return copy_oldmem_page(pfn, buf, csize, offset, userbuf);
203 }
204 
205 /*
206  * Copy to either kernel or user space
207  */
208 static int copy_to(void *target, void *src, size_t size, int userbuf)
209 {
210 	if (userbuf) {
211 		if (copy_to_user((char __user *) target, src, size))
212 			return -EFAULT;
213 	} else {
214 		memcpy(target, src, size);
215 	}
216 	return 0;
217 }
218 
219 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
220 static int vmcoredd_copy_dumps(void *dst, u64 start, size_t size, int userbuf)
221 {
222 	struct vmcoredd_node *dump;
223 	u64 offset = 0;
224 	int ret = 0;
225 	size_t tsz;
226 	char *buf;
227 
228 	mutex_lock(&vmcoredd_mutex);
229 	list_for_each_entry(dump, &vmcoredd_list, list) {
230 		if (start < offset + dump->size) {
231 			tsz = min(offset + (u64)dump->size - start, (u64)size);
232 			buf = dump->buf + start - offset;
233 			if (copy_to(dst, buf, tsz, userbuf)) {
234 				ret = -EFAULT;
235 				goto out_unlock;
236 			}
237 
238 			size -= tsz;
239 			start += tsz;
240 			dst += tsz;
241 
242 			/* Leave now if buffer filled already */
243 			if (!size)
244 				goto out_unlock;
245 		}
246 		offset += dump->size;
247 	}
248 
249 out_unlock:
250 	mutex_unlock(&vmcoredd_mutex);
251 	return ret;
252 }
253 
254 #ifdef CONFIG_MMU
255 static int vmcoredd_mmap_dumps(struct vm_area_struct *vma, unsigned long dst,
256 			       u64 start, size_t size)
257 {
258 	struct vmcoredd_node *dump;
259 	u64 offset = 0;
260 	int ret = 0;
261 	size_t tsz;
262 	char *buf;
263 
264 	mutex_lock(&vmcoredd_mutex);
265 	list_for_each_entry(dump, &vmcoredd_list, list) {
266 		if (start < offset + dump->size) {
267 			tsz = min(offset + (u64)dump->size - start, (u64)size);
268 			buf = dump->buf + start - offset;
269 			if (remap_vmalloc_range_partial(vma, dst, buf, 0,
270 							tsz)) {
271 				ret = -EFAULT;
272 				goto out_unlock;
273 			}
274 
275 			size -= tsz;
276 			start += tsz;
277 			dst += tsz;
278 
279 			/* Leave now if buffer filled already */
280 			if (!size)
281 				goto out_unlock;
282 		}
283 		offset += dump->size;
284 	}
285 
286 out_unlock:
287 	mutex_unlock(&vmcoredd_mutex);
288 	return ret;
289 }
290 #endif /* CONFIG_MMU */
291 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
292 
293 /* Read from the ELF header and then the crash dump. On error, negative value is
294  * returned otherwise number of bytes read are returned.
295  */
296 static ssize_t __read_vmcore(char *buffer, size_t buflen, loff_t *fpos,
297 			     int userbuf)
298 {
299 	ssize_t acc = 0, tmp;
300 	size_t tsz;
301 	u64 start;
302 	struct vmcore *m = NULL;
303 
304 	if (buflen == 0 || *fpos >= vmcore_size)
305 		return 0;
306 
307 	/* trim buflen to not go beyond EOF */
308 	if (buflen > vmcore_size - *fpos)
309 		buflen = vmcore_size - *fpos;
310 
311 	/* Read ELF core header */
312 	if (*fpos < elfcorebuf_sz) {
313 		tsz = min(elfcorebuf_sz - (size_t)*fpos, buflen);
314 		if (copy_to(buffer, elfcorebuf + *fpos, tsz, userbuf))
315 			return -EFAULT;
316 		buflen -= tsz;
317 		*fpos += tsz;
318 		buffer += tsz;
319 		acc += tsz;
320 
321 		/* leave now if filled buffer already */
322 		if (buflen == 0)
323 			return acc;
324 	}
325 
326 	/* Read Elf note segment */
327 	if (*fpos < elfcorebuf_sz + elfnotes_sz) {
328 		void *kaddr;
329 
330 		/* We add device dumps before other elf notes because the
331 		 * other elf notes may not fill the elf notes buffer
332 		 * completely and we will end up with zero-filled data
333 		 * between the elf notes and the device dumps. Tools will
334 		 * then try to decode this zero-filled data as valid notes
335 		 * and we don't want that. Hence, adding device dumps before
336 		 * the other elf notes ensure that zero-filled data can be
337 		 * avoided.
