1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *	fs/proc/vmcore.c Interface for accessing the crash
4  * 				 dump from the system's previous life.
5  * 	Heavily borrowed from fs/proc/kcore.c
6  *	Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
7  *	Copyright (C) IBM Corporation, 2004. All rights reserved
8  *
9  */
10 
11 #include <linux/mm.h>
12 #include <linux/kcore.h>
13 #include <linux/user.h>
14 #include <linux/elf.h>
15 #include <linux/elfcore.h>
16 #include <linux/export.h>
17 #include <linux/slab.h>
18 #include <linux/highmem.h>
19 #include <linux/printk.h>
20 #include <linux/memblock.h>
21 #include <linux/init.h>
22 #include <linux/crash_dump.h>
23 #include <linux/list.h>
24 #include <linux/moduleparam.h>
25 #include <linux/mutex.h>
26 #include <linux/vmalloc.h>
27 #include <linux/pagemap.h>
28 #include <linux/uio.h>
29 #include <linux/cc_platform.h>
30 #include <asm/io.h>
31 #include "internal.h"
32 
33 /* List representing chunks of contiguous memory areas and their offsets in
34  * vmcore file.
35  */
36 static LIST_HEAD(vmcore_list);
37 
38 /* Stores the pointer to the buffer containing kernel elf core headers. */
39 static char *elfcorebuf;
40 static size_t elfcorebuf_sz;
41 static size_t elfcorebuf_sz_orig;
42 
43 static char *elfnotes_buf;
44 static size_t elfnotes_sz;
45 /* Size of all notes minus the device dump notes */
46 static size_t elfnotes_orig_sz;
47 
48 /* Total size of vmcore file. */
49 static u64 vmcore_size;
50 
51 static struct proc_dir_entry *proc_vmcore;
52 
53 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
54 /* Device Dump list and mutex to synchronize access to list */
55 static LIST_HEAD(vmcoredd_list);
56 static DEFINE_MUTEX(vmcoredd_mutex);
57 
58 static bool vmcoredd_disabled;
59 core_param(novmcoredd, vmcoredd_disabled, bool, 0);
60 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
61 
62 /* Device Dump Size */
63 static size_t vmcoredd_orig_sz;
64 
65 static DEFINE_SPINLOCK(vmcore_cb_lock);
66 DEFINE_STATIC_SRCU(vmcore_cb_srcu);
67 /* List of registered vmcore callbacks. */
68 static LIST_HEAD(vmcore_cb_list);
69 /* Whether the vmcore has been opened once. */
70 static bool vmcore_opened;
71 
register_vmcore_cb(struct vmcore_cb * cb)72 void register_vmcore_cb(struct vmcore_cb *cb)
73 {
74 	INIT_LIST_HEAD(&cb->next);
75 	spin_lock(&vmcore_cb_lock);
76 	list_add_tail(&cb->next, &vmcore_cb_list);
77 	/*
78 	 * Registering a vmcore callback after the vmcore was opened is
79 	 * very unusual (e.g., manual driver loading).
80 	 */
81 	if (vmcore_opened)
82 		pr_warn_once("Unexpected vmcore callback registration\n");
83 	spin_unlock(&vmcore_cb_lock);
84 }
85 EXPORT_SYMBOL_GPL(register_vmcore_cb);
86 
unregister_vmcore_cb(struct vmcore_cb * cb)87 void unregister_vmcore_cb(struct vmcore_cb *cb)
88 {
89 	spin_lock(&vmcore_cb_lock);
90 	list_del_rcu(&cb->next);
91 	/*
92 	 * Unregistering a vmcore callback after the vmcore was opened is
93 	 * very unusual (e.g., forced driver removal), but we cannot stop
94 	 * unregistering.
95 	 */
96 	if (vmcore_opened)
97 		pr_warn_once("Unexpected vmcore callback unregistration\n");
98 	spin_unlock(&vmcore_cb_lock);
99 
100 	synchronize_srcu(&vmcore_cb_srcu);
101 }
102 EXPORT_SYMBOL_GPL(unregister_vmcore_cb);
103 
pfn_is_ram(unsigned long pfn)104 static bool pfn_is_ram(unsigned long pfn)
105 {
106 	struct vmcore_cb *cb;
107 	bool ret = true;
108 
109 	list_for_each_entry_srcu(cb, &vmcore_cb_list, next,
110 				 srcu_read_lock_held(&vmcore_cb_srcu)) {
111 		if (unlikely(!cb->pfn_is_ram))
112 			continue;
113 		ret = cb->pfn_is_ram(cb, pfn);
114 		if (!ret)
115 			break;
116 	}
117 
118 	return ret;
119 }
120 
open_vmcore(struct inode * inode,struct file * file)121 static int open_vmcore(struct inode *inode, struct file *file)
122 {
123 	spin_lock(&vmcore_cb_lock);
124 	vmcore_opened = true;
125 	spin_unlock(&vmcore_cb_lock);
126 
127 	return 0;
128 }
129 
130 /* Reads a page from the oldmem device from given offset. */
read_from_oldmem(struct iov_iter * iter,size_t count,u64 * ppos,bool encrypted)131 ssize_t read_from_oldmem(struct iov_iter *iter, size_t count,
132 			 u64 *ppos, bool encrypted)
133 {
134 	unsigned long pfn, offset;
135 	ssize_t nr_bytes;
136 	ssize_t read = 0, tmp;
137 	int idx;
138 
139 	if (!count)
140 		return 0;
141 
142 	offset = (unsigned long)(*ppos % PAGE_SIZE);
143 	pfn = (unsigned long)(*ppos / PAGE_SIZE);
144 
145 	idx = srcu_read_lock(&vmcore_cb_srcu);
146 	do {
147 		if (count > (PAGE_SIZE - offset))
148 			nr_bytes = PAGE_SIZE - offset;
149 		else
150 			nr_bytes = count;
151 
152 		/* If pfn is not ram, return zeros for sparse dump files */
153 		if (!pfn_is_ram(pfn)) {
154 			tmp = iov_iter_zero(nr_bytes, iter);
155 		} else {
156 			if (encrypted)
157 				tmp = copy_oldmem_page_encrypted(iter, pfn,
158 								 nr_bytes,
159 								 offset);
160 			else
161 				tmp = copy_oldmem_page(iter, pfn, nr_bytes,
162 						       offset);
163 		}
164 		if (tmp < nr_bytes) {
165 			srcu_read_unlock(&vmcore_cb_srcu, idx);
166 			return -EFAULT;
167 		}
168 
169 		*ppos += nr_bytes;
170 		count -= nr_bytes;
171 		read += nr_bytes;
172 		++pfn;
173 		offset = 0;
174 	} while (count);
175 	srcu_read_unlock(&vmcore_cb_srcu, idx);
176 
177 	return read;
178 }
179 
180 /*
181  * Architectures may override this function to allocate ELF header in 2nd kernel
182  */
elfcorehdr_alloc(unsigned long long * addr,unsigned long long * size)183 int __weak elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
184 {
185 	return 0;
186 }
187 
188 /*
189  * Architectures may override this function to free header
190  */
elfcorehdr_free(unsigned long long addr)191 void __weak elfcorehdr_free(unsigned long long addr)
192 {}
193 
194 /*
195  * Architectures may override this function to read from ELF header
196  */
elfcorehdr_read(char * buf,size_t count,u64 * ppos)197 ssize_t __weak elfcorehdr_read(char *buf, size_t count, u64 *ppos)
198 {
199 	struct kvec kvec = { .iov_base = buf, .iov_len = count };
200 	struct iov_iter iter;
201 
202 	iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, count);
203 
204 	return read_from_oldmem(&iter, count, ppos, false);
205 }
206 
207 /*
208  * Architectures may override this function to read from notes sections
209  */
elfcorehdr_read_notes(char * buf,size_t count,u64 * ppos)210 ssize_t __weak elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
211 {
212 	struct kvec kvec = { .iov_base = buf, .iov_len = count };
213 	struct iov_iter iter;
214 
215 	iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, count);
216 
217 	return read_from_oldmem(&iter, count, ppos,
218 			cc_platform_has(CC_ATTR_MEM_ENCRYPT));
219 }
220 
221 /*
222  * Architectures may override this function to map oldmem
223  */
remap_oldmem_pfn_range(struct vm_area_struct * vma,unsigned long from,unsigned long pfn,unsigned long size,pgprot_t prot)224 int __weak remap_oldmem_pfn_range(struct vm_area_struct *vma,
225 				  unsigned long from, unsigned long pfn,
226 				  unsigned long size, pgprot_t prot)
227 {
228 	prot = pgprot_encrypted(prot);
229 	return remap_pfn_range(vma, from, pfn, size, prot);
230 }
231 
232 /*
233  * Architectures which support memory encryption override this.
