Linux Audio

Check our new training course

Loading...
   1/*
   2 *	fs/proc/vmcore.c Interface for accessing the crash
   3 * 				 dump from the system's previous life.
   4 * 	Heavily borrowed from fs/proc/kcore.c
   5 *	Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
   6 *	Copyright (C) IBM Corporation, 2004. All rights reserved
   7 *
   8 */
   9
  10#include <linux/mm.h>
  11#include <linux/kcore.h>
  12#include <linux/user.h>
  13#include <linux/elf.h>
  14#include <linux/elfcore.h>
  15#include <linux/export.h>
  16#include <linux/slab.h>
  17#include <linux/highmem.h>
  18#include <linux/printk.h>
  19#include <linux/bootmem.h>
  20#include <linux/init.h>
  21#include <linux/crash_dump.h>
  22#include <linux/list.h>
  23#include <linux/vmalloc.h>
  24#include <linux/pagemap.h>
  25#include <linux/uaccess.h>
  26#include <asm/io.h>
  27#include "internal.h"
  28
  29/* List representing chunks of contiguous memory areas and their offsets in
  30 * vmcore file.
  31 */
  32static LIST_HEAD(vmcore_list);
  33
  34/* Stores the pointer to the buffer containing kernel elf core headers. */
  35static char *elfcorebuf;
  36static size_t elfcorebuf_sz;
  37static size_t elfcorebuf_sz_orig;
  38
  39static char *elfnotes_buf;
  40static size_t elfnotes_sz;
  41
  42/* Total size of vmcore file. */
  43static u64 vmcore_size;
  44
  45static struct proc_dir_entry *proc_vmcore;
  46
  47/*
  48 * Returns > 0 for RAM pages, 0 for non-RAM pages, < 0 on error
  49 * The called function has to take care of module refcounting.
  50 */
  51static int (*oldmem_pfn_is_ram)(unsigned long pfn);
  52
  53int register_oldmem_pfn_is_ram(int (*fn)(unsigned long pfn))
  54{
  55	if (oldmem_pfn_is_ram)
  56		return -EBUSY;
  57	oldmem_pfn_is_ram = fn;
  58	return 0;
  59}
  60EXPORT_SYMBOL_GPL(register_oldmem_pfn_is_ram);
  61
  62void unregister_oldmem_pfn_is_ram(void)
  63{
  64	oldmem_pfn_is_ram = NULL;
  65	wmb();
  66}
  67EXPORT_SYMBOL_GPL(unregister_oldmem_pfn_is_ram);
  68
  69static int pfn_is_ram(unsigned long pfn)
  70{
  71	int (*fn)(unsigned long pfn);
  72	/* pfn is ram unless fn() checks pagetype */
  73	int ret = 1;
  74
  75	/*
  76	 * Ask hypervisor if the pfn is really ram.
  77	 * A ballooned page contains no data and reading from such a page
  78	 * will cause high load in the hypervisor.
  79	 */
  80	fn = oldmem_pfn_is_ram;
  81	if (fn)
  82		ret = fn(pfn);
  83
  84	return ret;
  85}
  86
  87/* Reads a page from the oldmem device from given offset. */
  88static ssize_t read_from_oldmem(char *buf, size_t count,
  89				u64 *ppos, int userbuf)
  90{
  91	unsigned long pfn, offset;
  92	size_t nr_bytes;
  93	ssize_t read = 0, tmp;
  94
  95	if (!count)
  96		return 0;
  97
  98	offset = (unsigned long)(*ppos % PAGE_SIZE);
  99	pfn = (unsigned long)(*ppos / PAGE_SIZE);
 100
 101	do {
 102		if (count > (PAGE_SIZE - offset))
 103			nr_bytes = PAGE_SIZE - offset;
 104		else
 105			nr_bytes = count;
 106
 107		/* If pfn is not ram, return zeros for sparse dump files */
 108		if (pfn_is_ram(pfn) == 0)
 109			memset(buf, 0, nr_bytes);
 110		else {
 111			tmp = copy_oldmem_page(pfn, buf, nr_bytes,
 112						offset, userbuf);
 113			if (tmp < 0)
 114				return tmp;
 115		}
 116		*ppos += nr_bytes;
 117		count -= nr_bytes;
 118		buf += nr_bytes;
 119		read += nr_bytes;
 120		++pfn;
 121		offset = 0;
 122	} while (count);
 123
 124	return read;
 125}
 126
 127/*
 128 * Architectures may override this function to allocate ELF header in 2nd kernel
 129 */
 130int __weak elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
 131{
 132	return 0;
 133}
 134
 135/*
 136 * Architectures may override this function to free header
 137 */
 138void __weak elfcorehdr_free(unsigned long long addr)
 139{}
 140
 141/*
 142 * Architectures may override this function to read from ELF header
 143 */
 144ssize_t __weak elfcorehdr_read(char *buf, size_t count, u64 *ppos)
 145{
 146	return read_from_oldmem(buf, count, ppos, 0);
 147}
 148
 149/*
 150 * Architectures may override this function to read from notes sections
 151 */
 152ssize_t __weak elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
 153{
 154	return read_from_oldmem(buf, count, ppos, 0);
 155}
 156
 157/*
 158 * Architectures may override this function to map oldmem
 159 */
 160int __weak remap_oldmem_pfn_range(struct vm_area_struct *vma,
 161				  unsigned long from, unsigned long pfn,
 162				  unsigned long size, pgprot_t prot)
 163{
 164	return remap_pfn_range(vma, from, pfn, size, prot);
 165}
 166
 167/*
 168 * Copy to either kernel or user space
 169 */
 170static int copy_to(void *target, void *src, size_t size, int userbuf)
 171{
 172	if (userbuf) {
 173		if (copy_to_user((char __user *) target, src, size))
 174			return -EFAULT;
 175	} else {
 176		memcpy(target, src, size);
 177	}
 178	return 0;
 179}
 180
 181/* Read from the ELF header and then the crash dump. On error, negative value is
 182 * returned otherwise number of bytes read are returned.