338 		 */
339 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
340 		/* Read device dumps */
341 		if (*fpos < elfcorebuf_sz + vmcoredd_orig_sz) {
342 			tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
343 				  (size_t)*fpos, buflen);
344 			start = *fpos - elfcorebuf_sz;
345 			if (vmcoredd_copy_dumps(buffer, start, tsz, userbuf))
346 				return -EFAULT;
347 
348 			buflen -= tsz;
349 			*fpos += tsz;
350 			buffer += tsz;
351 			acc += tsz;
352 
353 			/* leave now if filled buffer already */
354 			if (!buflen)
355 				return acc;
356 		}
357 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
358 
359 		/* Read remaining elf notes */
360 		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)*fpos, buflen);
361 		kaddr = elfnotes_buf + *fpos - elfcorebuf_sz - vmcoredd_orig_sz;
362 		if (copy_to(buffer, kaddr, tsz, userbuf))
363 			return -EFAULT;
364 
365 		buflen -= tsz;
366 		*fpos += tsz;
367 		buffer += tsz;
368 		acc += tsz;
369 
370 		/* leave now if filled buffer already */
371 		if (buflen == 0)
372 			return acc;
373 	}
374 
375 	list_for_each_entry(m, &vmcore_list, list) {
376 		if (*fpos < m->offset + m->size) {
377 			tsz = (size_t)min_t(unsigned long long,
378 					    m->offset + m->size - *fpos,
379 					    buflen);
380 			start = m->paddr + *fpos - m->offset;
381 			tmp = read_from_oldmem(buffer, tsz, &start,
382 					       userbuf, mem_encrypt_active());
383 			if (tmp < 0)
384 				return tmp;
385 			buflen -= tsz;
386 			*fpos += tsz;
387 			buffer += tsz;
388 			acc += tsz;
389 
390 			/* leave now if filled buffer already */
391 			if (buflen == 0)
392 				return acc;
393 		}
394 	}
395 
396 	return acc;
397 }
398 
399 static ssize_t read_vmcore(struct file *file, char __user *buffer,
400 			   size_t buflen, loff_t *fpos)
401 {
402 	return __read_vmcore((__force char *) buffer, buflen, fpos, 1);
403 }
404 
405 /*
406  * The vmcore fault handler uses the page cache and fills data using the
407  * standard __vmcore_read() function.
408  *
409  * On s390 the fault handler is used for memory regions that can't be mapped
410  * directly with remap_pfn_range().
411  */
412 static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
413 {
414 #ifdef CONFIG_S390
415 	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
416 	pgoff_t index = vmf->pgoff;
417 	struct page *page;
418 	loff_t offset;
419 	char *buf;
420 	int rc;
421 
422 	page = find_or_create_page(mapping, index, GFP_KERNEL);
423 	if (!page)
424 		return VM_FAULT_OOM;
425 	if (!PageUptodate(page)) {
426 		offset = (loff_t) index << PAGE_SHIFT;
427 		buf = __va((page_to_pfn(page) << PAGE_SHIFT));
428 		rc = __read_vmcore(buf, PAGE_SIZE, &offset, 0);
429 		if (rc < 0) {
430 			unlock_page(page);
431 			put_page(page);
432 			return vmf_error(rc);
433 		}
434 		SetPageUptodate(page);
435 	}
436 	unlock_page(page);
437 	vmf->page = page;
438 	return 0;
439 #else
440 	return VM_FAULT_SIGBUS;
441 #endif
442 }
443 
444 static const struct vm_operations_struct vmcore_mmap_ops = {
445 	.fault = mmap_vmcore_fault,
446 };
447 
448 /**
449  * vmcore_alloc_buf - allocate buffer in vmalloc memory
450  * @sizez: size of buffer
451  *
452  * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
453  * the buffer to user-space by means of remap_vmalloc_range().
454  *
455  * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
456  * disabled and there's no need to allow users to mmap the buffer.
457  */
458 static inline char *vmcore_alloc_buf(size_t size)
459 {
460 #ifdef CONFIG_MMU
461 	return vmalloc_user(size);
462 #else
463 	return vzalloc(size);
464 #endif
465 }
466 
467 /*
468  * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
469  * essential for mmap_vmcore() in order to map physically
470  * non-contiguous objects (ELF header, ELF note segment and memory
471  * regions in the 1st kernel pointed to by PT_LOAD entries) into
472  * virtually contiguous user-space in ELF layout.
473  */
474 #ifdef CONFIG_MMU
475 /*
476  * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
477  * reported as not being ram with the zero page.
478  *
479  * @vma: vm_area_struct describing requested mapping
480  * @from: start remapping from
481  * @pfn: page frame number to start remapping to
482  * @size: remapping size
483  * @prot: protection bits
484  *
485  * Returns zero on success, -EAGAIN on failure.
486  */
487 static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
488 				    unsigned long from, unsigned long pfn,
489 				    unsigned long size, pgprot_t prot)
490 {
491 	unsigned long map_size;
492 	unsigned long pos_start, pos_end, pos;
493 	unsigned long zeropage_pfn = my_zero_pfn(0);
494 	size_t len = 0;
495 
496 	pos_start = pfn;
497 	pos_end = pfn + (size >> PAGE_SHIFT);
498 
499 	for (pos = pos_start; pos < pos_end; ++pos) {
500 		if (!pfn_is_ram(pos)) {
501 			/*
502 			 * We hit a page which is not ram. Remap the continuous
503 			 * region between pos_start and pos-1 and replace
504 			 * the non-ram page at pos with the zero page.
505 			 */
506 			if (pos > pos_start) {
507 				/* Remap continuous region */
508 				map_size = (pos - pos_start) << PAGE_SHIFT;
509 				if (remap_oldmem_pfn_range(vma, from + len,
510 							   pos_start, map_size,
511 							   prot))
512 					goto fail;
513 				len += map_size;
514 			}
515 			/* Remap the zero page */
516 			if (remap_oldmem_pfn_range(vma, from + len,
517 						   zeropage_pfn,
518 						   PAGE_SIZE, prot))
519 				goto fail;
520 			len += PAGE_SIZE;
521 			pos_start = pos + 1;
522 		}
523 	}
524 	if (pos > pos_start) {
525 		/* Remap the rest */
526 		map_size = (pos - pos_start) << PAGE_SHIFT;
527 		if (remap_oldmem_pfn_range(vma, from + len, pos_start,
528 					   map_size, prot))
529 			goto fail;
530 	}
531 	return 0;
532 fail:
533 	do_munmap(vma->vm_mm, from, len, NULL);
534 	return -EAGAIN;
535 }
536 
537 static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
538 			    unsigned long from, unsigned long pfn,
539 			    unsigned long size, pgprot_t prot)
540 {
541 	/*
542 	 * Check if oldmem_pfn_is_ram was registered to avoid
543 	 * looping over all pages without a reason.