234  */
copy_oldmem_page_encrypted(struct iov_iter * iter,unsigned long pfn,size_t csize,unsigned long offset)235 ssize_t __weak copy_oldmem_page_encrypted(struct iov_iter *iter,
236 		unsigned long pfn, size_t csize, unsigned long offset)
237 {
238 	return copy_oldmem_page(iter, pfn, csize, offset);
239 }
240 
241 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
vmcoredd_copy_dumps(struct iov_iter * iter,u64 start,size_t size)242 static int vmcoredd_copy_dumps(struct iov_iter *iter, u64 start, size_t size)
243 {
244 	struct vmcoredd_node *dump;
245 	u64 offset = 0;
246 	int ret = 0;
247 	size_t tsz;
248 	char *buf;
249 
250 	mutex_lock(&vmcoredd_mutex);
251 	list_for_each_entry(dump, &vmcoredd_list, list) {
252 		if (start < offset + dump->size) {
253 			tsz = min(offset + (u64)dump->size - start, (u64)size);
254 			buf = dump->buf + start - offset;
255 			if (copy_to_iter(buf, tsz, iter) < tsz) {
256 				ret = -EFAULT;
257 				goto out_unlock;
258 			}
259 
260 			size -= tsz;
261 			start += tsz;
262 
263 			/* Leave now if buffer filled already */
264 			if (!size)
265 				goto out_unlock;
266 		}
267 		offset += dump->size;
268 	}
269 
270 out_unlock:
271 	mutex_unlock(&vmcoredd_mutex);
272 	return ret;
273 }
274 
275 #ifdef CONFIG_MMU
vmcoredd_mmap_dumps(struct vm_area_struct * vma,unsigned long dst,u64 start,size_t size)276 static int vmcoredd_mmap_dumps(struct vm_area_struct *vma, unsigned long dst,
277 			       u64 start, size_t size)
278 {
279 	struct vmcoredd_node *dump;
280 	u64 offset = 0;
281 	int ret = 0;
282 	size_t tsz;
283 	char *buf;
284 
285 	mutex_lock(&vmcoredd_mutex);
286 	list_for_each_entry(dump, &vmcoredd_list, list) {
287 		if (start < offset + dump->size) {
288 			tsz = min(offset + (u64)dump->size - start, (u64)size);
289 			buf = dump->buf + start - offset;
290 			if (remap_vmalloc_range_partial(vma, dst, buf, 0,
291 							tsz)) {
292 				ret = -EFAULT;
293 				goto out_unlock;
294 			}
295 
296 			size -= tsz;
297 			start += tsz;
298 			dst += tsz;
299 
300 			/* Leave now if buffer filled already */
301 			if (!size)
302 				goto out_unlock;
303 		}
304 		offset += dump->size;
305 	}
306 
307 out_unlock:
308 	mutex_unlock(&vmcoredd_mutex);
309 	return ret;
310 }
311 #endif /* CONFIG_MMU */
312 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
313 
314 /* Read from the ELF header and then the crash dump. On error, negative value is
315  * returned otherwise number of bytes read are returned.
316  */
__read_vmcore(struct iov_iter * iter,loff_t * fpos)317 static ssize_t __read_vmcore(struct iov_iter *iter, loff_t *fpos)
318 {
319 	ssize_t acc = 0, tmp;
320 	size_t tsz;
321 	u64 start;
322 	struct vmcore *m = NULL;
323 
324 	if (!iov_iter_count(iter) || *fpos >= vmcore_size)
325 		return 0;
326 
327 	iov_iter_truncate(iter, vmcore_size - *fpos);
328 
329 	/* Read ELF core header */
330 	if (*fpos < elfcorebuf_sz) {
331 		tsz = min(elfcorebuf_sz - (size_t)*fpos, iov_iter_count(iter));
332 		if (copy_to_iter(elfcorebuf + *fpos, tsz, iter) < tsz)
333 			return -EFAULT;
334 		*fpos += tsz;
335 		acc += tsz;
336 
337 		/* leave now if filled buffer already */
338 		if (!iov_iter_count(iter))
339 			return acc;
340 	}
341 
342 	/* Read ELF note segment */
343 	if (*fpos < elfcorebuf_sz + elfnotes_sz) {
344 		void *kaddr;
345 
346 		/* We add device dumps before other elf notes because the
347 		 * other elf notes may not fill the elf notes buffer
348 		 * completely and we will end up with zero-filled data
349 		 * between the elf notes and the device dumps. Tools will
350 		 * then try to decode this zero-filled data as valid notes
351 		 * and we don't want that. Hence, adding device dumps before
352 		 * the other elf notes ensure that zero-filled data can be
353 		 * avoided.