 183 */
 184static ssize_t __read_vmcore(char *buffer, size_t buflen, loff_t *fpos,
 185			     int userbuf)
 186{
 187	ssize_t acc = 0, tmp;
 188	size_t tsz;
 189	u64 start;
 190	struct vmcore *m = NULL;
 191
 192	if (buflen == 0 || *fpos >= vmcore_size)
 193		return 0;
 194
 195	/* trim buflen to not go beyond EOF */
 196	if (buflen > vmcore_size - *fpos)
 197		buflen = vmcore_size - *fpos;
 198
 199	/* Read ELF core header */
 200	if (*fpos < elfcorebuf_sz) {
 201		tsz = min(elfcorebuf_sz - (size_t)*fpos, buflen);
 202		if (copy_to(buffer, elfcorebuf + *fpos, tsz, userbuf))
 203			return -EFAULT;
 204		buflen -= tsz;
 205		*fpos += tsz;
 206		buffer += tsz;
 207		acc += tsz;
 208
 209		/* leave now if filled buffer already */
 210		if (buflen == 0)
 211			return acc;
 212	}
 213
 214	/* Read Elf note segment */
 215	if (*fpos < elfcorebuf_sz + elfnotes_sz) {
 216		void *kaddr;
 217
 218		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)*fpos, buflen);
 219		kaddr = elfnotes_buf + *fpos - elfcorebuf_sz;
 220		if (copy_to(buffer, kaddr, tsz, userbuf))
 221			return -EFAULT;
 222		buflen -= tsz;
 223		*fpos += tsz;
 224		buffer += tsz;
 225		acc += tsz;
 226
 227		/* leave now if filled buffer already */
 228		if (buflen == 0)
 229			return acc;
 230	}
 231
 232	list_for_each_entry(m, &vmcore_list, list) {
 233		if (*fpos < m->offset + m->size) {
 234			tsz = (size_t)min_t(unsigned long long,
 235					    m->offset + m->size - *fpos,
 236					    buflen);
 237			start = m->paddr + *fpos - m->offset;
 238			tmp = read_from_oldmem(buffer, tsz, &start, userbuf);
 239			if (tmp < 0)
 240				return tmp;
 241			buflen -= tsz;
 242			*fpos += tsz;
 243			buffer += tsz;
 244			acc += tsz;
 245
 246			/* leave now if filled buffer already */
 247			if (buflen == 0)
 248				return acc;
 249		}
 250	}
 251
 252	return acc;
 253}
 254
 255static ssize_t read_vmcore(struct file *file, char __user *buffer,
 256			   size_t buflen, loff_t *fpos)
 257{
 258	return __read_vmcore((__force char *) buffer, buflen, fpos, 1);
 259}
 260
 261/*
 262 * The vmcore fault handler uses the page cache and fills data using the
 263 * standard __vmcore_read() function.
 264 *
 265 * On s390 the fault handler is used for memory regions that can't be mapped
 266 * directly with remap_pfn_range().
 267 */
 268static int mmap_vmcore_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 269{
 270#ifdef CONFIG_S390
 271	struct address_space *mapping = vma->vm_file->f_mapping;
 272	pgoff_t index = vmf->pgoff;
 273	struct page *page;
 274	loff_t offset;
 275	char *buf;
 276	int rc;
 277
 278	page = find_or_create_page(mapping, index, GFP_KERNEL);
 279	if (!page)
 280		return VM_FAULT_OOM;
 281	if (!PageUptodate(page)) {
 282		offset = (loff_t) index << PAGE_SHIFT;
 283		buf = __va((page_to_pfn(page) << PAGE_SHIFT));
 284		rc = __read_vmcore(buf, PAGE_SIZE, &offset, 0);
 285		if (rc < 0) {
 286			unlock_page(page);
 287			put_page(page);
 288			return (rc == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
 289		}
 290		SetPageUptodate(page);
 291	}
 292	unlock_page(page);
 293	vmf->page = page;
 294	return 0;
 295#else
 296	return VM_FAULT_SIGBUS;
 297#endif
 298}
 299
 300static const struct vm_operations_struct vmcore_mmap_ops = {
 301	.fault = mmap_vmcore_fault,
 302};
 303
 304/**
 305 * alloc_elfnotes_buf - allocate buffer for ELF note segment in
 306 *                      vmalloc memory
 307 *
 308 * @notes_sz: size of buffer
 309 *
 310 * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
 311 * the buffer to user-space by means of remap_vmalloc_range().
 312 *
 313 * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
 314 * disabled and there's no need to allow users to mmap the buffer.
