xref: /openbmc/linux/fs/proc/vmcore.c (revision 323dd2c3)
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, tsz)) {
270 				ret = -EFAULT;
271 				goto out_unlock;
272 			}
273 
274 			size -= tsz;
275 			start += tsz;
276 			dst += tsz;
277 
278 			/* Leave now if buffer filled already */
279 			if (!size)
280 				goto out_unlock;
281 		}
282 		offset += dump->size;
283 	}
284 
285 out_unlock:
286 	mutex_unlock(&vmcoredd_mutex);
287 	return ret;
288 }
289 #endif /* CONFIG_MMU */
290 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
291 
292 /* Read from the ELF header and then the crash dump. On error, negative value is
293  * returned otherwise number of bytes read are returned.
294  */
295 static ssize_t __read_vmcore(char *buffer, size_t buflen, loff_t *fpos,
296 			     int userbuf)
297 {
298 	ssize_t acc = 0, tmp;
299 	size_t tsz;
300 	u64 start;
301 	struct vmcore *m = NULL;
302 
303 	if (buflen == 0 || *fpos >= vmcore_size)
304 		return 0;
305 
306 	/* trim buflen to not go beyond EOF */
307 	if (buflen > vmcore_size - *fpos)
308 		buflen = vmcore_size - *fpos;
309 
310 	/* Read ELF core header */
311 	if (*fpos < elfcorebuf_sz) {
312 		tsz = min(elfcorebuf_sz - (size_t)*fpos, buflen);
313 		if (copy_to(buffer, elfcorebuf + *fpos, tsz, userbuf))
314 			return -EFAULT;
315 		buflen -= tsz;
316 		*fpos += tsz;
317 		buffer += tsz;
318 		acc += tsz;
319 
320 		/* leave now if filled buffer already */
321 		if (buflen == 0)
322 			return acc;
323 	}
324 
325 	/* Read Elf note segment */
326 	if (*fpos < elfcorebuf_sz + elfnotes_sz) {
327 		void *kaddr;
328 
329 		/* We add device dumps before other elf notes because the
330 		 * other elf notes may not fill the elf notes buffer
331 		 * completely and we will end up with zero-filled data
332 		 * between the elf notes and the device dumps. Tools will
333 		 * then try to decode this zero-filled data as valid notes
334 		 * and we don't want that. Hence, adding device dumps before
335 		 * the other elf notes ensure that zero-filled data can be
336 		 * avoided.
337 		 */
338 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
339 		/* Read device dumps */
340 		if (*fpos < elfcorebuf_sz + vmcoredd_orig_sz) {
341 			tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
342 				  (size_t)*fpos, buflen);
343 			start = *fpos - elfcorebuf_sz;
344 			if (vmcoredd_copy_dumps(buffer, start, tsz, userbuf))
345 				return -EFAULT;
346 
347 			buflen -= tsz;
348 			*fpos += tsz;
349 			buffer += tsz;
350 			acc += tsz;
351 
352 			/* leave now if filled buffer already */
353 			if (!buflen)
354 				return acc;
355 		}
356 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
357 
358 		/* Read remaining elf notes */
359 		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)*fpos, buflen);
360 		kaddr = elfnotes_buf + *fpos - elfcorebuf_sz - vmcoredd_orig_sz;
361 		if (copy_to(buffer, kaddr, tsz, userbuf))
362 			return -EFAULT;
363 
364 		buflen -= tsz;
365 		*fpos += tsz;
366 		buffer += tsz;
367 		acc += tsz;
368 
369 		/* leave now if filled buffer already */
370 		if (buflen == 0)
371 			return acc;
372 	}
373 
374 	list_for_each_entry(m, &vmcore_list, list) {
375 		if (*fpos < m->offset + m->size) {
376 			tsz = (size_t)min_t(unsigned long long,
377 					    m->offset + m->size - *fpos,
378 					    buflen);
379 			start = m->paddr + *fpos - m->offset;
380 			tmp = read_from_oldmem(buffer, tsz, &start,
381 					       userbuf, mem_encrypt_active());
382 			if (tmp < 0)
383 				return tmp;
384 			buflen -= tsz;
385 			*fpos += tsz;
386 			buffer += tsz;
387 			acc += tsz;
388 
389 			/* leave now if filled buffer already */
390 			if (buflen == 0)
391 				return acc;
392 		}
393 	}
394 
395 	return acc;
396 }
397 
398 static ssize_t read_vmcore(struct file *file, char __user *buffer,
399 			   size_t buflen, loff_t *fpos)
400 {
401 	return __read_vmcore((__force char *) buffer, buflen, fpos, 1);
402 }
403 
404 /*
405  * The vmcore fault handler uses the page cache and fills data using the
406  * standard __vmcore_read() function.
407  *
408  * On s390 the fault handler is used for memory regions that can't be mapped
409  * directly with remap_pfn_range().
410  */
411 static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
412 {
413 #ifdef CONFIG_S390
414 	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
415 	pgoff_t index = vmf->pgoff;
416 	struct page *page;
417 	loff_t offset;
418 	char *buf;
419 	int rc;
420 
421 	page = find_or_create_page(mapping, index, GFP_KERNEL);
422 	if (!page)
423 		return VM_FAULT_OOM;
424 	if (!PageUptodate(page)) {
425 		offset = (loff_t) index << PAGE_SHIFT;
426 		buf = __va((page_to_pfn(page) << PAGE_SHIFT));
427 		rc = __read_vmcore(buf, PAGE_SIZE, &offset, 0);
428 		if (rc < 0) {
429 			unlock_page(page);
430 			put_page(page);
431 			return vmf_error(rc);
432 		}
433 		SetPageUptodate(page);
434 	}
435 	unlock_page(page);
436 	vmf->page = page;
437 	return 0;
438 #else
439 	return VM_FAULT_SIGBUS;
440 #endif
441 }
442 
443 static const struct vm_operations_struct vmcore_mmap_ops = {
444 	.fault = mmap_vmcore_fault,
445 };
446 
447 /**
448  * vmcore_alloc_buf - allocate buffer in vmalloc memory
449  * @sizez: size of buffer
450  *
451  * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
452  * the buffer to user-space by means of remap_vmalloc_range().