354 		 */
355 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
356 		/* Read device dumps */
357 		if (*fpos < elfcorebuf_sz + vmcoredd_orig_sz) {
358 			tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
359 				  (size_t)*fpos, iov_iter_count(iter));
360 			start = *fpos - elfcorebuf_sz;
361 			if (vmcoredd_copy_dumps(iter, start, tsz))
362 				return -EFAULT;
363 
364 			*fpos += tsz;
365 			acc += tsz;
366 
367 			/* leave now if filled buffer already */
368 			if (!iov_iter_count(iter))
369 				return acc;
370 		}
371 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
372 
373 		/* Read remaining elf notes */
374 		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)*fpos,
375 			  iov_iter_count(iter));
376 		kaddr = elfnotes_buf + *fpos - elfcorebuf_sz - vmcoredd_orig_sz;
377 		if (copy_to_iter(kaddr, tsz, iter) < tsz)
378 			return -EFAULT;
379 
380 		*fpos += tsz;
381 		acc += tsz;
382 
383 		/* leave now if filled buffer already */
384 		if (!iov_iter_count(iter))
385 			return acc;
386 	}
387 
388 	list_for_each_entry(m, &vmcore_list, list) {
389 		if (*fpos < m->offset + m->size) {
390 			tsz = (size_t)min_t(unsigned long long,
391 					    m->offset + m->size - *fpos,
392 					    iov_iter_count(iter));
393 			start = m->paddr + *fpos - m->offset;
394 			tmp = read_from_oldmem(iter, tsz, &start,
395 					cc_platform_has(CC_ATTR_MEM_ENCRYPT));
396 			if (tmp < 0)
397 				return tmp;
398 			*fpos += tsz;
399 			acc += tsz;
400 
401 			/* leave now if filled buffer already */
402 			if (!iov_iter_count(iter))
403 				return acc;
404 		}
405 	}
406 
407 	return acc;
408 }
409 
read_vmcore(struct kiocb * iocb,struct iov_iter * iter)410 static ssize_t read_vmcore(struct kiocb *iocb, struct iov_iter *iter)
411 {
412 	return __read_vmcore(iter, &iocb->ki_pos);
413 }
414 
415 /*
416  * The vmcore fault handler uses the page cache and fills data using the
417  * standard __read_vmcore() function.
418  *
419  * On s390 the fault handler is used for memory regions that can't be mapped
420  * directly with remap_pfn_range().
421  */
mmap_vmcore_fault(struct vm_fault * vmf)422 static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
423 {
424 #ifdef CONFIG_S390
425 	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
426 	pgoff_t index = vmf->pgoff;
427 	struct iov_iter iter;
428 	struct kvec kvec;
429 	struct page *page;
430 	loff_t offset;
431 	int rc;
432 
433 	page = find_or_create_page(mapping, index, GFP_KERNEL);
434 	if (!page)
435 		return VM_FAULT_OOM;
436 	if (!PageUptodate(page)) {
437 		offset = (loff_t) index << PAGE_SHIFT;
438 		kvec.iov_base = page_address(page);
439 		kvec.iov_len = PAGE_SIZE;
440 		iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, PAGE_SIZE);
441 
442 		rc = __read_vmcore(&iter, &offset);
443 		if (rc < 0) {
444 			unlock_page(page);
445 			put_page(page);
446 			return vmf_error(rc);
447 		}
448 		SetPageUptodate(page);
449 	}
450 	unlock_page(page);
451 	vmf->page = page;
452 	return 0;
453 #else
454 	return VM_FAULT_SIGBUS;
455 #endif
456 }
457 
458 static const struct vm_operations_struct vmcore_mmap_ops = {
459 	.fault = mmap_vmcore_fault,
460 };
461 
462 /**
463  * vmcore_alloc_buf - allocate buffer in vmalloc memory
464  * @size: size of buffer
465  *
466  * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
467  * the buffer to user-space by means of remap_vmalloc_range().
468  *
469  * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
470  * disabled and there's no need to allow users to mmap the buffer.
471  */
vmcore_alloc_buf(size_t size)472 static inline char *vmcore_alloc_buf(size_t size)
473 {
474 #ifdef CONFIG_MMU
475 	return vmalloc_user(size);
476 #else
477 	return vzalloc(size);
478 #endif
479 }
480 
481 /*
482  * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
483  * essential for mmap_vmcore() in order to map physically
484  * non-contiguous objects (ELF header, ELF note segment and memory
485  * regions in the 1st kernel pointed to by PT_LOAD entries) into
486  * virtually contiguous user-space in ELF layout.
487  */
488 #ifdef CONFIG_MMU
489 /*
490  * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
491  * reported as not being ram with the zero page.
492  *
493  * @vma: vm_area_struct describing requested mapping
494  * @from: start remapping from
495  * @pfn: page frame number to start remapping to
496  * @size: remapping size
497  * @prot: protection bits
498  *
499  * Returns zero on success, -EAGAIN on failure.
500  */
remap_oldmem_pfn_checked(struct vm_area_struct * vma,unsigned long from,unsigned long pfn,unsigned long size,pgprot_t prot)501 static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
502 				    unsigned long from, unsigned long pfn,
503 				    unsigned long size, pgprot_t prot)
504 {
505 	unsigned long map_size;
506 	unsigned long pos_start, pos_end, pos;
507 	unsigned long zeropage_pfn = my_zero_pfn(0);
508 	size_t len = 0;
509 
510 	pos_start = pfn;
511 	pos_end = pfn + (size >> PAGE_SHIFT);
512 
513 	for (pos = pos_start; pos < pos_end; ++pos) {
514 		if (!pfn_is_ram(pos)) {
515 			/*
516 			 * We hit a page which is not ram. Remap the continuous
517 			 * region between pos_start and pos-1 and replace
518 			 * the non-ram page at pos with the zero page.
519 			 */
520 			if (pos > pos_start) {
521 				/* Remap continuous region */
522 				map_size = (pos - pos_start) << PAGE_SHIFT;
523 				if (remap_oldmem_pfn_range(vma, from + len,
524 							   pos_start, map_size,
525 							   prot))
526 					goto fail;
527 				len += map_size;
528 			}
529 			/* Remap the zero page */
530 			if (remap_oldmem_pfn_range(vma, from + len,
531 						   zeropage_pfn,
532 						   PAGE_SIZE, prot))
533 				goto fail;
534 			len += PAGE_SIZE;
535 			pos_start = pos + 1;
536 		}
537 	}
538 	if (pos > pos_start) {
539 		/* Remap the rest */
540 		map_size = (pos - pos_start) << PAGE_SHIFT;
541 		if (remap_oldmem_pfn_range(vma, from + len, pos_start,
542 					   map_size, prot))
543 			goto fail;
544 	}
545 	return 0;
546 fail:
547 	do_munmap(vma->vm_mm, from, len, NULL);
548 	return -EAGAIN;
549 }
550 
vmcore_remap_oldmem_pfn(struct vm_area_struct * vma,unsigned long from,unsigned long pfn,unsigned long size,pgprot_t prot)551 static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
552 			    unsigned long from, unsigned long pfn,
553 			    unsigned long size, pgprot_t prot)
554 {
555 	int ret, idx;
556 
557 	/*
558 	 * Check if a callback was registered to avoid looping over all
559 	 * pages without a reason.