544 	 */
545 	if (oldmem_pfn_is_ram)
546 		return remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
547 	else
548 		return remap_oldmem_pfn_range(vma, from, pfn, size, prot);
549 }
550 
551 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
552 {
553 	size_t size = vma->vm_end - vma->vm_start;
554 	u64 start, end, len, tsz;
555 	struct vmcore *m;
556 
557 	start = (u64)vma->vm_pgoff << PAGE_SHIFT;
558 	end = start + size;
559 
560 	if (size > vmcore_size || end > vmcore_size)
561 		return -EINVAL;
562 
563 	if (vma->vm_flags & (VM_WRITE | VM_EXEC))
564 		return -EPERM;
565 
566 	vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
567 	vma->vm_flags |= VM_MIXEDMAP;
568 	vma->vm_ops = &vmcore_mmap_ops;
569 
570 	len = 0;
571 
572 	if (start < elfcorebuf_sz) {
573 		u64 pfn;
574 
575 		tsz = min(elfcorebuf_sz - (size_t)start, size);
576 		pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
577 		if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
578 				    vma->vm_page_prot))
579 			return -EAGAIN;
580 		size -= tsz;
581 		start += tsz;
582 		len += tsz;
583 
584 		if (size == 0)
585 			return 0;
586 	}
587 
588 	if (start < elfcorebuf_sz + elfnotes_sz) {
589 		void *kaddr;
590 
591 		/* We add device dumps before other elf notes because the
592 		 * other elf notes may not fill the elf notes buffer
593 		 * completely and we will end up with zero-filled data
594 		 * between the elf notes and the device dumps. Tools will
595 		 * then try to decode this zero-filled data as valid notes
596 		 * and we don't want that. Hence, adding device dumps before
597 		 * the other elf notes ensure that zero-filled data can be
598 		 * avoided. This also ensures that the device dumps and
599 		 * other elf notes can be properly mmaped at page aligned
600 		 * address.
601 		 */
602 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
603 		/* Read device dumps */
604 		if (start < elfcorebuf_sz + vmcoredd_orig_sz) {
605 			u64 start_off;
606 
607 			tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
608 				  (size_t)start, size);
609 			start_off = start - elfcorebuf_sz;
610 			if (vmcoredd_mmap_dumps(vma, vma->vm_start + len,
611 						start_off, tsz))
612 				goto fail;
613 
614 			size -= tsz;
615 			start += tsz;
616 			len += tsz;
617 
618 			/* leave now if filled buffer already */
619 			if (!size)
620 				return 0;
621 		}
622 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
623 
624 		/* Read remaining elf notes */
625 		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
626 		kaddr = elfnotes_buf + start - elfcorebuf_sz - vmcoredd_orig_sz;
627 		if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
628 						kaddr, 0, tsz))
629 			goto fail;
630 
631 		size -= tsz;
632 		start += tsz;
633 		len += tsz;
634 
635 		if (size == 0)
636 			return 0;
637 	}
638 
639 	list_for_each_entry(m, &vmcore_list, list) {
640 		if (start < m->offset + m->size) {
641 			u64 paddr = 0;
642 
643 			tsz = (size_t)min_t(unsigned long long,
644 					    m->offset + m->size - start, size);
645 			paddr = m->paddr + start - m->offset;
646 			if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
647 						    paddr >> PAGE_SHIFT, tsz,
648 						    vma->vm_page_prot))
649 				goto fail;
650 			size -= tsz;
651 			start += tsz;
652 			len += tsz;
653 
654 			if (size == 0)
655 				return 0;
656 		}
657 	}
658 
659 	return 0;
660 fail:
661 	do_munmap(vma->vm_mm, vma->vm_start, len, NULL);
662 	return -EAGAIN;
663 }
664 #else
665 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
666 {
667 	return -ENOSYS;
668 }
669 #endif
670 
671 static const struct proc_ops vmcore_proc_ops = {
672 	.proc_read	= read_vmcore,
673 	.proc_lseek	= default_llseek,
674 	.proc_mmap	= mmap_vmcore,
675 };
676 
677 static struct vmcore* __init get_new_element(void)
678 {
679 	return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
680 }
681 
682 static u64 get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
683 			   struct list_head *vc_list)
684 {
685 	u64 size;
686 	struct vmcore *m;
687 
688 	size = elfsz + elfnotesegsz;
689 	list_for_each_entry(m, vc_list, list) {
690 		size += m->size;
691 	}
692 	return size;
693 }
694 
695 /**
696  * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
697  *
698  * @ehdr_ptr: ELF header
699  *
700  * This function updates p_memsz member of each PT_NOTE entry in the
701  * program header table pointed to by @ehdr_ptr to real size of ELF
702  * note segment.
703  */
704 static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
705 {
706 	int i, rc=0;
707 	Elf64_Phdr *phdr_ptr;
708 	Elf64_Nhdr *nhdr_ptr;
709 
710 	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
711 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
712 		void *notes_section;
713 		u64 offset, max_sz, sz, real_sz = 0;
714 		if (phdr_ptr->p_type != PT_NOTE)
715 			continue;
716 		max_sz = phdr_ptr->p_memsz;
717 		offset = phdr_ptr->p_offset;
718 		notes_section = kmalloc(max_sz, GFP_KERNEL);
719 		if (!notes_section)
720 			return -ENOMEM;
721 		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
722 		if (rc < 0) {
723 			kfree(notes_section);
724 			return rc;
725 		}
726 		nhdr_ptr = notes_section;
727 		while (nhdr_ptr->n_namesz != 0) {
728 			sz = sizeof(Elf64_Nhdr) +
729 				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
730 				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
731 			if ((real_sz + sz) > max_sz) {
732 				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
733 					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
734 				break;
735 			}
736 			real_sz += sz;
737 			nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
738 		}
739 		kfree(notes_section);
740 		phdr_ptr->p_memsz = real_sz;
741 		if (real_sz == 0) {
742 			pr_warn("Warning: Zero PT_NOTE entries found\n");
743 		}
744 	}
745 
746 	return 0;
747 }
748 
749 /**
750  * get_note_number_and_size_elf64 - get the number of PT_NOTE program
751  * headers and sum of real size of their ELF note segment headers and
752  * data.