453  *
454  * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
455  * disabled and there's no need to allow users to mmap the buffer.
456  */
457 static inline char *vmcore_alloc_buf(size_t size)
458 {
459 #ifdef CONFIG_MMU
460 	return vmalloc_user(size);
461 #else
462 	return vzalloc(size);
463 #endif
464 }
465 
466 /*
467  * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
468  * essential for mmap_vmcore() in order to map physically
469  * non-contiguous objects (ELF header, ELF note segment and memory
470  * regions in the 1st kernel pointed to by PT_LOAD entries) into
471  * virtually contiguous user-space in ELF layout.
472  */
473 #ifdef CONFIG_MMU
474 /*
475  * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
476  * reported as not being ram with the zero page.
477  *
478  * @vma: vm_area_struct describing requested mapping
479  * @from: start remapping from
480  * @pfn: page frame number to start remapping to
481  * @size: remapping size
482  * @prot: protection bits
483  *
484  * Returns zero on success, -EAGAIN on failure.
485  */
486 static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
487 				    unsigned long from, unsigned long pfn,
488 				    unsigned long size, pgprot_t prot)
489 {
490 	unsigned long map_size;
491 	unsigned long pos_start, pos_end, pos;
492 	unsigned long zeropage_pfn = my_zero_pfn(0);
493 	size_t len = 0;
494 
495 	pos_start = pfn;
496 	pos_end = pfn + (size >> PAGE_SHIFT);
497 
498 	for (pos = pos_start; pos < pos_end; ++pos) {
499 		if (!pfn_is_ram(pos)) {
500 			/*
501 			 * We hit a page which is not ram. Remap the continuous
502 			 * region between pos_start and pos-1 and replace
503 			 * the non-ram page at pos with the zero page.
504 			 */
505 			if (pos > pos_start) {
506 				/* Remap continuous region */
507 				map_size = (pos - pos_start) << PAGE_SHIFT;
508 				if (remap_oldmem_pfn_range(vma, from + len,
509 							   pos_start, map_size,
510 							   prot))
511 					goto fail;
512 				len += map_size;
513 			}
514 			/* Remap the zero page */
515 			if (remap_oldmem_pfn_range(vma, from + len,
516 						   zeropage_pfn,
517 						   PAGE_SIZE, prot))
518 				goto fail;
519 			len += PAGE_SIZE;
520 			pos_start = pos + 1;
521 		}
522 	}
523 	if (pos > pos_start) {
524 		/* Remap the rest */
525 		map_size = (pos - pos_start) << PAGE_SHIFT;
526 		if (remap_oldmem_pfn_range(vma, from + len, pos_start,
527 					   map_size, prot))
528 			goto fail;
529 	}
530 	return 0;
531 fail:
532 	do_munmap(vma->vm_mm, from, len, NULL);
533 	return -EAGAIN;
534 }
535 
536 static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
537 			    unsigned long from, unsigned long pfn,
538 			    unsigned long size, pgprot_t prot)
539 {
540 	/*
541 	 * Check if oldmem_pfn_is_ram was registered to avoid
542 	 * looping over all pages without a reason.
543 	 */
544 	if (oldmem_pfn_is_ram)
545 		return remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
546 	else
547 		return remap_oldmem_pfn_range(vma, from, pfn, size, prot);
548 }
549 
550 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
551 {
552 	size_t size = vma->vm_end - vma->vm_start;
553 	u64 start, end, len, tsz;
554 	struct vmcore *m;
555 
556 	start = (u64)vma->vm_pgoff << PAGE_SHIFT;
557 	end = start + size;
558 
559 	if (size > vmcore_size || end > vmcore_size)
560 		return -EINVAL;
561 
562 	if (vma->vm_flags & (VM_WRITE | VM_EXEC))
563 		return -EPERM;
564 
565 	vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
566 	vma->vm_flags |= VM_MIXEDMAP;
567 	vma->vm_ops = &vmcore_mmap_ops;
568 
569 	len = 0;
570 
571 	if (start < elfcorebuf_sz) {
572 		u64 pfn;
573 
574 		tsz = min(elfcorebuf_sz - (size_t)start, size);
575 		pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
576 		if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
577 				    vma->vm_page_prot))
578 			return -EAGAIN;
579 		size -= tsz;
580 		start += tsz;
581 		len += tsz;
582 
583 		if (size == 0)
584 			return 0;
585 	}
586 
587 	if (start < elfcorebuf_sz + elfnotes_sz) {
588 		void *kaddr;
589 
590 		/* We add device dumps before other elf notes because the
591 		 * other elf notes may not fill the elf notes buffer
592 		 * completely and we will end up with zero-filled data
593 		 * between the elf notes and the device dumps. Tools will
594 		 * then try to decode this zero-filled data as valid notes
595 		 * and we don't want that. Hence, adding device dumps before
596 		 * the other elf notes ensure that zero-filled data can be
597 		 * avoided. This also ensures that the device dumps and
598 		 * other elf notes can be properly mmaped at page aligned
599 		 * address.