 315 */
 316static inline char *alloc_elfnotes_buf(size_t notes_sz)
 317{
 318#ifdef CONFIG_MMU
 319	return vmalloc_user(notes_sz);
 320#else
 321	return vzalloc(notes_sz);
 322#endif
 323}
 324
 325/*
 326 * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
 327 * essential for mmap_vmcore() in order to map physically
 328 * non-contiguous objects (ELF header, ELF note segment and memory
 329 * regions in the 1st kernel pointed to by PT_LOAD entries) into
 330 * virtually contiguous user-space in ELF layout.
 331 */
 332#ifdef CONFIG_MMU
 333/*
 334 * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
 335 * reported as not being ram with the zero page.
 336 *
 337 * @vma: vm_area_struct describing requested mapping
 338 * @from: start remapping from
 339 * @pfn: page frame number to start remapping to
 340 * @size: remapping size
 341 * @prot: protection bits
 342 *
 343 * Returns zero on success, -EAGAIN on failure.
 344 */
 345static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
 346				    unsigned long from, unsigned long pfn,
 347				    unsigned long size, pgprot_t prot)
 348{
 349	unsigned long map_size;
 350	unsigned long pos_start, pos_end, pos;
 351	unsigned long zeropage_pfn = my_zero_pfn(0);
 352	size_t len = 0;
 353
 354	pos_start = pfn;
 355	pos_end = pfn + (size >> PAGE_SHIFT);
 356
 357	for (pos = pos_start; pos < pos_end; ++pos) {
 358		if (!pfn_is_ram(pos)) {
 359			/*
 360			 * We hit a page which is not ram. Remap the continuous
 361			 * region between pos_start and pos-1 and replace
 362			 * the non-ram page at pos with the zero page.
 363			 */
 364			if (pos > pos_start) {
 365				/* Remap continuous region */
 366				map_size = (pos - pos_start) << PAGE_SHIFT;
 367				if (remap_oldmem_pfn_range(vma, from + len,
 368							   pos_start, map_size,
 369							   prot))
 370					goto fail;
 371				len += map_size;
 372			}
 373			/* Remap the zero page */
 374			if (remap_oldmem_pfn_range(vma, from + len,
 375						   zeropage_pfn,
 376						   PAGE_SIZE, prot))
 377				goto fail;
 378			len += PAGE_SIZE;
 379			pos_start = pos + 1;
 380		}
 381	}
 382	if (pos > pos_start) {
 383		/* Remap the rest */
 384		map_size = (pos - pos_start) << PAGE_SHIFT;
 385		if (remap_oldmem_pfn_range(vma, from + len, pos_start,
 386					   map_size, prot))
 387			goto fail;
 388	}
 389	return 0;
 390fail:
 391	do_munmap(vma->vm_mm, from, len);
 392	return -EAGAIN;
 393}
 394
 395static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
 396			    unsigned long from, unsigned long pfn,
 397			    unsigned long size, pgprot_t prot)
 398{
 399	/*
 400	 * Check if oldmem_pfn_is_ram was registered to avoid
 401	 * looping over all pages without a reason.
 402	 */
 403	if (oldmem_pfn_is_ram)
 404		return remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
 405	else
 406		return remap_oldmem_pfn_range(vma, from, pfn, size, prot);
 407}
 408
 409static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 410{
 411	size_t size = vma->vm_end - vma->vm_start;
 412	u64 start, end, len, tsz;
 413	struct vmcore *m;
 414
 415	start = (u64)vma->vm_pgoff << PAGE_SHIFT;
 416	end = start + size;
 417
 418	if (size > vmcore_size || end > vmcore_size)
 419		return -EINVAL;
 420
 421	if (vma->vm_flags & (VM_WRITE | VM_EXEC))
 422		return -EPERM;
 423
 424	vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
 425	vma->vm_flags |= VM_MIXEDMAP;
 426	vma->vm_ops = &vmcore_mmap_ops;
 427
 428	len = 0;
 429
 430	if (start < elfcorebuf_sz) {
 431		u64 pfn;
 432
 433		tsz = min(elfcorebuf_sz - (size_t)start, size);
 434		pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
 435		if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
 436				    vma->vm_page_prot))
 437			return -EAGAIN;
 438		size -= tsz;
 439		start += tsz;
 440		len += tsz;
 441
 442		if (size == 0)
 443			return 0;
 444	}
 445
 446	if (start < elfcorebuf_sz + elfnotes_sz) {
 447		void *kaddr;
 448
 449		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
 450		kaddr = elfnotes_buf + start - elfcorebuf_sz;
 451		if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
 452						kaddr, tsz))
 453			goto fail;
 454		size -= tsz;
 455		start += tsz;
 456		len += tsz;
 457
 458		if (size == 0)
 459			return 0;
 460	}
 461
 462	list_for_each_entry(m, &vmcore_list, list) {
 463		if (start < m->offset + m->size) {
 464			u64 paddr = 0;
 465
 466			tsz = (size_t)min_t(unsigned long long,
 467					    m->offset + m->size - start, size);
 468			paddr = m->paddr + start - m->offset;
 469			if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
 470						    paddr >> PAGE_SHIFT, tsz,
 471						    vma->vm_page_prot))
 472				goto fail;
 473			size -= tsz;
 474			start += tsz;
 475			len += tsz;
 476
 477			if (size == 0)
 478				return 0;
 479		}
 480	}
 481
 482	return 0;
 483fail:
 484	do_munmap(vma->vm_mm, vma->vm_start, len);
 485	return -EAGAIN;
 486}
 487#else
 488static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 489{
 490	return -ENOSYS;
 491}
 492#endif
 493
 494static const struct file_operations proc_vmcore_operations = {
 495	.read		= read_vmcore,
 496	.llseek		= default_llseek,
 497	.mmap		= mmap_vmcore,
 498};
 499
 500static struct vmcore* __init get_new_element(void)
 501{
 502	return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
 503}
 504
 505static u64 __init get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
 506				  struct list_head *vc_list)
 507{
 508	u64 size;
 509	struct vmcore *m;
 510
 511	size = elfsz + elfnotesegsz;
 512	list_for_each_entry(m, vc_list, list) {
 513		size += m->size;
 514	}
 515	return size;
 516}
 517
 518/**
 519 * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
 520 *
 521 * @ehdr_ptr: ELF header
 522 *
 523 * This function updates p_memsz member of each PT_NOTE entry in the
 524 * program header table pointed to by @ehdr_ptr to real size of ELF
 525 * note segment.