753  *
754  * @ehdr_ptr: ELF header
755  * @nr_ptnote: buffer for the number of PT_NOTE program headers
756  * @sz_ptnote: buffer for size of unique PT_NOTE program header
757  *
758  * This function is used to merge multiple PT_NOTE program headers
759  * into a unique single one. The resulting unique entry will have
760  * @sz_ptnote in its phdr->p_mem.
761  *
762  * It is assumed that program headers with PT_NOTE type pointed to by
763  * @ehdr_ptr has already been updated by update_note_header_size_elf64
764  * and each of PT_NOTE program headers has actual ELF note segment
765  * size in its p_memsz member.
766  */
767 static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
768 						 int *nr_ptnote, u64 *sz_ptnote)
769 {
770 	int i;
771 	Elf64_Phdr *phdr_ptr;
772 
773 	*nr_ptnote = *sz_ptnote = 0;
774 
775 	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
776 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
777 		if (phdr_ptr->p_type != PT_NOTE)
778 			continue;
779 		*nr_ptnote += 1;
780 		*sz_ptnote += phdr_ptr->p_memsz;
781 	}
782 
783 	return 0;
784 }
785 
786 /**
787  * copy_notes_elf64 - copy ELF note segments in a given buffer
788  *
789  * @ehdr_ptr: ELF header
790  * @notes_buf: buffer into which ELF note segments are copied
791  *
792  * This function is used to copy ELF note segment in the 1st kernel
793  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
794  * size of the buffer @notes_buf is equal to or larger than sum of the
795  * real ELF note segment headers and data.
796  *
797  * It is assumed that program headers with PT_NOTE type pointed to by
798  * @ehdr_ptr has already been updated by update_note_header_size_elf64
799  * and each of PT_NOTE program headers has actual ELF note segment
800  * size in its p_memsz member.
801  */
802 static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
803 {
804 	int i, rc=0;
805 	Elf64_Phdr *phdr_ptr;
806 
807 	phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
808 
809 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
810 		u64 offset;
811 		if (phdr_ptr->p_type != PT_NOTE)
812 			continue;
813 		offset = phdr_ptr->p_offset;
814 		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
815 					   &offset);
816 		if (rc < 0)
817 			return rc;
818 		notes_buf += phdr_ptr->p_memsz;
819 	}
820 
821 	return 0;
822 }
823 
824 /* Merges all the PT_NOTE headers into one. */
825 static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
826 					   char **notes_buf, size_t *notes_sz)
827 {
828 	int i, nr_ptnote=0, rc=0;
829 	char *tmp;
830 	Elf64_Ehdr *ehdr_ptr;
831 	Elf64_Phdr phdr;
832 	u64 phdr_sz = 0, note_off;
833 
834 	ehdr_ptr = (Elf64_Ehdr *)elfptr;
835 
836 	rc = update_note_header_size_elf64(ehdr_ptr);
837 	if (rc < 0)
838 		return rc;
839 
840 	rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
841 	if (rc < 0)
842 		return rc;
843 
844 	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
845 	*notes_buf = vmcore_alloc_buf(*notes_sz);
846 	if (!*notes_buf)
847 		return -ENOMEM;
848 
849 	rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
850 	if (rc < 0)
851 		return rc;
852 
853 	/* Prepare merged PT_NOTE program header. */
854 	phdr.p_type    = PT_NOTE;
855 	phdr.p_flags   = 0;
856 	note_off = sizeof(Elf64_Ehdr) +
857 			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
858 	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
859 	phdr.p_vaddr   = phdr.p_paddr = 0;
860 	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
861 	phdr.p_align   = 0;
862 
863 	/* Add merged PT_NOTE program header*/
864 	tmp = elfptr + sizeof(Elf64_Ehdr);
865 	memcpy(tmp, &phdr, sizeof(phdr));
866 	tmp += sizeof(phdr);
867 
868 	/* Remove unwanted PT_NOTE program headers. */
869 	i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
870 	*elfsz = *elfsz - i;
871 	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
872 	memset(elfptr + *elfsz, 0, i);
873 	*elfsz = roundup(*elfsz, PAGE_SIZE);
874 
875 	/* Modify e_phnum to reflect merged headers. */
876 	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
877 
878 	/* Store the size of all notes.  We need this to update the note
879 	 * header when the device dumps will be added.
880 	 */
881 	elfnotes_orig_sz = phdr.p_memsz;
882 
883 	return 0;
884 }
885 
886 /**
887  * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
888  *
889  * @ehdr_ptr: ELF header
890  *
891  * This function updates p_memsz member of each PT_NOTE entry in the
892  * program header table pointed to by @ehdr_ptr to real size of ELF
893  * note segment.