600 		 */
601 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
602 		/* Read device dumps */
603 		if (start < elfcorebuf_sz + vmcoredd_orig_sz) {
604 			u64 start_off;
605 
606 			tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
607 				  (size_t)start, size);
608 			start_off = start - elfcorebuf_sz;
609 			if (vmcoredd_mmap_dumps(vma, vma->vm_start + len,
610 						start_off, tsz))
611 				goto fail;
612 
613 			size -= tsz;
614 			start += tsz;
615 			len += tsz;
616 
617 			/* leave now if filled buffer already */
618 			if (!size)
619 				return 0;
620 		}
621 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
622 
623 		/* Read remaining elf notes */
624 		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
625 		kaddr = elfnotes_buf + start - elfcorebuf_sz - vmcoredd_orig_sz;
626 		if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
627 						kaddr, tsz))
628 			goto fail;
629 
630 		size -= tsz;
631 		start += tsz;
632 		len += tsz;
633 
634 		if (size == 0)
635 			return 0;
636 	}
637 
638 	list_for_each_entry(m, &vmcore_list, list) {
639 		if (start < m->offset + m->size) {
640 			u64 paddr = 0;
641 
642 			tsz = (size_t)min_t(unsigned long long,
643 					    m->offset + m->size - start, size);
644 			paddr = m->paddr + start - m->offset;
645 			if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
646 						    paddr >> PAGE_SHIFT, tsz,
647 						    vma->vm_page_prot))
648 				goto fail;
649 			size -= tsz;
650 			start += tsz;
651 			len += tsz;
652 
653 			if (size == 0)
654 				return 0;
655 		}
656 	}
657 
658 	return 0;
659 fail:
660 	do_munmap(vma->vm_mm, vma->vm_start, len, NULL);
661 	return -EAGAIN;
662 }
663 #else
664 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
665 {
666 	return -ENOSYS;
667 }
668 #endif
669 
670 static const struct file_operations proc_vmcore_operations = {
671 	.read		= read_vmcore,
672 	.llseek		= default_llseek,
673 	.mmap		= mmap_vmcore,
674 };
675 
676 static struct vmcore* __init get_new_element(void)
677 {
678 	return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
679 }
680 
681 static u64 get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
682 			   struct list_head *vc_list)
683 {
684 	u64 size;
685 	struct vmcore *m;
686 
687 	size = elfsz + elfnotesegsz;
688 	list_for_each_entry(m, vc_list, list) {
689 		size += m->size;
690 	}
691 	return size;
692 }
693 
694 /**
695  * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
696  *
697  * @ehdr_ptr: ELF header
698  *
699  * This function updates p_memsz member of each PT_NOTE entry in the
700  * program header table pointed to by @ehdr_ptr to real size of ELF
701  * note segment.
702  */
703 static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
704 {
705 	int i, rc=0;
706 	Elf64_Phdr *phdr_ptr;
707 	Elf64_Nhdr *nhdr_ptr;
708 
709 	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
710 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
711 		void *notes_section;
712 		u64 offset, max_sz, sz, real_sz = 0;
713 		if (phdr_ptr->p_type != PT_NOTE)
714 			continue;
715 		max_sz = phdr_ptr->p_memsz;
716 		offset = phdr_ptr->p_offset;
717 		notes_section = kmalloc(max_sz, GFP_KERNEL);
718 		if (!notes_section)
719 			return -ENOMEM;
720 		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
721 		if (rc < 0) {
722 			kfree(notes_section);
723 			return rc;
724 		}
725 		nhdr_ptr = notes_section;
726 		while (nhdr_ptr->n_namesz != 0) {
727 			sz = sizeof(Elf64_Nhdr) +
728 				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
729 				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
730 			if ((real_sz + sz) > max_sz) {
731 				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
732 					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
733 				break;
734 			}
735 			real_sz += sz;
736 			nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
737 		}
738 		kfree(notes_section);
739 		phdr_ptr->p_memsz = real_sz;
740 		if (real_sz == 0) {
741 			pr_warn("Warning: Zero PT_NOTE entries found\n");
742 		}
743 	}
744 
745 	return 0;
746 }
747 
748 /**
749  * get_note_number_and_size_elf64 - get the number of PT_NOTE program
750  * headers and sum of real size of their ELF note segment headers and
751  * data.
752  *
753  * @ehdr_ptr: ELF header
754  * @nr_ptnote: buffer for the number of PT_NOTE program headers
755  * @sz_ptnote: buffer for size of unique PT_NOTE program header
756  *
757  * This function is used to merge multiple PT_NOTE program headers
758  * into a unique single one. The resulting unique entry will have
759  * @sz_ptnote in its phdr->p_mem.
760  *
761  * It is assumed that program headers with PT_NOTE type pointed to by
762  * @ehdr_ptr has already been updated by update_note_header_size_elf64
763  * and each of PT_NOTE program headers has actual ELF note segment
764  * size in its p_memsz member.
765  */
766 static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
767 						 int *nr_ptnote, u64 *sz_ptnote)
768 {
769 	int i;
770 	Elf64_Phdr *phdr_ptr;
771 
772 	*nr_ptnote = *sz_ptnote = 0;
773 
774 	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
775 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
776 		if (phdr_ptr->p_type != PT_NOTE)
777 			continue;
778 		*nr_ptnote += 1;
779 		*sz_ptnote += phdr_ptr->p_memsz;
780 	}
781 
782 	return 0;
783 }
784 
785 /**
786  * copy_notes_elf64 - copy ELF note segments in a given buffer
787  *
788  * @ehdr_ptr: ELF header
789  * @notes_buf: buffer into which ELF note segments are copied
790  *
791  * This function is used to copy ELF note segment in the 1st kernel
792  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
793  * size of the buffer @notes_buf is equal to or larger than sum of the
794  * real ELF note segment headers and data.