560 	 */
561 	idx = srcu_read_lock(&vmcore_cb_srcu);
562 	if (!list_empty(&vmcore_cb_list))
563 		ret = remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
564 	else
565 		ret = remap_oldmem_pfn_range(vma, from, pfn, size, prot);
566 	srcu_read_unlock(&vmcore_cb_srcu, idx);
567 	return ret;
568 }
569 
mmap_vmcore(struct file * file,struct vm_area_struct * vma)570 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
571 {
572 	size_t size = vma->vm_end - vma->vm_start;
573 	u64 start, end, len, tsz;
574 	struct vmcore *m;
575 
576 	start = (u64)vma->vm_pgoff << PAGE_SHIFT;
577 	end = start + size;
578 
579 	if (size > vmcore_size || end > vmcore_size)
580 		return -EINVAL;
581 
582 	if (vma->vm_flags & (VM_WRITE | VM_EXEC))
583 		return -EPERM;
584 
585 	vm_flags_mod(vma, VM_MIXEDMAP, VM_MAYWRITE | VM_MAYEXEC);
586 	vma->vm_ops = &vmcore_mmap_ops;
587 
588 	len = 0;
589 
590 	if (start < elfcorebuf_sz) {
591 		u64 pfn;
592 
593 		tsz = min(elfcorebuf_sz - (size_t)start, size);
594 		pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
595 		if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
596 				    vma->vm_page_prot))
597 			return -EAGAIN;
598 		size -= tsz;
599 		start += tsz;
600 		len += tsz;
601 
602 		if (size == 0)
603 			return 0;
604 	}
605 
606 	if (start < elfcorebuf_sz + elfnotes_sz) {
607 		void *kaddr;
608 
609 		/* We add device dumps before other elf notes because the
610 		 * other elf notes may not fill the elf notes buffer
611 		 * completely and we will end up with zero-filled data
612 		 * between the elf notes and the device dumps. Tools will
613 		 * then try to decode this zero-filled data as valid notes
614 		 * and we don't want that. Hence, adding device dumps before
615 		 * the other elf notes ensure that zero-filled data can be
616 		 * avoided. This also ensures that the device dumps and
617 		 * other elf notes can be properly mmaped at page aligned
618 		 * address.
619 		 */
620 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
621 		/* Read device dumps */
622 		if (start < elfcorebuf_sz + vmcoredd_orig_sz) {
623 			u64 start_off;
624 
625 			tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
626 				  (size_t)start, size);
627 			start_off = start - elfcorebuf_sz;
628 			if (vmcoredd_mmap_dumps(vma, vma->vm_start + len,
629 						start_off, tsz))
630 				goto fail;
631 
632 			size -= tsz;
633 			start += tsz;
634 			len += tsz;
635 
636 			/* leave now if filled buffer already */
637 			if (!size)
638 				return 0;
639 		}
640 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
641 
642 		/* Read remaining elf notes */
643 		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
644 		kaddr = elfnotes_buf + start - elfcorebuf_sz - vmcoredd_orig_sz;
645 		if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
646 						kaddr, 0, tsz))
647 			goto fail;
648 
649 		size -= tsz;
650 		start += tsz;
651 		len += tsz;
652 
653 		if (size == 0)
654 			return 0;
655 	}
656 
657 	list_for_each_entry(m, &vmcore_list, list) {
658 		if (start < m->offset + m->size) {
659 			u64 paddr = 0;
660 
661 			tsz = (size_t)min_t(unsigned long long,
662 					    m->offset + m->size - start, size);
663 			paddr = m->paddr + start - m->offset;
664 			if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
665 						    paddr >> PAGE_SHIFT, tsz,
666 						    vma->vm_page_prot))
667 				goto fail;
668 			size -= tsz;
669 			start += tsz;
670 			len += tsz;
671 
672 			if (size == 0)
673 				return 0;
674 		}
675 	}
676 
677 	return 0;
678 fail:
679 	do_munmap(vma->vm_mm, vma->vm_start, len, NULL);
680 	return -EAGAIN;
681 }
682 #else
mmap_vmcore(struct file * file,struct vm_area_struct * vma)683 static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
684 {
685 	return -ENOSYS;
686 }
687 #endif
688 
689 static const struct proc_ops vmcore_proc_ops = {
690 	.proc_open	= open_vmcore,
691 	.proc_read_iter	= read_vmcore,
692 	.proc_lseek	= default_llseek,
693 	.proc_mmap	= mmap_vmcore,
694 };
695 
get_new_element(void)696 static struct vmcore* __init get_new_element(void)
697 {
698 	return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
699 }
700 
get_vmcore_size(size_t elfsz,size_t elfnotesegsz,struct list_head * vc_list)701 static u64 get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
702 			   struct list_head *vc_list)
703 {
704 	u64 size;
705 	struct vmcore *m;
706 
707 	size = elfsz + elfnotesegsz;
708 	list_for_each_entry(m, vc_list, list) {
709 		size += m->size;
710 	}
711 	return size;
712 }
713 
714 /**
715  * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
716  *
717  * @ehdr_ptr: ELF header
718  *
719  * This function updates p_memsz member of each PT_NOTE entry in the
720  * program header table pointed to by @ehdr_ptr to real size of ELF
721  * note segment.
722  */
update_note_header_size_elf64(const Elf64_Ehdr * ehdr_ptr)723 static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
724 {
725 	int i, rc=0;
726 	Elf64_Phdr *phdr_ptr;
727 	Elf64_Nhdr *nhdr_ptr;
728 
729 	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
730 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
731 		void *notes_section;
732 		u64 offset, max_sz, sz, real_sz = 0;
733 		if (phdr_ptr->p_type != PT_NOTE)
734 			continue;
735 		max_sz = phdr_ptr->p_memsz;
736 		offset = phdr_ptr->p_offset;
737 		notes_section = kmalloc(max_sz, GFP_KERNEL);
738 		if (!notes_section)
739 			return -ENOMEM;
740 		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
741 		if (rc < 0) {
742 			kfree(notes_section);
743 			return rc;
744 		}
745 		nhdr_ptr = notes_section;
746 		while (nhdr_ptr->n_namesz != 0) {
747 			sz = sizeof(Elf64_Nhdr) +
748 				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
749 				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
750 			if ((real_sz + sz) > max_sz) {
751 				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
752 					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
753 				break;
754 			}
755 			real_sz += sz;
756 			nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
757 		}
758 		kfree(notes_section);
759 		phdr_ptr->p_memsz = real_sz;
760 		if (real_sz == 0) {
761 			pr_warn("Warning: Zero PT_NOTE entries found\n");
762 		}
763 	}
764 
765 	return 0;
766 }
767 
768 /**
769  * get_note_number_and_size_elf64 - get the number of PT_NOTE program
770  * headers and sum of real size of their ELF note segment headers and
771  * data.
772  *
773  * @ehdr_ptr: ELF header
774  * @nr_ptnote: buffer for the number of PT_NOTE program headers
775  * @sz_ptnote: buffer for size of unique PT_NOTE program header
776  *
777  * This function is used to merge multiple PT_NOTE program headers
778  * into a unique single one. The resulting unique entry will have
779  * @sz_ptnote in its phdr->p_mem.
780  *
781  * It is assumed that program headers with PT_NOTE type pointed to by
782  * @ehdr_ptr has already been updated by update_note_header_size_elf64
783  * and each of PT_NOTE program headers has actual ELF note segment
784  * size in its p_memsz member.