894  */
895 static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
896 {
897 	int i, rc=0;
898 	Elf32_Phdr *phdr_ptr;
899 	Elf32_Nhdr *nhdr_ptr;
900 
901 	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
902 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
903 		void *notes_section;
904 		u64 offset, max_sz, sz, real_sz = 0;
905 		if (phdr_ptr->p_type != PT_NOTE)
906 			continue;
907 		max_sz = phdr_ptr->p_memsz;
908 		offset = phdr_ptr->p_offset;
909 		notes_section = kmalloc(max_sz, GFP_KERNEL);
910 		if (!notes_section)
911 			return -ENOMEM;
912 		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
913 		if (rc < 0) {
914 			kfree(notes_section);
915 			return rc;
916 		}
917 		nhdr_ptr = notes_section;
918 		while (nhdr_ptr->n_namesz != 0) {
919 			sz = sizeof(Elf32_Nhdr) +
920 				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
921 				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
922 			if ((real_sz + sz) > max_sz) {
923 				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
924 					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
925 				break;
926 			}
927 			real_sz += sz;
928 			nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
929 		}
930 		kfree(notes_section);
931 		phdr_ptr->p_memsz = real_sz;
932 		if (real_sz == 0) {
933 			pr_warn("Warning: Zero PT_NOTE entries found\n");
934 		}
935 	}
936 
937 	return 0;
938 }
939 
940 /**
941  * get_note_number_and_size_elf32 - get the number of PT_NOTE program
942  * headers and sum of real size of their ELF note segment headers and
943  * data.
944  *
945  * @ehdr_ptr: ELF header
946  * @nr_ptnote: buffer for the number of PT_NOTE program headers
947  * @sz_ptnote: buffer for size of unique PT_NOTE program header
948  *
949  * This function is used to merge multiple PT_NOTE program headers
950  * into a unique single one. The resulting unique entry will have
951  * @sz_ptnote in its phdr->p_mem.
952  *
953  * It is assumed that program headers with PT_NOTE type pointed to by
954  * @ehdr_ptr has already been updated by update_note_header_size_elf32
955  * and each of PT_NOTE program headers has actual ELF note segment
956  * size in its p_memsz member.
957  */
958 static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
959 						 int *nr_ptnote, u64 *sz_ptnote)
960 {
961 	int i;
962 	Elf32_Phdr *phdr_ptr;
963 
964 	*nr_ptnote = *sz_ptnote = 0;
965 
966 	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
967 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
968 		if (phdr_ptr->p_type != PT_NOTE)
969 			continue;
970 		*nr_ptnote += 1;
971 		*sz_ptnote += phdr_ptr->p_memsz;
972 	}
973 
974 	return 0;
975 }
976 
977 /**
978  * copy_notes_elf32 - copy ELF note segments in a given buffer
979  *
980  * @ehdr_ptr: ELF header
981  * @notes_buf: buffer into which ELF note segments are copied
982  *
983  * This function is used to copy ELF note segment in the 1st kernel
984  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
985  * size of the buffer @notes_buf is equal to or larger than sum of the
986  * real ELF note segment headers and data.
987  *
988  * It is assumed that program headers with PT_NOTE type pointed to by
989  * @ehdr_ptr has already been updated by update_note_header_size_elf32
990  * and each of PT_NOTE program headers has actual ELF note segment
991  * size in its p_memsz member.
992  */
993 static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
994 {
995 	int i, rc=0;
996 	Elf32_Phdr *phdr_ptr;
997 
998 	phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
999 
1000 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1001 		u64 offset;
1002 		if (phdr_ptr->p_type != PT_NOTE)
1003 			continue;
1004 		offset = phdr_ptr->p_offset;
1005 		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
1006 					   &offset);
1007 		if (rc < 0)
1008 			return rc;
1009 		notes_buf += phdr_ptr->p_memsz;
1010 	}
1011 
1012 	return 0;
1013 }
1014 
1015 /* Merges all the PT_NOTE headers into one. */
1016 static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
1017 					   char **notes_buf, size_t *notes_sz)
1018 {
1019 	int i, nr_ptnote=0, rc=0;
1020 	char *tmp;
1021 	Elf32_Ehdr *ehdr_ptr;
1022 	Elf32_Phdr phdr;
1023 	u64 phdr_sz = 0, note_off;
1024 
1025 	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1026 
1027 	rc = update_note_header_size_elf32(ehdr_ptr);
1028 	if (rc < 0)
1029 		return rc;
1030 
1031 	rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
1032 	if (rc < 0)
1033 		return rc;
1034 
1035 	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
1036 	*notes_buf = vmcore_alloc_buf(*notes_sz);
1037 	if (!*notes_buf)
1038 		return -ENOMEM;
1039 
1040 	rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
1041 	if (rc < 0)
1042 		return rc;
1043 
1044 	/* Prepare merged PT_NOTE program header. */
1045 	phdr.p_type    = PT_NOTE;
1046 	phdr.p_flags   = 0;
1047 	note_off = sizeof(Elf32_Ehdr) +
1048 			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
1049 	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
1050 	phdr.p_vaddr   = phdr.p_paddr = 0;
1051 	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
1052 	phdr.p_align   = 0;
1053 
1054 	/* Add merged PT_NOTE program header*/
1055 	tmp = elfptr + sizeof(Elf32_Ehdr);
1056 	memcpy(tmp, &phdr, sizeof(phdr));
1057 	tmp += sizeof(phdr);
1058 
1059 	/* Remove unwanted PT_NOTE program headers. */
1060 	i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
1061 	*elfsz = *elfsz - i;
1062 	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
1063 	memset(elfptr + *elfsz, 0, i);
1064 	*elfsz = roundup(*elfsz, PAGE_SIZE);
1065 
1066 	/* Modify e_phnum to reflect merged headers. */
1067 	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
1068 
1069 	/* Store the size of all notes.  We need this to update the note
1070 	 * header when the device dumps will be added.