795  *
796  * It is assumed that program headers with PT_NOTE type pointed to by
797  * @ehdr_ptr has already been updated by update_note_header_size_elf64
798  * and each of PT_NOTE program headers has actual ELF note segment
799  * size in its p_memsz member.
800  */
801 static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
802 {
803 	int i, rc=0;
804 	Elf64_Phdr *phdr_ptr;
805 
806 	phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
807 
808 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
809 		u64 offset;
810 		if (phdr_ptr->p_type != PT_NOTE)
811 			continue;
812 		offset = phdr_ptr->p_offset;
813 		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
814 					   &offset);
815 		if (rc < 0)
816 			return rc;
817 		notes_buf += phdr_ptr->p_memsz;
818 	}
819 
820 	return 0;
821 }
822 
823 /* Merges all the PT_NOTE headers into one. */
824 static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
825 					   char **notes_buf, size_t *notes_sz)
826 {
827 	int i, nr_ptnote=0, rc=0;
828 	char *tmp;
829 	Elf64_Ehdr *ehdr_ptr;
830 	Elf64_Phdr phdr;
831 	u64 phdr_sz = 0, note_off;
832 
833 	ehdr_ptr = (Elf64_Ehdr *)elfptr;
834 
835 	rc = update_note_header_size_elf64(ehdr_ptr);
836 	if (rc < 0)
837 		return rc;
838 
839 	rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
840 	if (rc < 0)
841 		return rc;
842 
843 	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
844 	*notes_buf = vmcore_alloc_buf(*notes_sz);
845 	if (!*notes_buf)
846 		return -ENOMEM;
847 
848 	rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
849 	if (rc < 0)
850 		return rc;
851 
852 	/* Prepare merged PT_NOTE program header. */
853 	phdr.p_type    = PT_NOTE;
854 	phdr.p_flags   = 0;
855 	note_off = sizeof(Elf64_Ehdr) +
856 			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
857 	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
858 	phdr.p_vaddr   = phdr.p_paddr = 0;
859 	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
860 	phdr.p_align   = 0;
861 
862 	/* Add merged PT_NOTE program header*/
863 	tmp = elfptr + sizeof(Elf64_Ehdr);
864 	memcpy(tmp, &phdr, sizeof(phdr));
865 	tmp += sizeof(phdr);
866 
867 	/* Remove unwanted PT_NOTE program headers. */
868 	i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
869 	*elfsz = *elfsz - i;
870 	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
871 	memset(elfptr + *elfsz, 0, i);
872 	*elfsz = roundup(*elfsz, PAGE_SIZE);
873 
874 	/* Modify e_phnum to reflect merged headers. */
875 	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
876 
877 	/* Store the size of all notes.  We need this to update the note
878 	 * header when the device dumps will be added.
879 	 */
880 	elfnotes_orig_sz = phdr.p_memsz;
881 
882 	return 0;
883 }
884 
885 /**
886  * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
887  *
888  * @ehdr_ptr: ELF header
889  *
890  * This function updates p_memsz member of each PT_NOTE entry in the
891  * program header table pointed to by @ehdr_ptr to real size of ELF
892  * note segment.
893  */
894 static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
895 {
896 	int i, rc=0;
897 	Elf32_Phdr *phdr_ptr;
898 	Elf32_Nhdr *nhdr_ptr;
899 
900 	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
901 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
902 		void *notes_section;
903 		u64 offset, max_sz, sz, real_sz = 0;
904 		if (phdr_ptr->p_type != PT_NOTE)
905 			continue;
906 		max_sz = phdr_ptr->p_memsz;
907 		offset = phdr_ptr->p_offset;
908 		notes_section = kmalloc(max_sz, GFP_KERNEL);
909 		if (!notes_section)
910 			return -ENOMEM;
911 		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
912 		if (rc < 0) {
913 			kfree(notes_section);
914 			return rc;
915 		}
916 		nhdr_ptr = notes_section;
917 		while (nhdr_ptr->n_namesz != 0) {
918 			sz = sizeof(Elf32_Nhdr) +
919 				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
920 				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
921 			if ((real_sz + sz) > max_sz) {
922 				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
923 					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
924 				break;
925 			}
926 			real_sz += sz;
927 			nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
928 		}
929 		kfree(notes_section);
930 		phdr_ptr->p_memsz = real_sz;
931 		if (real_sz == 0) {
932 			pr_warn("Warning: Zero PT_NOTE entries found\n");
933 		}
934 	}
935 
936 	return 0;
937 }
938 
939 /**
940  * get_note_number_and_size_elf32 - get the number of PT_NOTE program
941  * headers and sum of real size of their ELF note segment headers and
942  * data.
943  *
944  * @ehdr_ptr: ELF header
945  * @nr_ptnote: buffer for the number of PT_NOTE program headers
946  * @sz_ptnote: buffer for size of unique PT_NOTE program header
947  *
948  * This function is used to merge multiple PT_NOTE program headers
949  * into a unique single one. The resulting unique entry will have
950  * @sz_ptnote in its phdr->p_mem.