785  */
get_note_number_and_size_elf64(const Elf64_Ehdr * ehdr_ptr,int * nr_ptnote,u64 * sz_ptnote)786 static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
787 						 int *nr_ptnote, u64 *sz_ptnote)
788 {
789 	int i;
790 	Elf64_Phdr *phdr_ptr;
791 
792 	*nr_ptnote = *sz_ptnote = 0;
793 
794 	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
795 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
796 		if (phdr_ptr->p_type != PT_NOTE)
797 			continue;
798 		*nr_ptnote += 1;
799 		*sz_ptnote += phdr_ptr->p_memsz;
800 	}
801 
802 	return 0;
803 }
804 
805 /**
806  * copy_notes_elf64 - copy ELF note segments in a given buffer
807  *
808  * @ehdr_ptr: ELF header
809  * @notes_buf: buffer into which ELF note segments are copied
810  *
811  * This function is used to copy ELF note segment in the 1st kernel
812  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
813  * size of the buffer @notes_buf is equal to or larger than sum of the
814  * real ELF note segment headers and data.
815  *
816  * It is assumed that program headers with PT_NOTE type pointed to by
817  * @ehdr_ptr has already been updated by update_note_header_size_elf64
818  * and each of PT_NOTE program headers has actual ELF note segment
819  * size in its p_memsz member.
820  */
copy_notes_elf64(const Elf64_Ehdr * ehdr_ptr,char * notes_buf)821 static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
822 {
823 	int i, rc=0;
824 	Elf64_Phdr *phdr_ptr;
825 
826 	phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
827 
828 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
829 		u64 offset;
830 		if (phdr_ptr->p_type != PT_NOTE)
831 			continue;
832 		offset = phdr_ptr->p_offset;
833 		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
834 					   &offset);
835 		if (rc < 0)
836 			return rc;
837 		notes_buf += phdr_ptr->p_memsz;
838 	}
839 
840 	return 0;
841 }
842 
843 /* Merges all the PT_NOTE headers into one. */
merge_note_headers_elf64(char * elfptr,size_t * elfsz,char ** notes_buf,size_t * notes_sz)844 static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
845 					   char **notes_buf, size_t *notes_sz)
846 {
847 	int i, nr_ptnote=0, rc=0;
848 	char *tmp;
849 	Elf64_Ehdr *ehdr_ptr;
850 	Elf64_Phdr phdr;
851 	u64 phdr_sz = 0, note_off;
852 
853 	ehdr_ptr = (Elf64_Ehdr *)elfptr;
854 
855 	rc = update_note_header_size_elf64(ehdr_ptr);
856 	if (rc < 0)
857 		return rc;
858 
859 	rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
860 	if (rc < 0)
861 		return rc;
862 
863 	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
864 	*notes_buf = vmcore_alloc_buf(*notes_sz);
865 	if (!*notes_buf)
866 		return -ENOMEM;
867 
868 	rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
869 	if (rc < 0)
870 		return rc;
871 
872 	/* Prepare merged PT_NOTE program header. */
873 	phdr.p_type    = PT_NOTE;
874 	phdr.p_flags   = 0;
875 	note_off = sizeof(Elf64_Ehdr) +
876 			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
877 	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
878 	phdr.p_vaddr   = phdr.p_paddr = 0;
879 	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
880 	phdr.p_align   = 4;
881 
882 	/* Add merged PT_NOTE program header*/
883 	tmp = elfptr + sizeof(Elf64_Ehdr);
884 	memcpy(tmp, &phdr, sizeof(phdr));
885 	tmp += sizeof(phdr);
886 
887 	/* Remove unwanted PT_NOTE program headers. */
888 	i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
889 	*elfsz = *elfsz - i;
890 	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
891 	memset(elfptr + *elfsz, 0, i);
892 	*elfsz = roundup(*elfsz, PAGE_SIZE);
893 
894 	/* Modify e_phnum to reflect merged headers. */
895 	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
896 
897 	/* Store the size of all notes.  We need this to update the note
898 	 * header when the device dumps will be added.
899 	 */
900 	elfnotes_orig_sz = phdr.p_memsz;
901 
902 	return 0;
903 }
904 
905 /**
906  * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
907  *
908  * @ehdr_ptr: ELF header
909  *
910  * This function updates p_memsz member of each PT_NOTE entry in the
911  * program header table pointed to by @ehdr_ptr to real size of ELF
912  * note segment.
913  */
update_note_header_size_elf32(const Elf32_Ehdr * ehdr_ptr)914 static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
915 {
916 	int i, rc=0;
917 	Elf32_Phdr *phdr_ptr;
918 	Elf32_Nhdr *nhdr_ptr;
919 
920 	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
921 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
922 		void *notes_section;
923 		u64 offset, max_sz, sz, real_sz = 0;
924 		if (phdr_ptr->p_type != PT_NOTE)
925 			continue;
926 		max_sz = phdr_ptr->p_memsz;
927 		offset = phdr_ptr->p_offset;
928 		notes_section = kmalloc(max_sz, GFP_KERNEL);
929 		if (!notes_section)
930 			return -ENOMEM;
931 		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
932 		if (rc < 0) {
933 			kfree(notes_section);
934 			return rc;
935 		}
936 		nhdr_ptr = notes_section;
937 		while (nhdr_ptr->n_namesz != 0) {
938 			sz = sizeof(Elf32_Nhdr) +
939 				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
940 				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
941 			if ((real_sz + sz) > max_sz) {
942 				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
943 					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
944 				break;
945 			}
946 			real_sz += sz;
947 			nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
948 		}
949 		kfree(notes_section);
950 		phdr_ptr->p_memsz = real_sz;
951 		if (real_sz == 0) {
952 			pr_warn("Warning: Zero PT_NOTE entries found\n");
953 		}
954 	}
955 
956 	return 0;
957 }
958 
959 /**
960  * get_note_number_and_size_elf32 - get the number of PT_NOTE program
961  * headers and sum of real size of their ELF note segment headers and
962  * data.
963  *
964  * @ehdr_ptr: ELF header
965  * @nr_ptnote: buffer for the number of PT_NOTE program headers
966  * @sz_ptnote: buffer for size of unique PT_NOTE program header
967  *
968  * This function is used to merge multiple PT_NOTE program headers
969  * into a unique single one. The resulting unique entry will have
970  * @sz_ptnote in its phdr->p_mem.
971  *
972  * It is assumed that program headers with PT_NOTE type pointed to by
973  * @ehdr_ptr has already been updated by update_note_header_size_elf32
974  * and each of PT_NOTE program headers has actual ELF note segment
975  * size in its p_memsz member.
976  */
get_note_number_and_size_elf32(const Elf32_Ehdr * ehdr_ptr,int * nr_ptnote,u64 * sz_ptnote)977 static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
978 						 int *nr_ptnote, u64 *sz_ptnote)
979 {
980 	int i;
981 	Elf32_Phdr *phdr_ptr;
982 
983 	*nr_ptnote = *sz_ptnote = 0;
984 
985 	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
986 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
987 		if (phdr_ptr->p_type != PT_NOTE)
988 			continue;
989 		*nr_ptnote += 1;
990 		*sz_ptnote += phdr_ptr->p_memsz;
991 	}
992 
993 	return 0;
994 }
995 
996 /**
997  * copy_notes_elf32 - copy ELF note segments in a given buffer
998  *
999  * @ehdr_ptr: ELF header
1000  * @notes_buf: buffer into which ELF note segments are copied
1001  *
1002  * This function is used to copy ELF note segment in the 1st kernel
1003  * into the buffer @notes_buf in the 2nd kernel. It is assumed that
1004  * size of the buffer @notes_buf is equal to or larger than sum of the
1005  * real ELF note segment headers and data.