1071 	 */
1072 	elfnotes_orig_sz = phdr.p_memsz;
1073 
1074 	return 0;
1075 }
1076 
1077 /* Add memory chunks represented by program headers to vmcore list. Also update
1078  * the new offset fields of exported program headers. */
1079 static int __init process_ptload_program_headers_elf64(char *elfptr,
1080 						size_t elfsz,
1081 						size_t elfnotes_sz,
1082 						struct list_head *vc_list)
1083 {
1084 	int i;
1085 	Elf64_Ehdr *ehdr_ptr;
1086 	Elf64_Phdr *phdr_ptr;
1087 	loff_t vmcore_off;
1088 	struct vmcore *new;
1089 
1090 	ehdr_ptr = (Elf64_Ehdr *)elfptr;
1091 	phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
1092 
1093 	/* Skip Elf header, program headers and Elf note segment. */
1094 	vmcore_off = elfsz + elfnotes_sz;
1095 
1096 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1097 		u64 paddr, start, end, size;
1098 
1099 		if (phdr_ptr->p_type != PT_LOAD)
1100 			continue;
1101 
1102 		paddr = phdr_ptr->p_offset;
1103 		start = rounddown(paddr, PAGE_SIZE);
1104 		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1105 		size = end - start;
1106 
1107 		/* Add this contiguous chunk of memory to vmcore list.*/
1108 		new = get_new_element();
1109 		if (!new)
1110 			return -ENOMEM;
1111 		new->paddr = start;
1112 		new->size = size;
1113 		list_add_tail(&new->list, vc_list);
1114 
1115 		/* Update the program header offset. */
1116 		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1117 		vmcore_off = vmcore_off + size;
1118 	}
1119 	return 0;
1120 }
1121 
1122 static int __init process_ptload_program_headers_elf32(char *elfptr,
1123 						size_t elfsz,
1124 						size_t elfnotes_sz,
1125 						struct list_head *vc_list)
1126 {
1127 	int i;
1128 	Elf32_Ehdr *ehdr_ptr;
1129 	Elf32_Phdr *phdr_ptr;
1130 	loff_t vmcore_off;
1131 	struct vmcore *new;
1132 
1133 	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1134 	phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
1135 
1136 	/* Skip Elf header, program headers and Elf note segment. */
1137 	vmcore_off = elfsz + elfnotes_sz;
1138 
1139 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1140 		u64 paddr, start, end, size;
1141 
1142 		if (phdr_ptr->p_type != PT_LOAD)
1143 			continue;
1144 
1145 		paddr = phdr_ptr->p_offset;
1146 		start = rounddown(paddr, PAGE_SIZE);
1147 		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1148 		size = end - start;
1149 
1150 		/* Add this contiguous chunk of memory to vmcore list.*/
1151 		new = get_new_element();
1152 		if (!new)
1153 			return -ENOMEM;
1154 		new->paddr = start;
1155 		new->size = size;
1156 		list_add_tail(&new->list, vc_list);
1157 
1158 		/* Update the program header offset */
1159 		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1160 		vmcore_off = vmcore_off + size;
1161 	}
1162 	return 0;
1163 }
1164 
1165 /* Sets offset fields of vmcore elements. */
1166 static void set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
1167 				    struct list_head *vc_list)
1168 {
1169 	loff_t vmcore_off;
1170 	struct vmcore *m;
1171 
1172 	/* Skip Elf header, program headers and Elf note segment. */
1173 	vmcore_off = elfsz + elfnotes_sz;
1174 
1175 	list_for_each_entry(m, vc_list, list) {
1176 		m->offset = vmcore_off;
1177 		vmcore_off += m->size;
1178 	}
1179 }
1180 
1181 static void free_elfcorebuf(void)
1182 {
1183 	free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
1184 	elfcorebuf = NULL;
1185 	vfree(elfnotes_buf);
1186 	elfnotes_buf = NULL;
1187 }
1188 
1189 static int __init parse_crash_elf64_headers(void)
1190 {
1191 	int rc=0;
1192 	Elf64_Ehdr ehdr;
1193 	u64 addr;
1194 
1195 	addr = elfcorehdr_addr;
1196 
1197 	/* Read Elf header */
1198 	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
1199 	if (rc < 0)
1200 		return rc;
1201 
1202 	/* Do some basic Verification. */
1203 	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1204 		(ehdr.e_type != ET_CORE) ||
1205 		!vmcore_elf64_check_arch(&ehdr) ||
1206 		ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
1207 		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1208 		ehdr.e_version != EV_CURRENT ||
1209 		ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
1210 		ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
1211 		ehdr.e_phnum == 0) {
1212 		pr_warn("Warning: Core image elf header is not sane\n");
1213 		return -EINVAL;
1214 	}
1215 
1216 	/* Read in all elf headers. */
1217 	elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
1218 				ehdr.e_phnum * sizeof(Elf64_Phdr);
1219 	elfcorebuf_sz = elfcorebuf_sz_orig;
1220 	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1221 					      get_order(elfcorebuf_sz_orig));
1222 	if (!elfcorebuf)
1223 		return -ENOMEM;
1224 	addr = elfcorehdr_addr;
1225 	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1226 	if (rc < 0)
1227 		goto fail;
1228 
1229 	/* Merge all PT_NOTE headers into one. */
1230 	rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
1231 				      &elfnotes_buf, &elfnotes_sz);
1232 	if (rc)
1233 		goto fail;
1234 	rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
1235 						  elfnotes_sz, &vmcore_list);
1236 	if (rc)
1237 		goto fail;
1238 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1239 	return 0;
1240 fail:
1241 	free_elfcorebuf();
1242 	return rc;
1243 }
1244 
1245 static int __init parse_crash_elf32_headers(void)
1246 {
1247 	int rc=0;
1248 	Elf32_Ehdr ehdr;
1249 	u64 addr;
1250 
1251 	addr = elfcorehdr_addr;
1252 
1253 	/* Read Elf header */