951  *
952  * It is assumed that program headers with PT_NOTE type pointed to by
953  * @ehdr_ptr has already been updated by update_note_header_size_elf32
954  * and each of PT_NOTE program headers has actual ELF note segment
955  * size in its p_memsz member.
956  */
957 static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
958 						 int *nr_ptnote, u64 *sz_ptnote)
959 {
960 	int i;
961 	Elf32_Phdr *phdr_ptr;
962 
963 	*nr_ptnote = *sz_ptnote = 0;
964 
965 	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
966 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
967 		if (phdr_ptr->p_type != PT_NOTE)
968 			continue;
969 		*nr_ptnote += 1;
970 		*sz_ptnote += phdr_ptr->p_memsz;
971 	}
972 
973 	return 0;
974 }
975 
976 /**
977  * copy_notes_elf32 - copy ELF note segments in a given buffer
978  *
979  * @ehdr_ptr: ELF header
980  * @notes_buf: buffer into which ELF note segments are copied
981  *
982  * This function is used to copy ELF note segment in the 1st kernel
983  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
984  * size of the buffer @notes_buf is equal to or larger than sum of the
985  * real ELF note segment headers and data.
986  *
987  * It is assumed that program headers with PT_NOTE type pointed to by
988  * @ehdr_ptr has already been updated by update_note_header_size_elf32
989  * and each of PT_NOTE program headers has actual ELF note segment
990  * size in its p_memsz member.
991  */
992 static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
993 {
994 	int i, rc=0;
995 	Elf32_Phdr *phdr_ptr;
996 
997 	phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
998 
999 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1000 		u64 offset;
1001 		if (phdr_ptr->p_type != PT_NOTE)
1002 			continue;
1003 		offset = phdr_ptr->p_offset;
1004 		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
1005 					   &offset);
1006 		if (rc < 0)
1007 			return rc;
1008 		notes_buf += phdr_ptr->p_memsz;
1009 	}
1010 
1011 	return 0;
1012 }
1013 
1014 /* Merges all the PT_NOTE headers into one. */
1015 static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
1016 					   char **notes_buf, size_t *notes_sz)
1017 {
1018 	int i, nr_ptnote=0, rc=0;
1019 	char *tmp;
1020 	Elf32_Ehdr *ehdr_ptr;
1021 	Elf32_Phdr phdr;
1022 	u64 phdr_sz = 0, note_off;
1023 
1024 	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1025 
1026 	rc = update_note_header_size_elf32(ehdr_ptr);
1027 	if (rc < 0)
1028 		return rc;
1029 
1030 	rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
1031 	if (rc < 0)
1032 		return rc;
1033 
1034 	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
1035 	*notes_buf = vmcore_alloc_buf(*notes_sz);
1036 	if (!*notes_buf)
1037 		return -ENOMEM;
1038 
1039 	rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
1040 	if (rc < 0)
1041 		return rc;
1042 
1043 	/* Prepare merged PT_NOTE program header. */
1044 	phdr.p_type    = PT_NOTE;
1045 	phdr.p_flags   = 0;
1046 	note_off = sizeof(Elf32_Ehdr) +
1047 			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
1048 	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
1049 	phdr.p_vaddr   = phdr.p_paddr = 0;
1050 	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
1051 	phdr.p_align   = 0;
1052 
1053 	/* Add merged PT_NOTE program header*/
1054 	tmp = elfptr + sizeof(Elf32_Ehdr);
1055 	memcpy(tmp, &phdr, sizeof(phdr));
1056 	tmp += sizeof(phdr);
1057 
1058 	/* Remove unwanted PT_NOTE program headers. */
1059 	i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
1060 	*elfsz = *elfsz - i;
1061 	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
1062 	memset(elfptr + *elfsz, 0, i);
1063 	*elfsz = roundup(*elfsz, PAGE_SIZE);
1064 
1065 	/* Modify e_phnum to reflect merged headers. */
1066 	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
1067 
1068 	/* Store the size of all notes.  We need this to update the note
1069 	 * header when the device dumps will be added.
