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