1006  *
1007  * It is assumed that program headers with PT_NOTE type pointed to by
1008  * @ehdr_ptr has already been updated by update_note_header_size_elf32
1009  * and each of PT_NOTE program headers has actual ELF note segment
1010  * size in its p_memsz member.
1011  */
copy_notes_elf32(const Elf32_Ehdr * ehdr_ptr,char * notes_buf)1012 static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
1013 {
1014 	int i, rc=0;
1015 	Elf32_Phdr *phdr_ptr;
1016 
1017 	phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
1018 
1019 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1020 		u64 offset;
1021 		if (phdr_ptr->p_type != PT_NOTE)
1022 			continue;
1023 		offset = phdr_ptr->p_offset;
1024 		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
1025 					   &offset);
1026 		if (rc < 0)
1027 			return rc;
1028 		notes_buf += phdr_ptr->p_memsz;
1029 	}
1030 
1031 	return 0;
1032 }
1033 
1034 /* Merges all the PT_NOTE headers into one. */
merge_note_headers_elf32(char * elfptr,size_t * elfsz,char ** notes_buf,size_t * notes_sz)1035 static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
1036 					   char **notes_buf, size_t *notes_sz)
1037 {
1038 	int i, nr_ptnote=0, rc=0;
1039 	char *tmp;
1040 	Elf32_Ehdr *ehdr_ptr;
1041 	Elf32_Phdr phdr;
1042 	u64 phdr_sz = 0, note_off;
1043 
1044 	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1045 
1046 	rc = update_note_header_size_elf32(ehdr_ptr);
1047 	if (rc < 0)
1048 		return rc;
1049 
1050 	rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
1051 	if (rc < 0)
1052 		return rc;
1053 
1054 	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
1055 	*notes_buf = vmcore_alloc_buf(*notes_sz);
1056 	if (!*notes_buf)
1057 		return -ENOMEM;
1058 
1059 	rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
1060 	if (rc < 0)
1061 		return rc;
1062 
1063 	/* Prepare merged PT_NOTE program header. */
1064 	phdr.p_type    = PT_NOTE;
1065 	phdr.p_flags   = 0;
1066 	note_off = sizeof(Elf32_Ehdr) +
1067 			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
1068 	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
1069 	phdr.p_vaddr   = phdr.p_paddr = 0;
1070 	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
1071 	phdr.p_align   = 4;
1072 
1073 	/* Add merged PT_NOTE program header*/
1074 	tmp = elfptr + sizeof(Elf32_Ehdr);
1075 	memcpy(tmp, &phdr, sizeof(phdr));
1076 	tmp += sizeof(phdr);
1077 
1078 	/* Remove unwanted PT_NOTE program headers. */
1079 	i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
1080 	*elfsz = *elfsz - i;
1081 	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
1082 	memset(elfptr + *elfsz, 0, i);
1083 	*elfsz = roundup(*elfsz, PAGE_SIZE);
1084 
1085 	/* Modify e_phnum to reflect merged headers. */
1086 	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
1087 
1088 	/* Store the size of all notes.  We need this to update the note
1089 	 * header when the device dumps will be added.
1090 	 */
1091 	elfnotes_orig_sz = phdr.p_memsz;
1092 
1093 	return 0;
1094 }
1095 
1096 /* Add memory chunks represented by program headers to vmcore list. Also update
1097  * 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)1098 static int __init process_ptload_program_headers_elf64(char *elfptr,
1099 						size_t elfsz,
1100 						size_t elfnotes_sz,
1101 						struct list_head *vc_list)
1102 {
1103 	int i;
1104 	Elf64_Ehdr *ehdr_ptr;
1105 	Elf64_Phdr *phdr_ptr;
1106 	loff_t vmcore_off;
1107 	struct vmcore *new;
1108 
1109 	ehdr_ptr = (Elf64_Ehdr *)elfptr;
1110 	phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
1111 
1112 	/* Skip ELF header, program headers and ELF note segment. */
1113 	vmcore_off = elfsz + elfnotes_sz;
1114 
1115 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1116 		u64 paddr, start, end, size;
1117 
1118 		if (phdr_ptr->p_type != PT_LOAD)
1119 			continue;
1120 
1121 		paddr = phdr_ptr->p_offset;
1122 		start = rounddown(paddr, PAGE_SIZE);
1123 		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1124 		size = end - start;
1125 
1126 		/* Add this contiguous chunk of memory to vmcore list.*/
1127 		new = get_new_element();
1128 		if (!new)
1129 			return -ENOMEM;
1130 		new->paddr = start;
1131 		new->size = size;
1132 		list_add_tail(&new->list, vc_list);
1133 
1134 		/* Update the program header offset. */
1135 		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1136 		vmcore_off = vmcore_off + size;
1137 	}
1138 	return 0;
1139 }
1140 
process_ptload_program_headers_elf32(char * elfptr,size_t elfsz,size_t elfnotes_sz,struct list_head * vc_list)1141 static int __init process_ptload_program_headers_elf32(char *elfptr,
1142 						size_t elfsz,
1143 						size_t elfnotes_sz,
1144 						struct list_head *vc_list)
1145 {
1146 	int i;
1147 	Elf32_Ehdr *ehdr_ptr;
1148 	Elf32_Phdr *phdr_ptr;
1149 	loff_t vmcore_off;
1150 	struct vmcore *new;
1151 
1152 	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1153 	phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
1154 
1155 	/* Skip ELF header, program headers and ELF note segment. */
1156 	vmcore_off = elfsz + elfnotes_sz;
1157 
1158 	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1159 		u64 paddr, start, end, size;
1160 
1161 		if (phdr_ptr->p_type != PT_LOAD)
1162 			continue;
1163 
1164 		paddr = phdr_ptr->p_offset;
1165 		start = rounddown(paddr, PAGE_SIZE);
1166 		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1167 		size = end - start;
1168 
1169 		/* Add this contiguous chunk of memory to vmcore list.*/
1170 		new = get_new_element();
1171 		if (!new)
1172 			return -ENOMEM;
1173 		new->paddr = start;
1174 		new->size = size;
1175 		list_add_tail(&new->list, vc_list);
1176 
1177 		/* Update the program header offset */
1178 		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1179 		vmcore_off = vmcore_off + size;
1180 	}
1181 	return 0;
1182 }
1183 
1184 /* Sets offset fields of vmcore elements. */
set_vmcore_list_offsets(size_t elfsz,size_t elfnotes_sz,struct list_head * vc_list)1185 static void set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
1186 				    struct list_head *vc_list)
1187 {
1188 	loff_t vmcore_off;
1189 	struct vmcore *m;
1190 
1191 	/* Skip ELF header, program headers and ELF note segment. */
1192 	vmcore_off = elfsz + elfnotes_sz;
1193 
1194 	list_for_each_entry(m, vc_list, list) {