1070 	 */
1071 	elfnotes_orig_sz = phdr.p_memsz;
1072 
1073 	return 0;
1074 }
1075 
1076 /* Add memory chunks represented by program headers to vmcore list. Also update
1077  * the new offset fields of exported program headers. */
1078 static int __init process_ptload_program_headers_elf64(char *elfptr,
1079 						size_t elfsz,
1080 						size_t elfnotes_sz,
1081 						struct list_head *vc_list)
1082 {
1083 	int i;
1084 	Elf64_Ehdr *ehdr_ptr;
1085 	Elf64_Phdr *phdr_ptr;
1086 	loff_t vmcore_off;
1087 	struct vmcore *new;
1088 
1089 	ehdr_ptr = (Elf64_Ehdr *)elfptr;
1090 	phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
1091 
1092 	/* Skip Elf header, program headers and Elf note segment. */
1093 	vmcore_off = elfsz + elfnotes_sz;
1094 
1095 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1096 		u64 paddr, start, end, size;
1097 
1098 		if (phdr_ptr->p_type != PT_LOAD)
1099 			continue;
1100 
1101 		paddr = phdr_ptr->p_offset;
1102 		start = rounddown(paddr, PAGE_SIZE);
1103 		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1104 		size = end - start;
1105 
1106 		/* Add this contiguous chunk of memory to vmcore list.*/
1107 		new = get_new_element();
1108 		if (!new)
1109 			return -ENOMEM;
1110 		new->paddr = start;
1111 		new->size = size;
1112 		list_add_tail(&new->list, vc_list);
1113 
1114 		/* Update the program header offset. */
1115 		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1116 		vmcore_off = vmcore_off + size;
1117 	}
1118 	return 0;
1119 }
1120 
1121 static int __init process_ptload_program_headers_elf32(char *elfptr,
1122 						size_t elfsz,
1123 						size_t elfnotes_sz,
1124 						struct list_head *vc_list)
1125 {
1126 	int i;
1127 	Elf32_Ehdr *ehdr_ptr;
1128 	Elf32_Phdr *phdr_ptr;
1129 	loff_t vmcore_off;
1130 	struct vmcore *new;
1131 
1132 	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1133 	phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
1134 
1135 	/* Skip Elf header, program headers and Elf note segment. */
1136 	vmcore_off = elfsz + elfnotes_sz;
1137 
1138 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1139 		u64 paddr, start, end, size;
1140 
1141 		if (phdr_ptr->p_type != PT_LOAD)
1142 			continue;
1143 
1144 		paddr = phdr_ptr->p_offset;
1145 		start = rounddown(paddr, PAGE_SIZE);
1146 		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1147 		size = end - start;
1148 
1149 		/* Add this contiguous chunk of memory to vmcore list.*/
1150 		new = get_new_element();
1151 		if (!new)
1152 			return -ENOMEM;
1153 		new->paddr = start;
1154 		new->size = size;
1155 		list_add_tail(&new->list, vc_list);
1156 
1157 		/* Update the program header offset */
1158 		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1159 		vmcore_off = vmcore_off + size;
1160 	}
1161 	return 0;
1162 }
1163 
1164 /* Sets offset fields of vmcore elements. */
1165 static void set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
1166 				    struct list_head *vc_list)
1167 {
1168 	loff_t vmcore_off;
1169 	struct vmcore *m;
1170 
1171 	/* Skip Elf header, program headers and Elf note segment. */
1172 	vmcore_off = elfsz + elfnotes_sz;
1173 
1174 	list_for_each_entry(m, vc_list, list) {
1175 		m->offset = vmcore_off;
1176 		vmcore_off += m->size;
1177 	}
1178 }
1179 
1180 static void free_elfcorebuf(void)
1181 {
1182 	free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
1183 	elfcorebuf = NULL;
1184 	vfree(elfnotes_buf);
1185 	elfnotes_buf = NULL;
1186 }
1187 
1188 static int __init parse_crash_elf64_headers(void)
1189 {
1190 	int rc=0;
1191 	Elf64_Ehdr ehdr;
1192 	u64 addr;
1193 
1194 	addr = elfcorehdr_addr;
1195 
1196 	/* Read Elf header */
1197 	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
1198 	if (rc < 0)
1199 		return rc;
1200 
1201 	/* Do some basic Verification. */
1202 	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1203 		(ehdr.e_type != ET_CORE) ||
1204 		!vmcore_elf64_check_arch(&ehdr) ||
1205 		ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
1206 		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1207 		ehdr.e_version != EV_CURRENT ||
1208 		ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
1209 		ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
1210 		ehdr.e_phnum == 0) {
1211 		pr_warn("Warning: Core image elf header is not sane\n");
1212 		return -EINVAL;
1213 	}
1214 
1215 	/* Read in all elf headers. */
1216 	elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
1217 				ehdr.e_phnum * sizeof(Elf64_Phdr);
1218 	elfcorebuf_sz = elfcorebuf_sz_orig;
1219 	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1220 					      get_order(elfcorebuf_sz_orig));
1221 	if (!elfcorebuf)
1222 		return -ENOMEM;
1223 	addr = elfcorehdr_addr;
1224 	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1225 	if (rc < 0)
1226 		goto fail;
1227 
1228 	/* Merge all PT_NOTE headers into one. */
1229 	rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
1230 				      &elfnotes_buf, &elfnotes_sz);
1231 	if (rc)
1232 		goto fail;
1233 	rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
1234 						  elfnotes_sz, &vmcore_list);
1235 	if (rc)
1236 		goto fail;
1237 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1238 	return 0;
1239 fail:
1240 	free_elfcorebuf();
1241 	return rc;
1242 }
1243 
1244 static int __init parse_crash_elf32_headers(void)
1245 {
1246 	int rc=0;
1247 	Elf32_Ehdr ehdr;
1248 	u64 addr;
1249 
1250 	addr = elfcorehdr_addr;
1251 
1252 	/* Read Elf header */
1253 	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
1254 	if (rc < 0)
1255 		return rc;
1256 
1257 	/* Do some basic Verification. */
1258 	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1259 		(ehdr.e_type != ET_CORE) ||
1260 		!vmcore_elf32_check_arch(&ehdr) ||
1261 		ehdr.e_ident[EI_CLASS] != ELFCLASS32||
1262 		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1263 		ehdr.e_version != EV_CURRENT ||
1264 		ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1265 		ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1266 		ehdr.e_phnum == 0) {
1267 		pr_warn("Warning: Core image elf header is not sane\n");
1268 		return -EINVAL;
1269 	}
1270 
1271 	/* Read in all elf headers. */
1272 	elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1273 	elfcorebuf_sz = elfcorebuf_sz_orig;
1274 	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1275 					      get_order(elfcorebuf_sz_orig));
1276 	if (!elfcorebuf)
1277 		return -ENOMEM;
1278 	addr = elfcorehdr_addr;
1279 	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1280 	if (rc < 0)
1281 		goto fail;
1282 
1283 	/* Merge all PT_NOTE headers into one. */
1284 	rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1285 				      &elfnotes_buf, &elfnotes_sz);
1286 	if (rc)
1287 		goto fail;
1288 	rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1289 						  elfnotes_sz, &vmcore_list);
1290 	if (rc)
1291 		goto fail;
1292 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1293 	return 0;
1294 fail:
1295 	free_elfcorebuf();
1296 	return rc;
1297 }
1298 
1299 static int __init parse_crash_elf_headers(void)
1300 {
1301 	unsigned char e_ident[EI_NIDENT];
1302 	u64 addr;
1303 	int rc=0;
1304 
1305 	addr = elfcorehdr_addr;
1306 	rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1307 	if (rc < 0)
1308 		return rc;
1309 	if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1310 		pr_warn("Warning: Core image elf header not found\n");
1311 		return -EINVAL;
1312 	}
1313 
1314 	if (e_ident[EI_CLASS] == ELFCLASS64) {
1315 		rc = parse_crash_elf64_headers();
1316 		if (rc)
1317 			return rc;
1318 	} else if (e_ident[EI_CLASS] == ELFCLASS32) {
1319 		rc = parse_crash_elf32_headers();
1320 		if (rc)
1321 			return rc;
1322 	} else {
1323 		pr_warn("Warning: Core image elf header is not sane\n");
1324 		return -EINVAL;
1325 	}
1326 
1327 	/* Determine vmcore size. */
1328 	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1329 				      &vmcore_list);
1330 
1331 	return 0;
1332 }
1333 
1334 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1335 /**
1336  * vmcoredd_write_header - Write vmcore device dump header at the
1337  * beginning of the dump's buffer.