 526 */
 527static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
 528{
 529	int i, rc=0;
 530	Elf64_Phdr *phdr_ptr;
 531	Elf64_Nhdr *nhdr_ptr;
 532
 533	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
 534	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 535		void *notes_section;
 536		u64 offset, max_sz, sz, real_sz = 0;
 537		if (phdr_ptr->p_type != PT_NOTE)
 538			continue;
 539		max_sz = phdr_ptr->p_memsz;
 540		offset = phdr_ptr->p_offset;
 541		notes_section = kmalloc(max_sz, GFP_KERNEL);
 542		if (!notes_section)
 543			return -ENOMEM;
 544		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
 545		if (rc < 0) {
 546			kfree(notes_section);
 547			return rc;
 548		}
 549		nhdr_ptr = notes_section;
 550		while (nhdr_ptr->n_namesz != 0) {
 551			sz = sizeof(Elf64_Nhdr) +
 552				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
 553				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
 554			if ((real_sz + sz) > max_sz) {
 555				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
 556					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
 557				break;
 558			}
 559			real_sz += sz;
 560			nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
 561		}
 562		kfree(notes_section);
 563		phdr_ptr->p_memsz = real_sz;
 564		if (real_sz == 0) {
 565			pr_warn("Warning: Zero PT_NOTE entries found\n");
 566		}
 567	}
 568
 569	return 0;
 570}
 571
 572/**
 573 * get_note_number_and_size_elf64 - get the number of PT_NOTE program
 574 * headers and sum of real size of their ELF note segment headers and
 575 * data.
 576 *
 577 * @ehdr_ptr: ELF header
 578 * @nr_ptnote: buffer for the number of PT_NOTE program headers
 579 * @sz_ptnote: buffer for size of unique PT_NOTE program header
 580 *
 581 * This function is used to merge multiple PT_NOTE program headers
 582 * into a unique single one. The resulting unique entry will have
 583 * @sz_ptnote in its phdr->p_mem.
 584 *
 585 * It is assumed that program headers with PT_NOTE type pointed to by
 586 * @ehdr_ptr has already been updated by update_note_header_size_elf64
 587 * and each of PT_NOTE program headers has actual ELF note segment
 588 * size in its p_memsz member.
 589 */
 590static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
 591						 int *nr_ptnote, u64 *sz_ptnote)
 592{
 593	int i;
 594	Elf64_Phdr *phdr_ptr;
 595
 596	*nr_ptnote = *sz_ptnote = 0;
 597
 598	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
 599	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 600		if (phdr_ptr->p_type != PT_NOTE)
 601			continue;
 602		*nr_ptnote += 1;
 603		*sz_ptnote += phdr_ptr->p_memsz;
 604	}
 605
 606	return 0;
 607}
 608
 609/**
 610 * copy_notes_elf64 - copy ELF note segments in a given buffer
 611 *
 612 * @ehdr_ptr: ELF header
 613 * @notes_buf: buffer into which ELF note segments are copied
 614 *
 615 * This function is used to copy ELF note segment in the 1st kernel
 616 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
 617 * size of the buffer @notes_buf is equal to or larger than sum of the
 618 * real ELF note segment headers and data.
 619 *
 620 * It is assumed that program headers with PT_NOTE type pointed to by
 621 * @ehdr_ptr has already been updated by update_note_header_size_elf64
 622 * and each of PT_NOTE program headers has actual ELF note segment
 623 * size in its p_memsz member.