1195 		m->offset = vmcore_off;
1196 		vmcore_off += m->size;
1197 	}
1198 }
1199 
free_elfcorebuf(void)1200 static void free_elfcorebuf(void)
1201 {
1202 	free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
1203 	elfcorebuf = NULL;
1204 	vfree(elfnotes_buf);
1205 	elfnotes_buf = NULL;
1206 }
1207 
parse_crash_elf64_headers(void)1208 static int __init parse_crash_elf64_headers(void)
1209 {
1210 	int rc=0;
1211 	Elf64_Ehdr ehdr;
1212 	u64 addr;
1213 
1214 	addr = elfcorehdr_addr;
1215 
1216 	/* Read ELF header */
1217 	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
1218 	if (rc < 0)
1219 		return rc;
1220 
1221 	/* Do some basic Verification. */
1222 	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1223 		(ehdr.e_type != ET_CORE) ||
1224 		!vmcore_elf64_check_arch(&ehdr) ||
1225 		ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
1226 		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1227 		ehdr.e_version != EV_CURRENT ||
1228 		ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
1229 		ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
1230 		ehdr.e_phnum == 0) {
1231 		pr_warn("Warning: Core image elf header is not sane\n");
1232 		return -EINVAL;
1233 	}
1234 
1235 	/* Read in all elf headers. */
1236 	elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
1237 				ehdr.e_phnum * sizeof(Elf64_Phdr);
1238 	elfcorebuf_sz = elfcorebuf_sz_orig;
1239 	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1240 					      get_order(elfcorebuf_sz_orig));
1241 	if (!elfcorebuf)
1242 		return -ENOMEM;
1243 	addr = elfcorehdr_addr;
1244 	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1245 	if (rc < 0)
1246 		goto fail;
1247 
1248 	/* Merge all PT_NOTE headers into one. */
1249 	rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
1250 				      &elfnotes_buf, &elfnotes_sz);
1251 	if (rc)
1252 		goto fail;
1253 	rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
1254 						  elfnotes_sz, &vmcore_list);
1255 	if (rc)
1256 		goto fail;
1257 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1258 	return 0;
1259 fail:
1260 	free_elfcorebuf();
1261 	return rc;
1262 }
1263 
parse_crash_elf32_headers(void)1264 static int __init parse_crash_elf32_headers(void)
1265 {
1266 	int rc=0;
1267 	Elf32_Ehdr ehdr;
1268 	u64 addr;
1269 
1270 	addr = elfcorehdr_addr;
1271 
1272 	/* Read ELF header */
1273 	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
1274 	if (rc < 0)
1275 		return rc;
1276 
1277 	/* Do some basic Verification. */
1278 	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1279 		(ehdr.e_type != ET_CORE) ||
1280 		!vmcore_elf32_check_arch(&ehdr) ||
1281 		ehdr.e_ident[EI_CLASS] != ELFCLASS32||
1282 		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1283 		ehdr.e_version != EV_CURRENT ||
1284 		ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1285 		ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1286 		ehdr.e_phnum == 0) {
1287 		pr_warn("Warning: Core image elf header is not sane\n");
1288 		return -EINVAL;
1289 	}
1290 
1291 	/* Read in all elf headers. */
1292 	elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1293 	elfcorebuf_sz = elfcorebuf_sz_orig;
1294 	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1295 					      get_order(elfcorebuf_sz_orig));
1296 	if (!elfcorebuf)
1297 		return -ENOMEM;
1298 	addr = elfcorehdr_addr;
1299 	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1300 	if (rc < 0)
1301 		goto fail;
1302 
1303 	/* Merge all PT_NOTE headers into one. */
1304 	rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1305 				      &elfnotes_buf, &elfnotes_sz);
1306 	if (rc)
1307 		goto fail;
1308 	rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1309 						  elfnotes_sz, &vmcore_list);
1310 	if (rc)
1311 		goto fail;
1312 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1313 	return 0;
1314 fail:
1315 	free_elfcorebuf();
1316 	return rc;
1317 }
1318 
parse_crash_elf_headers(void)1319 static int __init parse_crash_elf_headers(void)
1320 {
1321 	unsigned char e_ident[EI_NIDENT];
1322 	u64 addr;
1323 	int rc=0;
1324 
1325 	addr = elfcorehdr_addr;
1326 	rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1327 	if (rc < 0)
1328 		return rc;
1329 	if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1330 		pr_warn("Warning: Core image elf header not found\n");
1331 		return -EINVAL;
1332 	}
1333 
1334 	if (e_ident[EI_CLASS] == ELFCLASS64) {
1335 		rc = parse_crash_elf64_headers();
1336 		if (rc)
1337 			return rc;
1338 	} else if (e_ident[EI_CLASS] == ELFCLASS32) {
1339 		rc = parse_crash_elf32_headers();
1340 		if (rc)
1341 			return rc;
1342 	} else {
1343 		pr_warn("Warning: Core image elf header is not sane\n");
1344 		return -EINVAL;
1345 	}
1346 
1347 	/* Determine vmcore size. */
1348 	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1349 				      &vmcore_list);
1350 
1351 	return 0;
1352 }
1353 
1354 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1355 /**
1356  * vmcoredd_write_header - Write vmcore device dump header at the
1357  * beginning of the dump's buffer.
1358  * @buf: Output buffer where the note is written
1359  * @data: Dump info
1360  * @size: Size of the dump
1361  *
1362  * Fills beginning of the dump's buffer with vmcore device dump header.
1363  */
vmcoredd_write_header(void * buf,struct vmcoredd_data * data,u32 size)1364 static void vmcoredd_write_header(void *buf, struct vmcoredd_data *data,
1365 				  u32 size)
1366 {
1367 	struct vmcoredd_header *vdd_hdr = (struct vmcoredd_header *)buf;
1368 
1369 	vdd_hdr->n_namesz = sizeof(vdd_hdr->name);
1370 	vdd_hdr->n_descsz = size + sizeof(vdd_hdr->dump_name);
1371 	vdd_hdr->n_type = NT_VMCOREDD;
1372 
1373 	strncpy((char *)vdd_hdr->name, VMCOREDD_NOTE_NAME,
1374 		sizeof(vdd_hdr->name));
1375 	memcpy(vdd_hdr->dump_name, data->dump_name, sizeof(vdd_hdr->dump_name));
1376 }
1377 
1378 /**
1379  * vmcoredd_update_program_headers - Update all ELF program headers
1380  * @elfptr: Pointer to elf header
1381  * @elfnotesz: Size of elf notes aligned to page size
1382  * @vmcoreddsz: Size of device dumps to be added to elf note header
1383  *
1384  * Determine type of ELF header (Elf64 or Elf32) and update the elf note size.
1385  * Also update the offsets of all the program headers after the elf note header.
