Linux Audio

Check our new training course

Loading...
v6.13.7
   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		cond_resched();
 388	}
 389
 390	list_for_each_entry(m, &vmcore_list, list) {
 391		if (*fpos < m->offset + m->size) {
 392			tsz = (size_t)min_t(unsigned long long,
 393					    m->offset + m->size - *fpos,
 394					    iov_iter_count(iter));
 395			start = m->paddr + *fpos - m->offset;
 396			tmp = read_from_oldmem(iter, tsz, &start,
 397					cc_platform_has(CC_ATTR_MEM_ENCRYPT));
 398			if (tmp < 0)
 399				return tmp;
 
 400			*fpos += tsz;
 
 401			acc += tsz;
 402
 403			/* leave now if filled buffer already */
 404			if (!iov_iter_count(iter))
 405				return acc;
 406		}
 407
 408		cond_resched();
 409	}
 410
 411	return acc;
 412}
 413
 414static ssize_t read_vmcore(struct kiocb *iocb, struct iov_iter *iter)
 
 415{
 416	return __read_vmcore(iter, &iocb->ki_pos);
 417}
 418
 419/**
 420 * vmcore_alloc_buf - allocate buffer in vmalloc memory
 421 * @size: size of buffer
 422 *
 423 * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
 424 * the buffer to user-space by means of remap_vmalloc_range().
 425 *
 426 * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
 427 * disabled and there's no need to allow users to mmap the buffer.
 428 */
 429static inline char *vmcore_alloc_buf(size_t size)
 430{
 431#ifdef CONFIG_MMU
 432	return vmalloc_user(size);
 433#else
 434	return vzalloc(size);
 435#endif
 436}
 437
 438/*
 439 * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
 440 * essential for mmap_vmcore() in order to map physically
 441 * non-contiguous objects (ELF header, ELF note segment and memory
 442 * regions in the 1st kernel pointed to by PT_LOAD entries) into
 443 * virtually contiguous user-space in ELF layout.
 444 */
 445#ifdef CONFIG_MMU
 446
 447/*
 448 * The vmcore fault handler uses the page cache and fills data using the
 449 * standard __read_vmcore() function.
 450 *
 451 * On s390 the fault handler is used for memory regions that can't be mapped
 452 * directly with remap_pfn_range().
 453 */
 454static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
 455{
 456#ifdef CONFIG_S390
 457	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
 458	pgoff_t index = vmf->pgoff;
 459	struct iov_iter iter;
 460	struct kvec kvec;
 461	struct page *page;
 462	loff_t offset;
 
 463	int rc;
 464
 465	page = find_or_create_page(mapping, index, GFP_KERNEL);
 466	if (!page)
 467		return VM_FAULT_OOM;
 468	if (!PageUptodate(page)) {
 469		offset = (loff_t) index << PAGE_SHIFT;
 470		kvec.iov_base = page_address(page);
 471		kvec.iov_len = PAGE_SIZE;
 472		iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, PAGE_SIZE);
 473
 474		rc = __read_vmcore(&iter, &offset);
 475		if (rc < 0) {
 476			unlock_page(page);
 477			put_page(page);
 478			return vmf_error(rc);
 479		}
 480		SetPageUptodate(page);
 481	}
 482	unlock_page(page);
 483	vmf->page = page;
 484	return 0;
 485#else
 486	return VM_FAULT_SIGBUS;
 487#endif
 488}
 489
 490static const struct vm_operations_struct vmcore_mmap_ops = {
 491	.fault = mmap_vmcore_fault,
 492};
 493
 494/*
 495 * remap_oldmem_pfn_checked - do remap_oldmem_pfn_range replacing all pages
 496 * reported as not being ram with the zero page.
 497 *
 498 * @vma: vm_area_struct describing requested mapping
 499 * @from: start remapping from
 500 * @pfn: page frame number to start remapping to
 501 * @size: remapping size
 502 * @prot: protection bits
 503 *
 504 * Returns zero on success, -EAGAIN on failure.
 
 
 
 
 505 */
 506static int remap_oldmem_pfn_checked(struct vm_area_struct *vma,
 507				    unsigned long from, unsigned long pfn,
 508				    unsigned long size, pgprot_t prot)
 509{
 510	unsigned long map_size;
 511	unsigned long pos_start, pos_end, pos;
 512	unsigned long zeropage_pfn = my_zero_pfn(0);
 513	size_t len = 0;
 514
 515	pos_start = pfn;
 516	pos_end = pfn + (size >> PAGE_SHIFT);
 517
 518	for (pos = pos_start; pos < pos_end; ++pos) {
 519		if (!pfn_is_ram(pos)) {
 520			/*
 521			 * We hit a page which is not ram. Remap the continuous
 522			 * region between pos_start and pos-1 and replace
 523			 * the non-ram page at pos with the zero page.
 524			 */
 525			if (pos > pos_start) {
 526				/* Remap continuous region */
 527				map_size = (pos - pos_start) << PAGE_SHIFT;
 528				if (remap_oldmem_pfn_range(vma, from + len,
 529							   pos_start, map_size,
 530							   prot))
 531					goto fail;
 532				len += map_size;
 533			}
 534			/* Remap the zero page */
 535			if (remap_oldmem_pfn_range(vma, from + len,
 536						   zeropage_pfn,
 537						   PAGE_SIZE, prot))
 538				goto fail;
 539			len += PAGE_SIZE;
 540			pos_start = pos + 1;
 541		}
 542	}
 543	if (pos > pos_start) {
 544		/* Remap the rest */
 545		map_size = (pos - pos_start) << PAGE_SHIFT;
 546		if (remap_oldmem_pfn_range(vma, from + len, pos_start,
 547					   map_size, prot))
 548			goto fail;
 549	}
 550	return 0;
 551fail:
 552	do_munmap(vma->vm_mm, from, len, NULL);
 553	return -EAGAIN;
 554}
 555
 556static int vmcore_remap_oldmem_pfn(struct vm_area_struct *vma,
 557			    unsigned long from, unsigned long pfn,
 558			    unsigned long size, pgprot_t prot)
 559{
 560	int ret, idx;
 561
 562	/*
 563	 * Check if a callback was registered to avoid looping over all
 564	 * pages without a reason.
 565	 */
 566	idx = srcu_read_lock(&vmcore_cb_srcu);
 567	if (!list_empty(&vmcore_cb_list))
 568		ret = remap_oldmem_pfn_checked(vma, from, pfn, size, prot);
 569	else
 570		ret = remap_oldmem_pfn_range(vma, from, pfn, size, prot);
 571	srcu_read_unlock(&vmcore_cb_srcu, idx);
 572	return ret;
 573}
 574
 