1338  * @buf: Output buffer where the note is written
1339  * @data: Dump info
1340  * @size: Size of the dump
1341  *
1342  * Fills beginning of the dump's buffer with vmcore device dump header.
1343  */
1344 static void vmcoredd_write_header(void *buf, struct vmcoredd_data *data,
1345 				  u32 size)
1346 {
1347 	struct vmcoredd_header *vdd_hdr = (struct vmcoredd_header *)buf;
1348 
1349 	vdd_hdr->n_namesz = sizeof(vdd_hdr->name);
1350 	vdd_hdr->n_descsz = size + sizeof(vdd_hdr->dump_name);
1351 	vdd_hdr->n_type = NT_VMCOREDD;
1352 
1353 	strncpy((char *)vdd_hdr->name, VMCOREDD_NOTE_NAME,
1354 		sizeof(vdd_hdr->name));
1355 	memcpy(vdd_hdr->dump_name, data->dump_name, sizeof(vdd_hdr->dump_name));
1356 }
1357 
1358 /**
1359  * vmcoredd_update_program_headers - Update all Elf program headers
1360  * @elfptr: Pointer to elf header
1361  * @elfnotesz: Size of elf notes aligned to page size
1362  * @vmcoreddsz: Size of device dumps to be added to elf note header
1363  *
1364  * Determine type of Elf header (Elf64 or Elf32) and update the elf note size.
1365  * Also update the offsets of all the program headers after the elf note header.
1366  */
1367 static void vmcoredd_update_program_headers(char *elfptr, size_t elfnotesz,
1368 					    size_t vmcoreddsz)
1369 {
1370 	unsigned char *e_ident = (unsigned char *)elfptr;
1371 	u64 start, end, size;
1372 	loff_t vmcore_off;
1373 	u32 i;
1374 
1375 	vmcore_off = elfcorebuf_sz + elfnotesz;
1376 
1377 	if (e_ident[EI_CLASS] == ELFCLASS64) {
1378 		Elf64_Ehdr *ehdr = (Elf64_Ehdr *)elfptr;
1379 		Elf64_Phdr *phdr = (Elf64_Phdr *)(elfptr + sizeof(Elf64_Ehdr));
1380 
1381 		/* Update all program headers */
1382 		for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1383 			if (phdr->p_type == PT_NOTE) {
1384 				/* Update note size */
1385 				phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1386 				phdr->p_filesz = phdr->p_memsz;
1387 				continue;
1388 			}
1389 
1390 			start = rounddown(phdr->p_offset, PAGE_SIZE);
1391 			end = roundup(phdr->p_offset + phdr->p_memsz,
1392 				      PAGE_SIZE);
1393 			size = end - start;
1394 			phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1395 			vmcore_off += size;
1396 		}
1397 	} else {
1398 		Elf32_Ehdr *ehdr = (Elf32_Ehdr *)elfptr;
1399 		Elf32_Phdr *phdr = (Elf32_Phdr *)(elfptr + sizeof(Elf32_Ehdr));
1400 
1401 		/* Update all program headers */
1402 		for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1403 			if (phdr->p_type == PT_NOTE) {
1404 				/* Update note size */
1405 				phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1406 				phdr->p_filesz = phdr->p_memsz;
1407 				continue;
1408 			}
1409 
1410 			start = rounddown(phdr->p_offset, PAGE_SIZE);
1411 			end = roundup(phdr->p_offset + phdr->p_memsz,
1412 				      PAGE_SIZE);
1413 			size = end - start;
1414 			phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1415 			vmcore_off += size;
1416 		}
1417 	}
1418 }
1419 
1420 /**
1421  * vmcoredd_update_size - Update the total size of the device dumps and update
1422  * Elf header
1423  * @dump_size: Size of the current device dump to be added to total size
1424  *
1425  * Update the total size of all the device dumps and update the Elf program
1426  * headers. Calculate the new offsets for the vmcore list and update the
1427  * total vmcore size.