 624 */
 625static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
 626{
 627	int i, rc=0;
 628	Elf64_Phdr *phdr_ptr;
 629
 630	phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
 631
 632	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 633		u64 offset;
 634		if (phdr_ptr->p_type != PT_NOTE)
 635			continue;
 636		offset = phdr_ptr->p_offset;
 637		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
 638					   &offset);
 639		if (rc < 0)
 640			return rc;
 641		notes_buf += phdr_ptr->p_memsz;
 642	}
 643
 644	return 0;
 645}
 646
 647/* Merges all the PT_NOTE headers into one. */
 648static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
 649					   char **notes_buf, size_t *notes_sz)
 650{
 651	int i, nr_ptnote=0, rc=0;
 652	char *tmp;
 653	Elf64_Ehdr *ehdr_ptr;
 654	Elf64_Phdr phdr;
 655	u64 phdr_sz = 0, note_off;
 656
 657	ehdr_ptr = (Elf64_Ehdr *)elfptr;
 658
 659	rc = update_note_header_size_elf64(ehdr_ptr);
 660	if (rc < 0)
 661		return rc;
 662
 663	rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
 664	if (rc < 0)
 665		return rc;
 666
 667	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
 668	*notes_buf = alloc_elfnotes_buf(*notes_sz);
 669	if (!*notes_buf)
 670		return -ENOMEM;
 671
 672	rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
 673	if (rc < 0)
 674		return rc;
 675
 676	/* Prepare merged PT_NOTE program header. */
 677	phdr.p_type    = PT_NOTE;
 678	phdr.p_flags   = 0;
 679	note_off = sizeof(Elf64_Ehdr) +
 680			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
 681	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
 682	phdr.p_vaddr   = phdr.p_paddr = 0;
 683	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
 684	phdr.p_align   = 0;
 685
 686	/* Add merged PT_NOTE program header*/
 687	tmp = elfptr + sizeof(Elf64_Ehdr);
 688	memcpy(tmp, &phdr, sizeof(phdr));
 689	tmp += sizeof(phdr);
 690
 691	/* Remove unwanted PT_NOTE program headers. */
 692	i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
 693	*elfsz = *elfsz - i;
 694	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
 695	memset(elfptr + *elfsz, 0, i);
 696	*elfsz = roundup(*elfsz, PAGE_SIZE);
 697
 698	/* Modify e_phnum to reflect merged headers. */
 699	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
 700
 701	return 0;
 702}
 703
 704/**
 705 * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
 706 *
 707 * @ehdr_ptr: ELF header
 708 *
 709 * This function updates p_memsz member of each PT_NOTE entry in the
 710 * program header table pointed to by @ehdr_ptr to real size of ELF
 711 * note segment.
 712 */
 713static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
 714{
 715	int i, rc=0;
 716	Elf32_Phdr *phdr_ptr;
 717	Elf32_Nhdr *nhdr_ptr;
 718
 719	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
 720	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 721		void *notes_section;
 722		u64 offset, max_sz, sz, real_sz = 0;
 723		if (phdr_ptr->p_type != PT_NOTE)
 724			continue;
 725		max_sz = phdr_ptr->p_memsz;
 726		offset = phdr_ptr->p_offset;
 727		notes_section = kmalloc(max_sz, GFP_KERNEL);
 728		if (!notes_section)
 729			return -ENOMEM;
 730		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
 731		if (rc < 0) {
 732			kfree(notes_section);
 733			return rc;
 734		}
 735		nhdr_ptr = notes_section;
 736		while (nhdr_ptr->n_namesz != 0) {
 737			sz = sizeof(Elf32_Nhdr) +
 738				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
 739				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
 740			if ((real_sz + sz) > max_sz) {
 741				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
 742					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
 743				break;
 744			}
 745			real_sz += sz;
 746			nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
 747		}
 748		kfree(notes_section);
 749		phdr_ptr->p_memsz = real_sz;
 750		if (real_sz == 0) {
 751			pr_warn("Warning: Zero PT_NOTE entries found\n");
 752		}
 753	}
 754
 755	return 0;
 756}
 757
 758/**
 759 * get_note_number_and_size_elf32 - get the number of PT_NOTE program
 760 * headers and sum of real size of their ELF note segment headers and
 761 * data.
 762 *
 763 * @ehdr_ptr: ELF header
 764 * @nr_ptnote: buffer for the number of PT_NOTE program headers
 765 * @sz_ptnote: buffer for size of unique PT_NOTE program header
 766 *
 767 * This function is used to merge multiple PT_NOTE program headers
 768 * into a unique single one. The resulting unique entry will have
 769 * @sz_ptnote in its phdr->p_mem.
 770 *
 771 * It is assumed that program headers with PT_NOTE type pointed to by
 772 * @ehdr_ptr has already been updated by update_note_header_size_elf32
 773 * and each of PT_NOTE program headers has actual ELF note segment
 774 * size in its p_memsz member.
 775 */
 776static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
 777						 int *nr_ptnote, u64 *sz_ptnote)
 778{
 779	int i;
 780	Elf32_Phdr *phdr_ptr;
 781
 782	*nr_ptnote = *sz_ptnote = 0;
 783
 784	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
 785	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 786		if (phdr_ptr->p_type != PT_NOTE)
 787			continue;
 788		*nr_ptnote += 1;
 789		*sz_ptnote += phdr_ptr->p_memsz;
 790	}
 791
 792	return 0;
 793}
 794
 795/**
 796 * copy_notes_elf32 - copy ELF note segments in a given buffer
 797 *
 798 * @ehdr_ptr: ELF header
 799 * @notes_buf: buffer into which ELF note segments are copied
 800 *
 801 * This function is used to copy ELF note segment in the 1st kernel
 802 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
 803 * size of the buffer @notes_buf is equal to or larger than sum of the
 804 * real ELF note segment headers and data.