 
 
 
 
 
 
 
 575static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 576{
 577	size_t size = vma->vm_end - vma->vm_start;
 578	u64 start, end, len, tsz;
 579	struct vmcore *m;
 580
 581	start = (u64)vma->vm_pgoff << PAGE_SHIFT;
 582	end = start + size;
 583
 584	if (size > vmcore_size || end > vmcore_size)
 585		return -EINVAL;
 586
 587	if (vma->vm_flags & (VM_WRITE | VM_EXEC))
 588		return -EPERM;
 589
 590	vm_flags_mod(vma, VM_MIXEDMAP, VM_MAYWRITE | VM_MAYEXEC);
 
 591	vma->vm_ops = &vmcore_mmap_ops;
 592
 593	len = 0;
 594
 595	if (start < elfcorebuf_sz) {
 596		u64 pfn;
 597
 598		tsz = min(elfcorebuf_sz - (size_t)start, size);
 599		pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
 600		if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
 601				    vma->vm_page_prot))
 602			return -EAGAIN;
 603		size -= tsz;
 604		start += tsz;
 605		len += tsz;
 606
 607		if (size == 0)
 608			return 0;
 609	}
 610
 611	if (start < elfcorebuf_sz + elfnotes_sz) {
 612		void *kaddr;
 613
 614		/* We add device dumps before other elf notes because the
 615		 * other elf notes may not fill the elf notes buffer
 616		 * completely and we will end up with zero-filled data
 617		 * between the elf notes and the device dumps. Tools will
 618		 * then try to decode this zero-filled data as valid notes
 619		 * and we don't want that. Hence, adding device dumps before
 620		 * the other elf notes ensure that zero-filled data can be
 621		 * avoided. This also ensures that the device dumps and
 622		 * other elf notes can be properly mmaped at page aligned
 623		 * address.
 624		 */
 625#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
 626		/* Read device dumps */
 627		if (start < elfcorebuf_sz + vmcoredd_orig_sz) {
 628			u64 start_off;
 629
 630			tsz = min(elfcorebuf_sz + vmcoredd_orig_sz -
 631				  (size_t)start, size);
 632			start_off = start - elfcorebuf_sz;
 633			if (vmcoredd_mmap_dumps(vma, vma->vm_start + len,
 634						start_off, tsz))
 635				goto fail;
 636
 637			size -= tsz;
 638			start += tsz;
 639			len += tsz;
 640
 641			/* leave now if filled buffer already */
 642			if (!size)
 643				return 0;
 644		}
 645#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
 646
 647		/* Read remaining elf notes */
 648		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
 649		kaddr = elfnotes_buf + start - elfcorebuf_sz - vmcoredd_orig_sz;
 650		if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
 651						kaddr, 0, tsz))
 652			goto fail;
 653
 654		size -= tsz;
 655		start += tsz;
 656		len += tsz;
 657
 658		if (size == 0)
 659			return 0;
 660	}
 661
 662	list_for_each_entry(m, &vmcore_list, list) {
 663		if (start < m->offset + m->size) {
 664			u64 paddr = 0;
 665
 666			tsz = (size_t)min_t(unsigned long long,
 667					    m->offset + m->size - start, size);
 668			paddr = m->paddr + start - m->offset;
 669			if (vmcore_remap_oldmem_pfn(vma, vma->vm_start + len,
 670						    paddr >> PAGE_SHIFT, tsz,
 671						    vma->vm_page_prot))
 672				goto fail;
 673			size -= tsz;
 674			start += tsz;
 675			len += tsz;
 676
 677			if (size == 0)
 678				return 0;
 679		}
 680	}
 681
 682	return 0;
 683fail:
 684	do_munmap(vma->vm_mm, vma->vm_start, len, NULL);
 685	return -EAGAIN;
 686}
 687#else
 688static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 689{
 690	return -ENOSYS;
 691}
 692#endif
 693
 694static const struct proc_ops vmcore_proc_ops = {
 695	.proc_open	= open_vmcore,
 696	.proc_read_iter	= read_vmcore,
 697	.proc_lseek	= default_llseek,
 698	.proc_mmap	= mmap_vmcore,
 699};
 700
 701static struct vmcore* __init get_new_element(void)
 702{
 703	return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
 704}
 705
 706static u64 get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
 707			   struct list_head *vc_list)
 708{
 709	u64 size;
 710	struct vmcore *m;
 711
 712	size = elfsz + elfnotesegsz;
 713	list_for_each_entry(m, vc_list, list) {
 714		size += m->size;
 715	}
 716	return size;
 717}
 718
 719/**
 720 * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
 721 *
 722 * @ehdr_ptr: ELF header
 723 *
 724 * This function updates p_memsz member of each PT_NOTE entry in the
 725 * program header table pointed to by @ehdr_ptr to real size of ELF
 726 * note segment.
 727 */
 728static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
 729{
 730	int i, rc=0;
 731	Elf64_Phdr *phdr_ptr;
 732	Elf64_Nhdr *nhdr_ptr;
 733
 734	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
 735	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 736		void *notes_section;
 737		u64 offset, max_sz, sz, real_sz = 0;
 738		if (phdr_ptr->p_type != PT_NOTE)
 739			continue;
 740		max_sz = phdr_ptr->p_memsz;
 741		offset = phdr_ptr->p_offset;
 742		notes_section = kmalloc(max_sz, GFP_KERNEL);
 743		if (!notes_section)
 744			return -ENOMEM;
 745		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
 746		if (rc < 0) {
 747			kfree(notes_section);
 748			return rc;
 749		}
 750		nhdr_ptr = notes_section;
 751		while (nhdr_ptr->n_namesz != 0) {
 752			sz = sizeof(Elf64_Nhdr) +
 753				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
 754				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
 755			if ((real_sz + sz) > max_sz) {
 756				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
 757					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
 758				break;
 759			}
 760			real_sz += sz;
 761			nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
 762		}
 763		kfree(notes_section);
 764		phdr_ptr->p_memsz = real_sz;
 765		if (real_sz == 0) {
 766			pr_warn("Warning: Zero PT_NOTE entries found\n");
 767		}
 768	}
 769
 770	return 0;
 771}
 772
 773/**
 774 * get_note_number_and_size_elf64 - get the number of PT_NOTE program
 775 * headers and sum of real size of their ELF note segment headers and
 776 * data.
 777 *
 778 * @ehdr_ptr: ELF header
 779 * @nr_ptnote: buffer for the number of PT_NOTE program headers
 780 * @sz_ptnote: buffer for size of unique PT_NOTE program header
 781 *
 782 * This function is used to merge multiple PT_NOTE program headers
 783 * into a unique single one. The resulting unique entry will have
 784 * @sz_ptnote in its phdr->p_mem.