1386  */
vmcoredd_update_program_headers(char * elfptr,size_t elfnotesz,size_t vmcoreddsz)1387 static void vmcoredd_update_program_headers(char *elfptr, size_t elfnotesz,
1388 					    size_t vmcoreddsz)
1389 {
1390 	unsigned char *e_ident = (unsigned char *)elfptr;
1391 	u64 start, end, size;
1392 	loff_t vmcore_off;
1393 	u32 i;
1394 
1395 	vmcore_off = elfcorebuf_sz + elfnotesz;
1396 
1397 	if (e_ident[EI_CLASS] == ELFCLASS64) {
1398 		Elf64_Ehdr *ehdr = (Elf64_Ehdr *)elfptr;
1399 		Elf64_Phdr *phdr = (Elf64_Phdr *)(elfptr + sizeof(Elf64_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 	} else {
1418 		Elf32_Ehdr *ehdr = (Elf32_Ehdr *)elfptr;
1419 		Elf32_Phdr *phdr = (Elf32_Phdr *)(elfptr + sizeof(Elf32_Ehdr));
1420 
1421 		/* Update all program headers */
1422 		for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1423 			if (phdr->p_type == PT_NOTE) {
1424 				/* Update note size */
1425 				phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1426 				phdr->p_filesz = phdr->p_memsz;
1427 				continue;
1428 			}
1429 
1430 			start = rounddown(phdr->p_offset, PAGE_SIZE);
1431 			end = roundup(phdr->p_offset + phdr->p_memsz,
1432 				      PAGE_SIZE);
1433 			size = end - start;
1434 			phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1435 			vmcore_off += size;
1436 		}
1437 	}
1438 }
1439 
1440 /**
1441  * vmcoredd_update_size - Update the total size of the device dumps and update
1442  * ELF header
1443  * @dump_size: Size of the current device dump to be added to total size
1444  *
1445  * Update the total size of all the device dumps and update the ELF program
1446  * headers. Calculate the new offsets for the vmcore list and update the
1447  * total vmcore size.
1448  */
vmcoredd_update_size(size_t dump_size)1449 static void vmcoredd_update_size(size_t dump_size)
1450 {
1451 	vmcoredd_orig_sz += dump_size;
1452 	elfnotes_sz = roundup(elfnotes_orig_sz, PAGE_SIZE) + vmcoredd_orig_sz;
1453 	vmcoredd_update_program_headers(elfcorebuf, elfnotes_sz,
1454 					vmcoredd_orig_sz);
1455 
1456 	/* Update vmcore list offsets */
1457 	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1458 
1459 	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1460 				      &vmcore_list);
1461 	proc_vmcore->size = vmcore_size;
1462 }
1463 
1464 /**
1465  * vmcore_add_device_dump - Add a buffer containing device dump to vmcore
1466  * @data: dump info.
1467  *
1468  * Allocate a buffer and invoke the calling driver's dump collect routine.
1469  * Write ELF note at the beginning of the buffer to indicate vmcore device
1470  * dump and add the dump to global list.
1471  */
vmcore_add_device_dump(struct vmcoredd_data * data)1472 int vmcore_add_device_dump(struct vmcoredd_data *data)
1473 {
1474 	struct vmcoredd_node *dump;
1475 	void *buf = NULL;
1476 	size_t data_size;
1477 	int ret;
1478 
1479 	if (vmcoredd_disabled) {
1480 		pr_err_once("Device dump is disabled\n");
1481 		return -EINVAL;
1482 	}
1483 
1484 	if (!data || !strlen(data->dump_name) ||
1485 	    !data->vmcoredd_callback || !data->size)
1486 		return -EINVAL;
1487 
1488 	dump = vzalloc(sizeof(*dump));
1489 	if (!dump) {
1490 		ret = -ENOMEM;
1491 		goto out_err;
1492 	}
1493 
1494 	/* Keep size of the buffer page aligned so that it can be mmaped */
1495 	data_size = roundup(sizeof(struct vmcoredd_header) + data->size,
1496 			    PAGE_SIZE);
1497 
1498 	/* Allocate buffer for driver's to write their dumps */
1499 	buf = vmcore_alloc_buf(data_size);
1500 	if (!buf) {
1501 		ret = -ENOMEM;
1502 		goto out_err;
1503 	}
1504 
1505 	vmcoredd_write_header(buf, data, data_size -
1506 			      sizeof(struct vmcoredd_header));
1507 
1508 	/* Invoke the driver's dump collection routing */
1509 	ret = data->vmcoredd_callback(data, buf +
1510 				      sizeof(struct vmcoredd_header));
1511 	if (ret)
1512 		goto out_err;
1513 
1514 	dump->buf = buf;
1515 	dump->size = data_size;
1516 
1517 	/* Add the dump to driver sysfs list */
1518 	mutex_lock(&vmcoredd_mutex);
1519 	list_add_tail(&dump->list, &vmcoredd_list);
1520 	mutex_unlock(&vmcoredd_mutex);
1521 
1522 	vmcoredd_update_size(data_size);
1523 	return 0;
1524 
1525 out_err:
1526 	vfree(buf);
1527 	vfree(dump);
1528 
1529 	return ret;
1530 }
1531 EXPORT_SYMBOL(vmcore_add_device_dump);
1532 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1533 
1534 /* Free all dumps in vmcore device dump list */
vmcore_free_device_dumps(void)1535 static void vmcore_free_device_dumps(void)
1536 {
1537 #ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1538 	mutex_lock(&vmcoredd_mutex);
1539 	while (!list_empty(&vmcoredd_list)) {
1540 		struct vmcoredd_node *dump;
1541 
1542 		dump = list_first_entry(&vmcoredd_list, struct vmcoredd_node,
1543 					list);
1544 		list_del(&dump->list);
1545 		vfree(dump->buf);
1546 		vfree(dump);
1547 	}
1548 	mutex_unlock(&vmcoredd_mutex);
1549 #endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1550 }
1551 
1552 /* Init function for vmcore module. */
vmcore_init(void)1553 static int __init vmcore_init(void)
1554 {
1555 	int rc = 0;
1556 
1557 	/* Allow architectures to allocate ELF header in 2nd kernel */
1558 	rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1559 	if (rc)
1560 		return rc;
1561 	/*
1562 	 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1563 	 * then capture the dump.
1564 	 */
1565 	if (!(is_vmcore_usable()))
1566 		return rc;
1567 	rc = parse_crash_elf_headers();
1568 	if (rc) {
1569 		elfcorehdr_free(elfcorehdr_addr);
1570 		pr_warn("Kdump: vmcore not initialized\n");
1571 		return rc;
1572 	}
1573 	elfcorehdr_free(elfcorehdr_addr);
1574 	elfcorehdr_addr = ELFCORE_ADDR_ERR;
1575 
1576 	proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &vmcore_proc_ops);
1577 	if (proc_vmcore)
1578 		proc_vmcore->size = vmcore_size;
1579 	return 0;
1580 }
1581 fs_initcall(vmcore_init);
1582 
1583 /* Cleanup function for vmcore module. */
vmcore_cleanup(void)1584 void vmcore_cleanup(void)
1585 {
1586 	if (proc_vmcore) {
1587 		proc_remove(proc_vmcore);
1588 		proc_vmcore = NULL;
1589 	}
1590 
1591 	/* clear the vmcore list. */
1592 	while (!list_empty(&vmcore_list)) {
1593 		struct vmcore *m;
1594 
1595 		m = list_first_entry(&vmcore_list, struct vmcore, list);
1596 		list_del(&m->list);
1597 		kfree(m);
1598 	}
1599 	free_elfcorebuf();
1600 
1601 	/* clear vmcore device dump list */
1602 	vmcore_free_device_dumps();
1603 }
1604