 805 *
 806 * It is assumed that program headers with PT_NOTE type pointed to by
 807 * @ehdr_ptr has already been updated by update_note_header_size_elf32
 808 * and each of PT_NOTE program headers has actual ELF note segment
 809 * size in its p_memsz member.
 810 */
 811static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
 812{
 813	int i, rc=0;
 814	Elf32_Phdr *phdr_ptr;
 815
 816	phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
 817
 818	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 819		u64 offset;
 820		if (phdr_ptr->p_type != PT_NOTE)
 821			continue;
 822		offset = phdr_ptr->p_offset;
 823		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
 824					   &offset);
 825		if (rc < 0)
 826			return rc;
 827		notes_buf += phdr_ptr->p_memsz;
 828	}
 829
 830	return 0;
 831}
 832
 833/* Merges all the PT_NOTE headers into one. */
 834static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
 835					   char **notes_buf, size_t *notes_sz)
 836{
 837	int i, nr_ptnote=0, rc=0;
 838	char *tmp;
 839	Elf32_Ehdr *ehdr_ptr;
 840	Elf32_Phdr phdr;
 841	u64 phdr_sz = 0, note_off;
 842
 843	ehdr_ptr = (Elf32_Ehdr *)elfptr;
 844
 845	rc = update_note_header_size_elf32(ehdr_ptr);
 846	if (rc < 0)
 847		return rc;
 848
 849	rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
 850	if (rc < 0)
 851		return rc;
 852
 853	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
 854	*notes_buf = alloc_elfnotes_buf(*notes_sz);
 855	if (!*notes_buf)
 856		return -ENOMEM;
 857
 858	rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
 859	if (rc < 0)
 860		return rc;
 861
 862	/* Prepare merged PT_NOTE program header. */
 863	phdr.p_type    = PT_NOTE;
 864	phdr.p_flags   = 0;
 865	note_off = sizeof(Elf32_Ehdr) +
 866			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
 867	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
 868	phdr.p_vaddr   = phdr.p_paddr = 0;
 869	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
 870	phdr.p_align   = 0;
 871
 872	/* Add merged PT_NOTE program header*/
 873	tmp = elfptr + sizeof(Elf32_Ehdr);
 874	memcpy(tmp, &phdr, sizeof(phdr));
 875	tmp += sizeof(phdr);
 876
 877	/* Remove unwanted PT_NOTE program headers. */
 878	i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
 879	*elfsz = *elfsz - i;
 880	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
 881	memset(elfptr + *elfsz, 0, i);
 882	*elfsz = roundup(*elfsz, PAGE_SIZE);
 883
 884	/* Modify e_phnum to reflect merged headers. */
 885	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
 886
 887	return 0;
 888}
 889
 890/* Add memory chunks represented by program headers to vmcore list. Also update
 891 * the new offset fields of exported program headers. */
 892static int __init process_ptload_program_headers_elf64(char *elfptr,
 893						size_t elfsz,
 894						size_t elfnotes_sz,
 895						struct list_head *vc_list)
 896{
 897	int i;
 898	Elf64_Ehdr *ehdr_ptr;
 899	Elf64_Phdr *phdr_ptr;
 900	loff_t vmcore_off;
 901	struct vmcore *new;
 902
 903	ehdr_ptr = (Elf64_Ehdr *)elfptr;
 904	phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
 905
 906	/* Skip Elf header, program headers and Elf note segment. */
 907	vmcore_off = elfsz + elfnotes_sz;
 908
 909	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 910		u64 paddr, start, end, size;
 911
 912		if (phdr_ptr->p_type != PT_LOAD)
 913			continue;
 914
 915		paddr = phdr_ptr->p_offset;
 916		start = rounddown(paddr, PAGE_SIZE);
 917		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
 918		size = end - start;
 919
 920		/* Add this contiguous chunk of memory to vmcore list.*/
 921		new = get_new_element();
 922		if (!new)
 923			return -ENOMEM;
 924		new->paddr = start;
 925		new->size = size;
 926		list_add_tail(&new->list, vc_list);
 927
 928		/* Update the program header offset. */
 929		phdr_ptr->p_offset = vmcore_off + (paddr - start);
 930		vmcore_off = vmcore_off + size;
 931	}
 932	return 0;
 933}
 934
 935static int __init process_ptload_program_headers_elf32(char *elfptr,
 936						size_t elfsz,
 937						size_t elfnotes_sz,
 938						struct list_head *vc_list)
 939{
 940	int i;
 941	Elf32_Ehdr *ehdr_ptr;
 942	Elf32_Phdr *phdr_ptr;
 943	loff_t vmcore_off;
 944	struct vmcore *new;
 945
 946	ehdr_ptr = (Elf32_Ehdr *)elfptr;
 947	phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
 948
 949	/* Skip Elf header, program headers and Elf note segment. */
 950	vmcore_off = elfsz + elfnotes_sz;
 951
 952	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 953		u64 paddr, start, end, size;
 954
 955		if (phdr_ptr->p_type != PT_LOAD)
 956			continue;
 957
 958		paddr = phdr_ptr->p_offset;
 959		start = rounddown(paddr, PAGE_SIZE);
 960		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
 961		size = end - start;
 962
 963		/* Add this contiguous chunk of memory to vmcore list.*/
 964		new = get_new_element();
 965		if (!new)
 966			return -ENOMEM;
 967		new->paddr = start;
 968		new->size = size;
 969		list_add_tail(&new->list, vc_list);
 970
 971		/* Update the program header offset */
 972		phdr_ptr->p_offset = vmcore_off + (paddr - start);
 973		vmcore_off = vmcore_off + size;
 974	}
 975	return 0;
 976}
 977
 978/* Sets offset fields of vmcore elements. */