 785 *
 786 * It is assumed that program headers with PT_NOTE type pointed to by
 787 * @ehdr_ptr has already been updated by update_note_header_size_elf64
 788 * and each of PT_NOTE program headers has actual ELF note segment
 789 * size in its p_memsz member.
 790 */
 791static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
 792						 int *nr_ptnote, u64 *sz_ptnote)
 793{
 794	int i;
 795	Elf64_Phdr *phdr_ptr;
 796
 797	*nr_ptnote = *sz_ptnote = 0;
 798
 799	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
 800	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 801		if (phdr_ptr->p_type != PT_NOTE)
 802			continue;
 803		*nr_ptnote += 1;
 804		*sz_ptnote += phdr_ptr->p_memsz;
 805	}
 806
 807	return 0;
 808}
 809
 810/**
 811 * copy_notes_elf64 - copy ELF note segments in a given buffer
 812 *
 813 * @ehdr_ptr: ELF header
 814 * @notes_buf: buffer into which ELF note segments are copied
 815 *
 816 * This function is used to copy ELF note segment in the 1st kernel
 817 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
 818 * size of the buffer @notes_buf is equal to or larger than sum of the
 819 * real ELF note segment headers and data.
 820 *
 821 * It is assumed that program headers with PT_NOTE type pointed to by
 822 * @ehdr_ptr has already been updated by update_note_header_size_elf64
 823 * and each of PT_NOTE program headers has actual ELF note segment
 824 * size in its p_memsz member.
 825 */
 826static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
 827{
 828	int i, rc=0;
 829	Elf64_Phdr *phdr_ptr;
 830
 831	phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
 832
 833	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 834		u64 offset;
 835		if (phdr_ptr->p_type != PT_NOTE)
 836			continue;
 837		offset = phdr_ptr->p_offset;
 838		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
 839					   &offset);
 840		if (rc < 0)
 841			return rc;
 842		notes_buf += phdr_ptr->p_memsz;
 843	}
 844
 845	return 0;
 846}
 847
 848/* Merges all the PT_NOTE headers into one. */
 849static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
 850					   char **notes_buf, size_t *notes_sz)
 851{
 852	int i, nr_ptnote=0, rc=0;
 853	char *tmp;
 854	Elf64_Ehdr *ehdr_ptr;
 855	Elf64_Phdr phdr;
 856	u64 phdr_sz = 0, note_off;
 857
 858	ehdr_ptr = (Elf64_Ehdr *)elfptr;
 859
 860	rc = update_note_header_size_elf64(ehdr_ptr);
 861	if (rc < 0)
 862		return rc;
 863
 864	rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
 865	if (rc < 0)
 866		return rc;
 867
 868	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
 869	*notes_buf = vmcore_alloc_buf(*notes_sz);
 870	if (!*notes_buf)
 871		return -ENOMEM;
 872
 873	rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
 874	if (rc < 0)
 875		return rc;
 876
 877	/* Prepare merged PT_NOTE program header. */
 878	phdr.p_type    = PT_NOTE;
 879	phdr.p_flags   = 0;
 880	note_off = sizeof(Elf64_Ehdr) +
 881			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
 882	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
 883	phdr.p_vaddr   = phdr.p_paddr = 0;
 884	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
 885	phdr.p_align   = 4;
 886
 887	/* Add merged PT_NOTE program header*/
 888	tmp = elfptr + sizeof(Elf64_Ehdr);
 889	memcpy(tmp, &phdr, sizeof(phdr));
 890	tmp += sizeof(phdr);
 891
 892	/* Remove unwanted PT_NOTE program headers. */
 893	i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
 894	*elfsz = *elfsz - i;
 895	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
 896	memset(elfptr + *elfsz, 0, i);
 897	*elfsz = roundup(*elfsz, PAGE_SIZE);
 898
 899	/* Modify e_phnum to reflect merged headers. */
 900	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
 901
 902	/* Store the size of all notes.  We need this to update the note
 903	 * header when the device dumps will be added.
 904	 */
 905	elfnotes_orig_sz = phdr.p_memsz;
 906
 907	return 0;
 908}
 909
 910/**
 911 * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
 912 *
 913 * @ehdr_ptr: ELF header
 914 *
 915 * This function updates p_memsz member of each PT_NOTE entry in the
 916 * program header table pointed to by @ehdr_ptr to real size of ELF
 917 * note segment.
 918 */
 919static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
 920{
 921	int i, rc=0;
 922	Elf32_Phdr *phdr_ptr;
 923	Elf32_Nhdr *nhdr_ptr;
 924
 925	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
 926	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 927		void *notes_section;
 928		u64 offset, max_sz, sz, real_sz = 0;
 929		if (phdr_ptr->p_type != PT_NOTE)
 930			continue;
 931		max_sz = phdr_ptr->p_memsz;
 932		offset = phdr_ptr->p_offset;
 933		notes_section = kmalloc(max_sz, GFP_KERNEL);
 934		if (!notes_section)
 935			return -ENOMEM;
 936		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
 937		if (rc < 0) {
 938			kfree(notes_section);
 939			return rc;
 940		}
 941		nhdr_ptr = notes_section;
 942		while (nhdr_ptr->n_namesz != 0) {
 943			sz = sizeof(Elf32_Nhdr) +
 944				(((u64)nhdr_ptr->n_namesz + 3) & ~3) +
 945				(((u64)nhdr_ptr->n_descsz + 3) & ~3);
 946			if ((real_sz + sz) > max_sz) {
 947				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
 948					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
 949				break;
 950			}
 951			real_sz += sz;
 952			nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
 953		}
 954		kfree(notes_section);
 955		phdr_ptr->p_memsz = real_sz;
 956		if (real_sz == 0) {
 957			pr_warn("Warning: Zero PT_NOTE entries found\n");
 958		}
 959	}
 960
 961	return 0;
 962}
 963
 964/**
 965 * get_note_number_and_size_elf32 - get the number of PT_NOTE program
 966 * headers and sum of real size of their ELF note segment headers and
 967 * data.
 968 *
 969 * @ehdr_ptr: ELF header
 970 * @nr_ptnote: buffer for the number of PT_NOTE program headers
 971 * @sz_ptnote: buffer for size of unique PT_NOTE program header
 972 *
 973 * This function is used to merge multiple PT_NOTE program headers
 974 * into a unique single one. The resulting unique entry will have
 975 * @sz_ptnote in its phdr->p_mem.
 976 *
 977 * It is assumed that program headers with PT_NOTE type pointed to by
 978 * @ehdr_ptr has already been updated by update_note_header_size_elf32