1254 	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
1255 	if (rc < 0)
1256 		return rc;
1257 
1258 	/* Do some basic Verification. */
1259 	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1260 		(ehdr.e_type != ET_CORE) ||
1261 		!vmcore_elf32_check_arch(&ehdr) ||
1262 		ehdr.e_ident[EI_CLASS] != ELFCLASS32||
1263 		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1264 		ehdr.e_version != EV_CURRENT ||
1265 		ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1266 		ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1267 		ehdr.e_phnum == 0) {
1268 		pr_warn("Warning: Core image elf header is not sane\n");
1269 		return -EINVAL;
1270 	}
1271 
1272 	/* Read in all elf headers. */
1273 	elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1274 	elfcorebuf_sz = elfcorebuf_sz_orig;
1275 	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1276 					      get_order(elfcorebuf_sz_orig));
1277 	if (!elfcorebuf)
1278 		return -ENOMEM;
1279 	addr = elfcorehdr_addr;
1280 	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1281 	if (rc < 0)
1282 		goto fail;
1283 
1284 	/* Merge all PT_NOTE headers into one. */
1285 	rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1286 				      &elfnotes_buf, &elfnotes_sz);
1287 	if (rc)
1288 		goto fail;
1289 	rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1290 						  elfnotes_sz, &vmcore_list);
1291 	if (rc)
1292 		goto fail;
1293 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1294 	return 0;
1295 fail:
1296 	free_elfcorebuf();
1297 	return rc;
1298 }
1299 
1300 static int __init parse_crash_elf_headers(void)
1301 {
1302 	unsigned char e_ident[EI_NIDENT];
1303 	u64 addr;
1304 	int rc=0;
1305 
1306 	addr = elfcorehdr_addr;
1307 	rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1308 	if (rc < 0)
1309 		return rc;
1310 	if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1311 		pr_warn("Warning: Core image elf header not found\n");
1312 		return -EINVAL;
1313 	}
1314 
1315 	if (e_ident[EI_CLASS] == ELFCLASS64) {
1316 		rc = parse_crash_elf64_headers();
1317 		if (rc)
1318 			return rc;
1319 	} else if (e_ident[EI_CLASS] == ELFCLASS32) {
1320 		rc = parse_crash_elf32_headers();
1321 		if (rc)
1322 			return rc;
1323 	} else {
1324 		pr_warn("Warning: Core image elf header is not sane\n");
1325 		return -EINVAL;
1326 	}
1327 
1328 	/* Determine vmcore size. */
1329 	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1330 				      &vmcore_list);
1331 
1332 	return 0;
1333 }
1334 
1335 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1336 /**
1337  * vmcoredd_write_header - Write vmcore device dump header at the
1338  * beginning of the dump's buffer.
1339  * @buf: Output buffer where the note is written
1340  * @data: Dump info
1341  * @size: Size of the dump
1342  *
1343  * Fills beginning of the dump's buffer with vmcore device dump header.
1344  */
1345 static void vmcoredd_write_header(void *buf, struct vmcoredd_data *data,
1346 				  u32 size)
1347 {
1348 	struct vmcoredd_header *vdd_hdr = (struct vmcoredd_header *)buf;
1349 
1350 	vdd_hdr->n_namesz = sizeof(vdd_hdr->name);
1351 	vdd_hdr->n_descsz = size + sizeof(vdd_hdr->dump_name);
1352 	vdd_hdr->n_type = NT_VMCOREDD;
1353 
1354 	strncpy((char *)vdd_hdr->name, VMCOREDD_NOTE_NAME,
1355 		sizeof(vdd_hdr->name));
1356 	memcpy(vdd_hdr->dump_name, data->dump_name, sizeof(vdd_hdr->dump_name));
1357 }
1358 
1359 /**
1360  * vmcoredd_update_program_headers - Update all Elf program headers
1361  * @elfptr: Pointer to elf header
1362  * @elfnotesz: Size of elf notes aligned to page size
1363  * @vmcoreddsz: Size of device dumps to be added to elf note header
1364  *
1365  * Determine type of Elf header (Elf64 or Elf32) and update the elf note size.
1366  * Also update the offsets of all the program headers after the elf note header.
1367  */
1368 static void vmcoredd_update_program_headers(char *elfptr, size_t elfnotesz,
1369 					    size_t vmcoreddsz)
1370 {
1371 	unsigned char *e_ident = (unsigned char *)elfptr;
1372 	u64 start, end, size;
1373 	loff_t vmcore_off;
1374 	u32 i;
1375 
1376 	vmcore_off = elfcorebuf_sz + elfnotesz;
1377 
1378 	if (e_ident[EI_CLASS] == ELFCLASS64) {
1379 		Elf64_Ehdr *ehdr = (Elf64_Ehdr *)elfptr;
1380 		Elf64_Phdr *phdr = (Elf64_Phdr *)(elfptr + sizeof(Elf64_Ehdr));
1381 
1382 		/* Update all program headers */
1383 		for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1384 			if (phdr->p_type == PT_NOTE) {
1385 				/* Update note size */
1386 				phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1387 				phdr->p_filesz = phdr->p_memsz;
1388 				continue;
1389 			}
1390 
1391 			start = rounddown(phdr->p_offset, PAGE_SIZE);
1392 			end = roundup(phdr->p_offset + phdr->p_memsz,
1393 				      PAGE_SIZE);
1394 			size = end - start;
1395 			phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1396 			vmcore_off += size;
1397 		}
1398 	} else {
1399 		Elf32_Ehdr *ehdr = (Elf32_Ehdr *)elfptr;
1400 		Elf32_Phdr *phdr = (Elf32_Phdr *)(elfptr + sizeof(Elf32_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 	}
1419 }
1420 
1421 /**
1422  * vmcoredd_update_size - Update the total size of the device dumps and update
1423  * Elf header
1424  * @dump_size: Size of the current device dump to be added to total size
1425  *
1426  * Update the total size of all the device dumps and update the Elf program
1427  * headers. Calculate the new offsets for the vmcore list and update the
1428  * total vmcore size.