 979static void __init set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
 980					   struct list_head *vc_list)
 981{
 982	loff_t vmcore_off;
 983	struct vmcore *m;
 984
 985	/* Skip Elf header, program headers and Elf note segment. */
 986	vmcore_off = elfsz + elfnotes_sz;
 987
 988	list_for_each_entry(m, vc_list, list) {
 989		m->offset = vmcore_off;
 990		vmcore_off += m->size;
 991	}
 992}
 993
 994static void free_elfcorebuf(void)
 995{
 996	free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
 997	elfcorebuf = NULL;
 998	vfree(elfnotes_buf);
 999	elfnotes_buf = NULL;
1000}
1001
1002static int __init parse_crash_elf64_headers(void)
1003{
1004	int rc=0;
1005	Elf64_Ehdr ehdr;
1006	u64 addr;
1007
1008	addr = elfcorehdr_addr;
1009
1010	/* Read Elf header */
1011	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
1012	if (rc < 0)
1013		return rc;
1014
1015	/* Do some basic Verification. */
1016	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1017		(ehdr.e_type != ET_CORE) ||
1018		!vmcore_elf64_check_arch(&ehdr) ||
1019		ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
1020		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1021		ehdr.e_version != EV_CURRENT ||
1022		ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
1023		ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
1024		ehdr.e_phnum == 0) {
1025		pr_warn("Warning: Core image elf header is not sane\n");
1026		return -EINVAL;
1027	}
1028
1029	/* Read in all elf headers. */
1030	elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
1031				ehdr.e_phnum * sizeof(Elf64_Phdr);
1032	elfcorebuf_sz = elfcorebuf_sz_orig;
1033	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1034					      get_order(elfcorebuf_sz_orig));
1035	if (!elfcorebuf)
1036		return -ENOMEM;
1037	addr = elfcorehdr_addr;
1038	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1039	if (rc < 0)
1040		goto fail;
1041
1042	/* Merge all PT_NOTE headers into one. */
1043	rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
1044				      &elfnotes_buf, &elfnotes_sz);
1045	if (rc)
1046		goto fail;
1047	rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
1048						  elfnotes_sz, &vmcore_list);
1049	if (rc)
1050		goto fail;
1051	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1052	return 0;
1053fail:
1054	free_elfcorebuf();
1055	return rc;
1056}
1057
1058static int __init parse_crash_elf32_headers(void)
1059{
1060	int rc=0;
1061	Elf32_Ehdr ehdr;
1062	u64 addr;
1063
1064	addr = elfcorehdr_addr;
1065
1066	/* Read Elf header */
1067	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
1068	if (rc < 0)
1069		return rc;
1070
1071	/* Do some basic Verification. */
1072	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1073		(ehdr.e_type != ET_CORE) ||
1074		!vmcore_elf32_check_arch(&ehdr) ||
1075		ehdr.e_ident[EI_CLASS] != ELFCLASS32||
1076		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1077		ehdr.e_version != EV_CURRENT ||
1078		ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1079		ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1080		ehdr.e_phnum == 0) {
1081		pr_warn("Warning: Core image elf header is not sane\n");
1082		return -EINVAL;
1083	}
1084
1085	/* Read in all elf headers. */
1086	elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1087	elfcorebuf_sz = elfcorebuf_sz_orig;
1088	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1089					      get_order(elfcorebuf_sz_orig));
1090	if (!elfcorebuf)
1091		return -ENOMEM;
1092	addr = elfcorehdr_addr;
1093	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1094	if (rc < 0)
1095		goto fail;
1096
1097	/* Merge all PT_NOTE headers into one. */
1098	rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1099				      &elfnotes_buf, &elfnotes_sz);
1100	if (rc)
1101		goto fail;
1102	rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1103						  elfnotes_sz, &vmcore_list);
1104	if (rc)
1105		goto fail;
1106	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1107	return 0;
1108fail:
1109	free_elfcorebuf();
1110	return rc;
1111}
1112
1113static int __init parse_crash_elf_headers(void)
1114{
1115	unsigned char e_ident[EI_NIDENT];
1116	u64 addr;
1117	int rc=0;
1118
1119	addr = elfcorehdr_addr;
1120	rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1121	if (rc < 0)
1122		return rc;
1123	if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1124		pr_warn("Warning: Core image elf header not found\n");
1125		return -EINVAL;
1126	}
1127
1128	if (e_ident[EI_CLASS] == ELFCLASS64) {
1129		rc = parse_crash_elf64_headers();
1130		if (rc)
1131			return rc;
1132	} else if (e_ident[EI_CLASS] == ELFCLASS32) {
1133		rc = parse_crash_elf32_headers();
1134		if (rc)
1135			return rc;
1136	} else {
1137		pr_warn("Warning: Core image elf header is not sane\n");
1138		return -EINVAL;
1139	}
1140
1141	/* Determine vmcore size. */
1142	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1143				      &vmcore_list);
1144
1145	return 0;
1146}
1147
1148/* Init function for vmcore module. */
1149static int __init vmcore_init(void)
1150{
1151	int rc = 0;
1152
1153	/* Allow architectures to allocate ELF header in 2nd kernel */
1154	rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1155	if (rc)
1156		return rc;
1157	/*
1158	 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1159	 * then capture the dump.