 979 * and each of PT_NOTE program headers has actual ELF note segment
 980 * size in its p_memsz member.
 981 */
 982static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
 983						 int *nr_ptnote, u64 *sz_ptnote)
 984{
 985	int i;
 986	Elf32_Phdr *phdr_ptr;
 987
 988	*nr_ptnote = *sz_ptnote = 0;
 989
 990	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
 991	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 992		if (phdr_ptr->p_type != PT_NOTE)
 993			continue;
 994		*nr_ptnote += 1;
 995		*sz_ptnote += phdr_ptr->p_memsz;
 996	}
 997
 998	return 0;
 999}
1000
1001/**
1002 * copy_notes_elf32 - copy ELF note segments in a given buffer
1003 *
1004 * @ehdr_ptr: ELF header
1005 * @notes_buf: buffer into which ELF note segments are copied
1006 *
1007 * This function is used to copy ELF note segment in the 1st kernel
1008 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
1009 * size of the buffer @notes_buf is equal to or larger than sum of the
1010 * real ELF note segment headers and data.
1011 *
1012 * It is assumed that program headers with PT_NOTE type pointed to by
1013 * @ehdr_ptr has already been updated by update_note_header_size_elf32
1014 * and each of PT_NOTE program headers has actual ELF note segment
1015 * size in its p_memsz member.
1016 */
1017static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
1018{
1019	int i, rc=0;
1020	Elf32_Phdr *phdr_ptr;
1021
1022	phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
1023
1024	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1025		u64 offset;
1026		if (phdr_ptr->p_type != PT_NOTE)
1027			continue;
1028		offset = phdr_ptr->p_offset;
1029		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
1030					   &offset);
1031		if (rc < 0)
1032			return rc;
1033		notes_buf += phdr_ptr->p_memsz;
1034	}
1035
1036	return 0;
1037}
1038
1039/* Merges all the PT_NOTE headers into one. */
1040static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
1041					   char **notes_buf, size_t *notes_sz)
1042{
1043	int i, nr_ptnote=0, rc=0;
1044	char *tmp;
1045	Elf32_Ehdr *ehdr_ptr;
1046	Elf32_Phdr phdr;
1047	u64 phdr_sz = 0, note_off;
1048
1049	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1050
1051	rc = update_note_header_size_elf32(ehdr_ptr);
1052	if (rc < 0)
1053		return rc;
1054
1055	rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
1056	if (rc < 0)
1057		return rc;
1058
1059	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
1060	*notes_buf = vmcore_alloc_buf(*notes_sz);
1061	if (!*notes_buf)
1062		return -ENOMEM;
1063
1064	rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
1065	if (rc < 0)
1066		return rc;
1067
1068	/* Prepare merged PT_NOTE program header. */
1069	phdr.p_type    = PT_NOTE;
1070	phdr.p_flags   = 0;
1071	note_off = sizeof(Elf32_Ehdr) +
1072			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
1073	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
1074	phdr.p_vaddr   = phdr.p_paddr = 0;
1075	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
1076	phdr.p_align   = 4;
1077
1078	/* Add merged PT_NOTE program header*/
1079	tmp = elfptr + sizeof(Elf32_Ehdr);
1080	memcpy(tmp, &phdr, sizeof(phdr));
1081	tmp += sizeof(phdr);
1082
1083	/* Remove unwanted PT_NOTE program headers. */
1084	i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
1085	*elfsz = *elfsz - i;
1086	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
1087	memset(elfptr + *elfsz, 0, i);
1088	*elfsz = roundup(*elfsz, PAGE_SIZE);
1089
1090	/* Modify e_phnum to reflect merged headers. */
1091	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
1092
1093	/* Store the size of all notes.  We need this to update the note
1094	 * header when the device dumps will be added.
1095	 */
1096	elfnotes_orig_sz = phdr.p_memsz;
1097
1098	return 0;
1099}
1100
1101/* Add memory chunks represented by program headers to vmcore list. Also update
1102 * the new offset fields of exported program headers. */
1103static int __init process_ptload_program_headers_elf64(char *elfptr,
1104						size_t elfsz,
1105						size_t elfnotes_sz,
1106						struct list_head *vc_list)
1107{
1108	int i;
1109	Elf64_Ehdr *ehdr_ptr;
1110	Elf64_Phdr *phdr_ptr;
1111	loff_t vmcore_off;
1112	struct vmcore *new;
1113
1114	ehdr_ptr = (Elf64_Ehdr *)elfptr;
1115	phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
1116
1117	/* Skip ELF header, program headers and ELF note segment. */
1118	vmcore_off = elfsz + elfnotes_sz;
1119
1120	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1121		u64 paddr, start, end, size;
1122
1123		if (phdr_ptr->p_type != PT_LOAD)
1124			continue;
1125
1126		paddr = phdr_ptr->p_offset;
1127		start = rounddown(paddr, PAGE_SIZE);
1128		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1129		size = end - start;
1130
1131		/* Add this contiguous chunk of memory to vmcore list.*/
1132		new = get_new_element();
1133		if (!new)
1134			return -ENOMEM;
1135		new->paddr = start;
1136		new->size = size;
1137		list_add_tail(&new->list, vc_list);
1138
1139		/* Update the program header offset. */
1140		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1141		vmcore_off = vmcore_off + size;
1142	}
1143	return 0;
1144}
1145
1146static int __init process_ptload_program_headers_elf32(char *elfptr,
1147						size_t elfsz,
1148						size_t elfnotes_sz,
1149						struct list_head *vc_list)
1150{
1151	int i;
1152	Elf32_Ehdr *ehdr_ptr;
1153	Elf32_Phdr *phdr_ptr;
1154	loff_t vmcore_off;
1155	struct vmcore *new;
1156
1157	ehdr_ptr = (Elf32_Ehdr *)elfptr;
1158	phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
1159
1160	/* Skip ELF header, program headers and ELF note segment. */
1161	vmcore_off = elfsz + elfnotes_sz;
1162
1163	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
1164		u64 paddr, start, end, size;
1165
1166		if (phdr_ptr->p_type != PT_LOAD)
1167			continue;
1168
1169		paddr = phdr_ptr->p_offset;
1170		start = rounddown(paddr, PAGE_SIZE);
1171		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
1172		size = end - start;
1173
1174		/* Add this contiguous chunk of memory to vmcore list.*/
1175		new = get_new_element();
1176		if (!new)
1177			return -ENOMEM;
1178		new->paddr = start;
1179		new->size = size;
1180		list_add_tail(&new->list, vc_list);