1428  */
1429 static void vmcoredd_update_size(size_t dump_size)
1430 {
1431 	vmcoredd_orig_sz += dump_size;
1432 	elfnotes_sz = roundup(elfnotes_orig_sz, PAGE_SIZE) + vmcoredd_orig_sz;
1433 	vmcoredd_update_program_headers(elfcorebuf, elfnotes_sz,
1434 					vmcoredd_orig_sz);
1435 
1436 	/* Update vmcore list offsets */
1437 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1438 
1439 	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1440 				      &vmcore_list);
1441 	proc_vmcore->size = vmcore_size;
1442 }
1443 
1444 /**
1445  * vmcore_add_device_dump - Add a buffer containing device dump to vmcore
1446  * @data: dump info.
1447  *
1448  * Allocate a buffer and invoke the calling driver's dump collect routine.
1449  * Write Elf note at the beginning of the buffer to indicate vmcore device
1450  * dump and add the dump to global list.
1451  */
1452 int vmcore_add_device_dump(struct vmcoredd_data *data)
1453 {
1454 	struct vmcoredd_node *dump;
1455 	void *buf = NULL;
1456 	size_t data_size;
1457 	int ret;
1458 
1459 	if (vmcoredd_disabled) {
1460 		pr_err_once("Device dump is disabled\n");
1461 		return -EINVAL;
1462 	}
1463 
1464 	if (!data || !strlen(data->dump_name) ||
1465 	    !data->vmcoredd_callback || !data->size)
1466 		return -EINVAL;
1467 
1468 	dump = vzalloc(sizeof(*dump));
1469 	if (!dump) {
1470 		ret = -ENOMEM;
1471 		goto out_err;
1472 	}
1473 
1474 	/* Keep size of the buffer page aligned so that it can be mmaped */
1475 	data_size = roundup(sizeof(struct vmcoredd_header) + data->size,
1476 			    PAGE_SIZE);
1477 
1478 	/* Allocate buffer for driver's to write their dumps */
1479 	buf = vmcore_alloc_buf(data_size);
1480 	if (!buf) {
1481 		ret = -ENOMEM;
1482 		goto out_err;
1483 	}
1484 
1485 	vmcoredd_write_header(buf, data, data_size -
1486 			      sizeof(struct vmcoredd_header));
1487 
1488 	/* Invoke the driver's dump collection routing */
1489 	ret = data->vmcoredd_callback(data, buf +
1490 				      sizeof(struct vmcoredd_header));
1491 	if (ret)
1492 		goto out_err;
1493 
1494 	dump->buf = buf;
1495 	dump->size = data_size;
1496 
1497 	/* Add the dump to driver sysfs list */
1498 	mutex_lock(&vmcoredd_mutex);
1499 	list_add_tail(&dump->list, &vmcoredd_list);
1500 	mutex_unlock(&vmcoredd_mutex);
1501 
1502 	vmcoredd_update_size(data_size);
1503 	return 0;
1504 
1505 out_err:
1506 	if (buf)
1507 		vfree(buf);
1508 
1509 	if (dump)
1510 		vfree(dump);
1511 
1512 	return ret;
1513 }
1514 EXPORT_SYMBOL(vmcore_add_device_dump);
1515 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1516 
1517 /* Free all dumps in vmcore device dump list */
1518 static void vmcore_free_device_dumps(void)
1519 {
1520 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1521 	mutex_lock(&vmcoredd_mutex);
1522 	while (!list_empty(&vmcoredd_list)) {
1523 		struct vmcoredd_node *dump;
1524 
1525 		dump = list_first_entry(&vmcoredd_list, struct vmcoredd_node,
1526 					list);
1527 		list_del(&dump->list);
1528 		vfree(dump->buf);
1529 		vfree(dump);
1530 	}
1531 	mutex_unlock(&vmcoredd_mutex);
1532 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1533 }
1534 
1535 /* Init function for vmcore module. */
1536 static int __init vmcore_init(void)
1537 {
1538 	int rc = 0;
1539 
1540 	/* Allow architectures to allocate ELF header in 2nd kernel */
1541 	rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1542 	if (rc)
1543 		return rc;
1544 	/*
1545 	 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1546 	 * then capture the dump.
1547 	 */
1548 	if (!(is_vmcore_usable()))
1549 		return rc;
1550 	rc = parse_crash_elf_headers();
1551 	if (rc) {
1552 		pr_warn("Kdump: vmcore not initialized\n");
1553 		return rc;
1554 	}
1555 	elfcorehdr_free(elfcorehdr_addr);
1556 	elfcorehdr_addr = ELFCORE_ADDR_ERR;
1557 
1558 	proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &proc_vmcore_operations);
1559 	if (proc_vmcore)
1560 		proc_vmcore->size = vmcore_size;
1561 	return 0;
1562 }
1563 fs_initcall(vmcore_init);
1564 
1565 /* Cleanup function for vmcore module. */
1566 void vmcore_cleanup(void)
1567 {
1568 	if (proc_vmcore) {
1569 		proc_remove(proc_vmcore);
1570 		proc_vmcore = NULL;
1571 	}
1572 
1573 	/* clear the vmcore list. */
1574 	while (!list_empty(&vmcore_list)) {
1575 		struct vmcore *m;
1576 
1577 		m = list_first_entry(&vmcore_list, struct vmcore, list);
1578 		list_del(&m->list);
1579 		kfree(m);
1580 	}
1581 	free_elfcorebuf();
1582 
1583 	/* clear vmcore device dump list */
1584 	vmcore_free_device_dumps();
1585 }
1586