1429  */
1430 static void vmcoredd_update_size(size_t dump_size)
1431 {
1432 	vmcoredd_orig_sz += dump_size;
1433 	elfnotes_sz = roundup(elfnotes_orig_sz, PAGE_SIZE) + vmcoredd_orig_sz;
1434 	vmcoredd_update_program_headers(elfcorebuf, elfnotes_sz,
1435 					vmcoredd_orig_sz);
1436 
1437 	/* Update vmcore list offsets */
1438 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1439 
1440 	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1441 				      &vmcore_list);
1442 	proc_vmcore->size = vmcore_size;
1443 }
1444 
1445 /**
1446  * vmcore_add_device_dump - Add a buffer containing device dump to vmcore
1447  * @data: dump info.
1448  *
1449  * Allocate a buffer and invoke the calling driver's dump collect routine.
1450  * Write Elf note at the beginning of the buffer to indicate vmcore device
1451  * dump and add the dump to global list.
1452  */
1453 int vmcore_add_device_dump(struct vmcoredd_data *data)
1454 {
1455 	struct vmcoredd_node *dump;
1456 	void *buf = NULL;
1457 	size_t data_size;
1458 	int ret;
1459 
1460 	if (vmcoredd_disabled) {
1461 		pr_err_once("Device dump is disabled\n");
1462 		return -EINVAL;
1463 	}
1464 
1465 	if (!data || !strlen(data->dump_name) ||
1466 	    !data->vmcoredd_callback || !data->size)
1467 		return -EINVAL;
1468 
1469 	dump = vzalloc(sizeof(*dump));
1470 	if (!dump) {
1471 		ret = -ENOMEM;
1472 		goto out_err;
1473 	}
1474 
1475 	/* Keep size of the buffer page aligned so that it can be mmaped */
1476 	data_size = roundup(sizeof(struct vmcoredd_header) + data->size,
1477 			    PAGE_SIZE);
1478 
1479 	/* Allocate buffer for driver's to write their dumps */
1480 	buf = vmcore_alloc_buf(data_size);
1481 	if (!buf) {
1482 		ret = -ENOMEM;
1483 		goto out_err;
1484 	}
1485 
1486 	vmcoredd_write_header(buf, data, data_size -
1487 			      sizeof(struct vmcoredd_header));
1488 
1489 	/* Invoke the driver's dump collection routing */
1490 	ret = data->vmcoredd_callback(data, buf +
1491 				      sizeof(struct vmcoredd_header));
1492 	if (ret)
1493 		goto out_err;
1494 
1495 	dump->buf = buf;
1496 	dump->size = data_size;
1497 
1498 	/* Add the dump to driver sysfs list */
1499 	mutex_lock(&vmcoredd_mutex);
1500 	list_add_tail(&dump->list, &vmcoredd_list);
1501 	mutex_unlock(&vmcoredd_mutex);
1502 
1503 	vmcoredd_update_size(data_size);
1504 	return 0;
1505 
1506 out_err:
1507 	if (buf)
1508 		vfree(buf);
1509 
1510 	if (dump)
1511 		vfree(dump);
1512 
1513 	return ret;
1514 }
1515 EXPORT_SYMBOL(vmcore_add_device_dump);
1516 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1517 
1518 /* Free all dumps in vmcore device dump list */
1519 static void vmcore_free_device_dumps(void)
1520 {
1521 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1522 	mutex_lock(&vmcoredd_mutex);
1523 	while (!list_empty(&vmcoredd_list)) {
1524 		struct vmcoredd_node *dump;
1525 
1526 		dump = list_first_entry(&vmcoredd_list, struct vmcoredd_node,
1527 					list);
1528 		list_del(&dump->list);
1529 		vfree(dump->buf);
1530 		vfree(dump);
1531 	}
1532 	mutex_unlock(&vmcoredd_mutex);
1533 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1534 }
1535 
1536 /* Init function for vmcore module. */
1537 static int __init vmcore_init(void)
1538 {
1539 	int rc = 0;
1540 
1541 	/* Allow architectures to allocate ELF header in 2nd kernel */
1542 	rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1543 	if (rc)
1544 		return rc;
1545 	/*
1546 	 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1547 	 * then capture the dump.
1548 	 */
1549 	if (!(is_vmcore_usable()))
1550 		return rc;
1551 	rc = parse_crash_elf_headers();
1552 	if (rc) {
1553 		pr_warn("Kdump: vmcore not initialized\n");
1554 		return rc;
1555 	}
1556 	elfcorehdr_free(elfcorehdr_addr);
1557 	elfcorehdr_addr = ELFCORE_ADDR_ERR;
1558 
1559 	proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &vmcore_proc_ops);
1560 	if (proc_vmcore)
1561 		proc_vmcore->size = vmcore_size;
1562 	return 0;
1563 }
1564 fs_initcall(vmcore_init);
1565 
1566 /* Cleanup function for vmcore module. */
1567 void vmcore_cleanup(void)
1568 {
1569 	if (proc_vmcore) {
1570 		proc_remove(proc_vmcore);
1571 		proc_vmcore = NULL;
1572 	}
1573 
1574 	/* clear the vmcore list. */
1575 	while (!list_empty(&vmcore_list)) {
1576 		struct vmcore *m;
1577 
1578 		m = list_first_entry(&vmcore_list, struct vmcore, list);
1579 		list_del(&m->list);
1580 		kfree(m);
1581 	}
1582 	free_elfcorebuf();
1583 
1584 	/* clear vmcore device dump list */
1585 	vmcore_free_device_dumps();
1586 }
1587