1181
1182		/* Update the program header offset */
1183		phdr_ptr->p_offset = vmcore_off + (paddr - start);
1184		vmcore_off = vmcore_off + size;
1185	}
1186	return 0;
1187}
1188
1189/* Sets offset fields of vmcore elements. */
1190static void set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
1191				    struct list_head *vc_list)
1192{
1193	loff_t vmcore_off;
1194	struct vmcore *m;
1195
1196	/* Skip ELF header, program headers and ELF note segment. */
1197	vmcore_off = elfsz + elfnotes_sz;
1198
1199	list_for_each_entry(m, vc_list, list) {
1200		m->offset = vmcore_off;
1201		vmcore_off += m->size;
1202	}
1203}
1204
1205static void free_elfcorebuf(void)
1206{
1207	free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
1208	elfcorebuf = NULL;
1209	vfree(elfnotes_buf);
1210	elfnotes_buf = NULL;
1211}
1212
1213static int __init parse_crash_elf64_headers(void)
1214{
1215	int rc=0;
1216	Elf64_Ehdr ehdr;
1217	u64 addr;
1218
1219	addr = elfcorehdr_addr;
1220
1221	/* Read ELF header */
1222	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
1223	if (rc < 0)
1224		return rc;
1225
1226	/* Do some basic Verification. */
1227	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1228		(ehdr.e_type != ET_CORE) ||
1229		!vmcore_elf64_check_arch(&ehdr) ||
1230		ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
1231		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1232		ehdr.e_version != EV_CURRENT ||
1233		ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
1234		ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
1235		ehdr.e_phnum == 0) {
1236		pr_warn("Warning: Core image elf header is not sane\n");
1237		return -EINVAL;
1238	}
1239
1240	/* Read in all elf headers. */
1241	elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
1242				ehdr.e_phnum * sizeof(Elf64_Phdr);
1243	elfcorebuf_sz = elfcorebuf_sz_orig;
1244	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1245					      get_order(elfcorebuf_sz_orig));
1246	if (!elfcorebuf)
1247		return -ENOMEM;
1248	addr = elfcorehdr_addr;
1249	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1250	if (rc < 0)
1251		goto fail;
1252
1253	/* Merge all PT_NOTE headers into one. */
1254	rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
1255				      &elfnotes_buf, &elfnotes_sz);
1256	if (rc)
1257		goto fail;
1258	rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
1259						  elfnotes_sz, &vmcore_list);
1260	if (rc)
1261		goto fail;
1262	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1263	return 0;
1264fail:
1265	free_elfcorebuf();
1266	return rc;
1267}
1268
1269static int __init parse_crash_elf32_headers(void)
1270{
1271	int rc=0;
1272	Elf32_Ehdr ehdr;
1273	u64 addr;
1274
1275	addr = elfcorehdr_addr;
1276
1277	/* Read ELF header */
1278	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
1279	if (rc < 0)
1280		return rc;
1281
1282	/* Do some basic Verification. */
1283	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
1284		(ehdr.e_type != ET_CORE) ||
1285		!vmcore_elf32_check_arch(&ehdr) ||
1286		ehdr.e_ident[EI_CLASS] != ELFCLASS32||
1287		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
1288		ehdr.e_version != EV_CURRENT ||
1289		ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1290		ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1291		ehdr.e_phnum == 0) {
1292		pr_warn("Warning: Core image elf header is not sane\n");
1293		return -EINVAL;
1294	}
1295
1296	/* Read in all elf headers. */
1297	elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1298	elfcorebuf_sz = elfcorebuf_sz_orig;
1299	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1300					      get_order(elfcorebuf_sz_orig));
1301	if (!elfcorebuf)
1302		return -ENOMEM;
1303	addr = elfcorehdr_addr;
1304	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1305	if (rc < 0)
1306		goto fail;
1307
1308	/* Merge all PT_NOTE headers into one. */
1309	rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1310				      &elfnotes_buf, &elfnotes_sz);
1311	if (rc)
1312		goto fail;
1313	rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1314						  elfnotes_sz, &vmcore_list);
1315	if (rc)
1316		goto fail;
1317	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1318	return 0;
1319fail:
1320	free_elfcorebuf();
1321	return rc;
1322}
1323
1324static int __init parse_crash_elf_headers(void)
1325{
1326	unsigned char e_ident[EI_NIDENT];
1327	u64 addr;
1328	int rc=0;
1329
1330	addr = elfcorehdr_addr;
1331	rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1332	if (rc < 0)
1333		return rc;
1334	if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1335		pr_warn("Warning: Core image elf header not found\n");
1336		return -EINVAL;
1337	}
1338
1339	if (e_ident[EI_CLASS] == ELFCLASS64) {
1340		rc = parse_crash_elf64_headers();
1341		if (rc)
1342			return rc;
1343	} else if (e_ident[EI_CLASS] == ELFCLASS32) {
1344		rc = parse_crash_elf32_headers();
1345		if (rc)
1346			return rc;
1347	} else {
1348		pr_warn("Warning: Core image elf header is not sane\n");
1349		return -EINVAL;
1350	}
1351
1352	/* Determine vmcore size. */
1353	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1354				      &vmcore_list);
1355
1356	return 0;
1357}
1358
1359#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1360/**
1361 * vmcoredd_write_header - Write vmcore device dump header at the
1362 * beginning of the dump's buffer.
1363 * @buf: Output buffer where the note is written
1364 * @data: Dump info
1365 * @size: Size of the dump
1366 *
1367 * Fills beginning of the dump's buffer with vmcore device dump header.
1368 */
1369static void vmcoredd_write_header(void *buf, struct vmcoredd_data *data,
1370				  u32 size)
1371{
1372	struct vmcoredd_header *vdd_hdr = (struct vmcoredd_header *)buf;
1373
1374	vdd_hdr->n_namesz = sizeof(vdd_hdr->name);
1375	vdd_hdr->n_descsz = size + sizeof(vdd_hdr->dump_name);
1376	vdd_hdr->n_type = NT_VMCOREDD;
1377
1378	strscpy_pad(vdd_hdr->name, VMCOREDD_NOTE_NAME);
1379	strscpy_pad(vdd_hdr->dump_name, data->dump_name);
1380}
1381
1382/**
1383 * vmcoredd_update_program_headers - Update all ELF program headers
1384 * @elfptr: Pointer to elf header
1385 * @elfnotesz: Size of elf notes aligned to page size
1386 * @vmcoreddsz: Size of device dumps to be added to elf note header