1160	 */
1161	if (!(is_vmcore_usable()))
1162		return rc;
1163	rc = parse_crash_elf_headers();
1164	if (rc) {
1165		pr_warn("Kdump: vmcore not initialized\n");
1166		return rc;
1167	}
1168	elfcorehdr_free(elfcorehdr_addr);
1169	elfcorehdr_addr = ELFCORE_ADDR_ERR;
1170
1171	proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &proc_vmcore_operations);
1172	if (proc_vmcore)
1173		proc_vmcore->size = vmcore_size;
1174	return 0;
1175}
1176fs_initcall(vmcore_init);
1177
1178/* Cleanup function for vmcore module. */
1179void vmcore_cleanup(void)
1180{
1181	struct list_head *pos, *next;
1182
1183	if (proc_vmcore) {
1184		proc_remove(proc_vmcore);
1185		proc_vmcore = NULL;
1186	}
1187
1188	/* clear the vmcore list. */
1189	list_for_each_safe(pos, next, &vmcore_list) {
1190		struct vmcore *m;
1191
1192		m = list_entry(pos, struct vmcore, list);
1193		list_del(&m->list);
1194		kfree(m);
1195	}
1196	free_elfcorebuf();
1197}
   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 */
  36static LIST_HEAD(vmcore_list);
  37
  38/* Stores the pointer to the buffer containing kernel elf core headers. */
  39static char *elfcorebuf;
  40static size_t elfcorebuf_sz;
  41static size_t elfcorebuf_sz_orig;
  42
  43static char *elfnotes_buf;
  44static size_t elfnotes_sz;
  45/* Size of all notes minus the device dump notes */
  46static size_t elfnotes_orig_sz;
  47
  48/* Total size of vmcore file. */
  49static u64 vmcore_size;
  50
  51static 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 */
  55static LIST_HEAD(vmcoredd_list);
  56static DEFINE_MUTEX(vmcoredd_mutex);
  57
  58static bool vmcoredd_disabled;
  59core_param(novmcoredd, vmcoredd_disabled, bool, 0);
  60#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
  61
  62/* Device Dump Size */
  63static size_t vmcoredd_orig_sz;
  64
  65static DEFINE_SPINLOCK(vmcore_cb_lock);
  66DEFINE_STATIC_SRCU(vmcore_cb_srcu);
  67/* List of registered vmcore callbacks. */
  68static LIST_HEAD(vmcore_cb_list);
  69/* Whether the vmcore has been opened once. */
  70static bool vmcore_opened;
  71
  72void 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}
  85EXPORT_SYMBOL_GPL(register_vmcore_cb);
  86
  87void 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}
 102EXPORT_SYMBOL_GPL(unregister_vmcore_cb);
 103
 104static 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
 121static 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. */
 131ssize_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 */
 183int __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 */
 191void __weak elfcorehdr_free(unsigned long long addr)
 192{}
 193
 194/*
 195 * Architectures may override this function to read from ELF header
 196 */
 197ssize_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 */
 210ssize_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 */
 224int __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 */
 235ssize_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
 242static 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
 270out_unlock:
 271	mutex_unlock(&vmcoredd_mutex);
 272	return ret;
 273}
 274
 275#ifdef CONFIG_MMU
 276static 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
 307out_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 */
 317static 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
 410static 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 */
 422static 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
 458static 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 */
 472static 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 */
 501static 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;
 546fail:
 547	do_munmap(vma->vm_mm, from, len, NULL);
 548	return -EAGAIN;
 549}
 550
 551static 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
 570static 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;
 678fail:
 679	do_munmap(vma->vm_mm, vma->vm_start, len, NULL);
 680	return -EAGAIN;
 681}
 682#else
 683static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 684{
 685	return -ENOSYS;
 686}
 687#endif
 688
 689static 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
 696static struct vmcore* __init get_new_element(void)
 697{
 698	return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
 699}
 700
 701static 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 */
 723static 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 */
 786static 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 */
 821static 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. */
 844static 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 */
 914static 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 */
 977static 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 */
1012static 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. */
1035static 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. */
1098static 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
1141static 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. */
1185static 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
1200static 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
1208static 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;
1259fail:
1260	free_elfcorebuf();
1261	return rc;
1262}
1263
1264static 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;
1314fail:
1315	free_elfcorebuf();
1316	return rc;
1317}
1318
1319static 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 */
1364static 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 */
1387static 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 */
1449static 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 */
1472int 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
1525out_err:
1526	vfree(buf);
1527	vfree(dump);
1528
1529	return ret;
1530}
1531EXPORT_SYMBOL(vmcore_add_device_dump);
1532#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1533
1534/* Free all dumps in vmcore device dump list */
1535static 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. */
1553static 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}
1581fs_initcall(vmcore_init);
1582
1583/* Cleanup function for vmcore module. */
1584void 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}