1387 *
1388 * Determine type of ELF header (Elf64 or Elf32) and update the elf note size.
1389 * Also update the offsets of all the program headers after the elf note header.
1390 */
1391static void vmcoredd_update_program_headers(char *elfptr, size_t elfnotesz,
1392					    size_t vmcoreddsz)
1393{
1394	unsigned char *e_ident = (unsigned char *)elfptr;
1395	u64 start, end, size;
1396	loff_t vmcore_off;
1397	u32 i;
1398
1399	vmcore_off = elfcorebuf_sz + elfnotesz;
1400
1401	if (e_ident[EI_CLASS] == ELFCLASS64) {
1402		Elf64_Ehdr *ehdr = (Elf64_Ehdr *)elfptr;
1403		Elf64_Phdr *phdr = (Elf64_Phdr *)(elfptr + sizeof(Elf64_Ehdr));
1404
1405		/* Update all program headers */
1406		for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1407			if (phdr->p_type == PT_NOTE) {
1408				/* Update note size */
1409				phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1410				phdr->p_filesz = phdr->p_memsz;
1411				continue;
1412			}
1413
1414			start = rounddown(phdr->p_offset, PAGE_SIZE);
1415			end = roundup(phdr->p_offset + phdr->p_memsz,
1416				      PAGE_SIZE);
1417			size = end - start;
1418			phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1419			vmcore_off += size;
1420		}
1421	} else {
1422		Elf32_Ehdr *ehdr = (Elf32_Ehdr *)elfptr;
1423		Elf32_Phdr *phdr = (Elf32_Phdr *)(elfptr + sizeof(Elf32_Ehdr));
1424
1425		/* Update all program headers */
1426		for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
1427			if (phdr->p_type == PT_NOTE) {
1428				/* Update note size */
1429				phdr->p_memsz = elfnotes_orig_sz + vmcoreddsz;
1430				phdr->p_filesz = phdr->p_memsz;
1431				continue;
1432			}
1433
1434			start = rounddown(phdr->p_offset, PAGE_SIZE);
1435			end = roundup(phdr->p_offset + phdr->p_memsz,
1436				      PAGE_SIZE);
1437			size = end - start;
1438			phdr->p_offset = vmcore_off + (phdr->p_offset - start);
1439			vmcore_off += size;
1440		}
1441	}
1442}
1443
1444/**
1445 * vmcoredd_update_size - Update the total size of the device dumps and update
1446 * ELF header
1447 * @dump_size: Size of the current device dump to be added to total size
1448 *
1449 * Update the total size of all the device dumps and update the ELF program
1450 * headers. Calculate the new offsets for the vmcore list and update the
1451 * total vmcore size.
1452 */
1453static void vmcoredd_update_size(size_t dump_size)
1454{
1455	vmcoredd_orig_sz += dump_size;
1456	elfnotes_sz = roundup(elfnotes_orig_sz, PAGE_SIZE) + vmcoredd_orig_sz;
1457	vmcoredd_update_program_headers(elfcorebuf, elfnotes_sz,
1458					vmcoredd_orig_sz);
1459
1460	/* Update vmcore list offsets */
1461	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1462
1463	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1464				      &vmcore_list);
1465	proc_vmcore->size = vmcore_size;
1466}
1467
1468/**
1469 * vmcore_add_device_dump - Add a buffer containing device dump to vmcore
1470 * @data: dump info.
1471 *
1472 * Allocate a buffer and invoke the calling driver's dump collect routine.
1473 * Write ELF note at the beginning of the buffer to indicate vmcore device
1474 * dump and add the dump to global list.
1475 */
1476int vmcore_add_device_dump(struct vmcoredd_data *data)
1477{
1478	struct vmcoredd_node *dump;
1479	void *buf = NULL;
1480	size_t data_size;
1481	int ret;
1482
1483	if (vmcoredd_disabled) {
1484		pr_err_once("Device dump is disabled\n");
1485		return -EINVAL;
1486	}
1487
1488	if (!data || !strlen(data->dump_name) ||
1489	    !data->vmcoredd_callback || !data->size)
1490		return -EINVAL;
1491
1492	dump = vzalloc(sizeof(*dump));
1493	if (!dump) {
1494		ret = -ENOMEM;
1495		goto out_err;
1496	}
1497
1498	/* Keep size of the buffer page aligned so that it can be mmaped */
1499	data_size = roundup(sizeof(struct vmcoredd_header) + data->size,
1500			    PAGE_SIZE);
1501
1502	/* Allocate buffer for driver's to write their dumps */
1503	buf = vmcore_alloc_buf(data_size);
1504	if (!buf) {
1505		ret = -ENOMEM;
1506		goto out_err;
1507	}
1508
1509	vmcoredd_write_header(buf, data, data_size -
1510			      sizeof(struct vmcoredd_header));
1511
1512	/* Invoke the driver's dump collection routing */
1513	ret = data->vmcoredd_callback(data, buf +
1514				      sizeof(struct vmcoredd_header));
1515	if (ret)
1516		goto out_err;
1517
1518	dump->buf = buf;
1519	dump->size = data_size;
1520
1521	/* Add the dump to driver sysfs list */
1522	mutex_lock(&vmcoredd_mutex);
1523	list_add_tail(&dump->list, &vmcoredd_list);
1524	mutex_unlock(&vmcoredd_mutex);
1525
1526	vmcoredd_update_size(data_size);
1527	return 0;
1528
1529out_err:
1530	vfree(buf);
1531	vfree(dump);
1532
1533	return ret;
1534}
1535EXPORT_SYMBOL(vmcore_add_device_dump);
1536#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1537
1538/* Free all dumps in vmcore device dump list */
1539static void vmcore_free_device_dumps(void)
1540{
1541#ifdef CONFIG_PROC_VMCORE_DEVICE_DUMP
1542	mutex_lock(&vmcoredd_mutex);
1543	while (!list_empty(&vmcoredd_list)) {
1544		struct vmcoredd_node *dump;
1545
1546		dump = list_first_entry(&vmcoredd_list, struct vmcoredd_node,
1547					list);
1548		list_del(&dump->list);
1549		vfree(dump->buf);
1550		vfree(dump);
1551	}
1552	mutex_unlock(&vmcoredd_mutex);
1553#endif /* CONFIG_PROC_VMCORE_DEVICE_DUMP */
1554}
1555
1556/* Init function for vmcore module. */
1557static int __init vmcore_init(void)
1558{
1559	int rc = 0;
1560
1561	/* Allow architectures to allocate ELF header in 2nd kernel */
1562	rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1563	if (rc)
1564		return rc;
1565	/*
1566	 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1567	 * then capture the dump.
1568	 */
1569	if (!(is_vmcore_usable()))
1570		return rc;
1571	rc = parse_crash_elf_headers();
1572	if (rc) {
1573		elfcorehdr_free(elfcorehdr_addr);
1574		pr_warn("Kdump: vmcore not initialized\n");
1575		return rc;
1576	}
1577	elfcorehdr_free(elfcorehdr_addr);
1578	elfcorehdr_addr = ELFCORE_ADDR_ERR;
1579
1580	proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &vmcore_proc_ops);
1581	if (proc_vmcore)
1582		proc_vmcore->size = vmcore_size;
1583	return 0;
1584}
1585fs_initcall(vmcore_init);
1586
1587/* Cleanup function for vmcore module. */
1588void vmcore_cleanup(void)
1589{
 
 
1590	if (proc_vmcore) {
1591		proc_remove(proc_vmcore);
1592		proc_vmcore = NULL;
1593	}
1594
1595	/* clear the vmcore list. */
1596	while (!list_empty(&vmcore_list)) {
1597		struct vmcore *m;
1598
1599		m = list_first_entry(&vmcore_list, struct vmcore, list);
1600		list_del(&m->list);
1601		kfree(m);
1602	}
1603	free_elfcorebuf();
1604
1605	/* clear vmcore device dump list */
1606	vmcore_free_device_dumps();
1607}
v3.15
 
   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 <asm/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 = NULL;
  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 = min_t(size_t, m->offset + m->size - *fpos, buflen);
 
 
 235			start = m->paddr + *fpos - m->offset;
 236			tmp = read_from_oldmem(buffer, tsz, &start, userbuf);
 
 237			if (tmp < 0)
 238				return tmp;
 239			buflen -= tsz;
 240			*fpos += tsz;
 241			buffer += tsz;
 242			acc += tsz;
 243
 244			/* leave now if filled buffer already */
 245			if (buflen == 0)
 246				return acc;
 247		}
 
 
 248	}
 249
 250	return acc;
 251}
 252
 253static ssize_t read_vmcore(struct file *file, char __user *buffer,
 254			   size_t buflen, loff_t *fpos)
 255{
 256	return __read_vmcore((__force char *) buffer, buflen, fpos, 1);
 257}
 258
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 259/*
 260 * The vmcore fault handler uses the page cache and fills data using the
 261 * standard __vmcore_read() function.
 262 *
 263 * On s390 the fault handler is used for memory regions that can't be mapped
 264 * directly with remap_pfn_range().
 265 */
 266static int mmap_vmcore_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 267{
 268#ifdef CONFIG_S390
 269	struct address_space *mapping = vma->vm_file->f_mapping;
 270	pgoff_t index = vmf->pgoff;
 
 
 271	struct page *page;
 272	loff_t offset;
 273	char *buf;
 274	int rc;
 275
 276	page = find_or_create_page(mapping, index, GFP_KERNEL);
 277	if (!page)
 278		return VM_FAULT_OOM;
 279	if (!PageUptodate(page)) {
 280		offset = (loff_t) index << PAGE_CACHE_SHIFT;
 281		buf = __va((page_to_pfn(page) << PAGE_SHIFT));
 282		rc = __read_vmcore(buf, PAGE_SIZE, &offset, 0);
 
 
 
 283		if (rc < 0) {
 284			unlock_page(page);
 285			page_cache_release(page);
 286			return (rc == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
 287		}
 288		SetPageUptodate(page);
 289	}
 290	unlock_page(page);
 291	vmf->page = page;
 292	return 0;
 293#else
 294	return VM_FAULT_SIGBUS;
 295#endif
 296}
 297
 298static const struct vm_operations_struct vmcore_mmap_ops = {
 299	.fault = mmap_vmcore_fault,
 300};
 301
 302/**
 303 * alloc_elfnotes_buf - allocate buffer for ELF note segment in
 304 *                      vmalloc memory
 305 *
 306 * @notes_sz: size of buffer
 
 
 
 
 307 *
 308 * If CONFIG_MMU is defined, use vmalloc_user() to allow users to mmap
 309 * the buffer to user-space by means of remap_vmalloc_range().
 310 *
 311 * If CONFIG_MMU is not defined, use vzalloc() since mmap_vmcore() is
 312 * disabled and there's no need to allow users to mmap the buffer.
 313 */
 314static inline char *alloc_elfnotes_buf(size_t notes_sz)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 315{
 316#ifdef CONFIG_MMU
 317	return vmalloc_user(notes_sz);
 318#else
 319	return vzalloc(notes_sz);
 320#endif
 
 
 
 
 
 
 
 
 321}
 322
 323/*
 324 * Disable mmap_vmcore() if CONFIG_MMU is not defined. MMU is
 325 * essential for mmap_vmcore() in order to map physically
 326 * non-contiguous objects (ELF header, ELF note segment and memory
 327 * regions in the 1st kernel pointed to by PT_LOAD entries) into
 328 * virtually contiguous user-space in ELF layout.
 329 */
 330#ifdef CONFIG_MMU
 331static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 332{
 333	size_t size = vma->vm_end - vma->vm_start;
 334	u64 start, end, len, tsz;
 335	struct vmcore *m;
 336
 337	start = (u64)vma->vm_pgoff << PAGE_SHIFT;
 338	end = start + size;
 339
 340	if (size > vmcore_size || end > vmcore_size)
 341		return -EINVAL;
 342
 343	if (vma->vm_flags & (VM_WRITE | VM_EXEC))
 344		return -EPERM;
 345
 346	vma->vm_flags &= ~(VM_MAYWRITE | VM_MAYEXEC);
 347	vma->vm_flags |= VM_MIXEDMAP;
 348	vma->vm_ops = &vmcore_mmap_ops;
 349
 350	len = 0;
 351
 352	if (start < elfcorebuf_sz) {
 353		u64 pfn;
 354
 355		tsz = min(elfcorebuf_sz - (size_t)start, size);
 356		pfn = __pa(elfcorebuf + start) >> PAGE_SHIFT;
 357		if (remap_pfn_range(vma, vma->vm_start, pfn, tsz,
 358				    vma->vm_page_prot))
 359			return -EAGAIN;
 360		size -= tsz;
 361		start += tsz;
 362		len += tsz;
 363
 364		if (size == 0)
 365			return 0;
 366	}
 367
 368	if (start < elfcorebuf_sz + elfnotes_sz) {
 369		void *kaddr;
 370
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 371		tsz = min(elfcorebuf_sz + elfnotes_sz - (size_t)start, size);
 372		kaddr = elfnotes_buf + start - elfcorebuf_sz;
 373		if (remap_vmalloc_range_partial(vma, vma->vm_start + len,
 374						kaddr, tsz))
 375			goto fail;
 
 376		size -= tsz;
 377		start += tsz;
 378		len += tsz;
 379
 380		if (size == 0)
 381			return 0;
 382	}
 383
 384	list_for_each_entry(m, &vmcore_list, list) {
 385		if (start < m->offset + m->size) {
 386			u64 paddr = 0;
 387
 388			tsz = min_t(size_t, m->offset + m->size - start, size);
 
 389			paddr = m->paddr + start - m->offset;
 390			if (remap_oldmem_pfn_range(vma, vma->vm_start + len,
 391						   paddr >> PAGE_SHIFT, tsz,
 392						   vma->vm_page_prot))
 393				goto fail;
 394			size -= tsz;
 395			start += tsz;
 396			len += tsz;
 397
 398			if (size == 0)
 399				return 0;
 400		}
 401	}
 402
 403	return 0;
 404fail:
 405	do_munmap(vma->vm_mm, vma->vm_start, len);
 406	return -EAGAIN;
 407}
 408#else
 409static int mmap_vmcore(struct file *file, struct vm_area_struct *vma)
 410{
 411	return -ENOSYS;
 412}
 413#endif
 414
 415static const struct file_operations proc_vmcore_operations = {
 416	.read		= read_vmcore,
 417	.llseek		= default_llseek,
 418	.mmap		= mmap_vmcore,
 
 419};
 420
 421static struct vmcore* __init get_new_element(void)
 422{
 423	return kzalloc(sizeof(struct vmcore), GFP_KERNEL);
 424}
 425
 426static u64 __init get_vmcore_size(size_t elfsz, size_t elfnotesegsz,
 427				  struct list_head *vc_list)
 428{
 429	u64 size;
 430	struct vmcore *m;
 431
 432	size = elfsz + elfnotesegsz;
 433	list_for_each_entry(m, vc_list, list) {
 434		size += m->size;
 435	}
 436	return size;
 437}
 438
 439/**
 440 * update_note_header_size_elf64 - update p_memsz member of each PT_NOTE entry
 441 *
 442 * @ehdr_ptr: ELF header
 443 *
 444 * This function updates p_memsz member of each PT_NOTE entry in the
 445 * program header table pointed to by @ehdr_ptr to real size of ELF
 446 * note segment.
 447 */
 448static int __init update_note_header_size_elf64(const Elf64_Ehdr *ehdr_ptr)
 449{
 450	int i, rc=0;
 451	Elf64_Phdr *phdr_ptr;
 452	Elf64_Nhdr *nhdr_ptr;
 453
 454	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
 455	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 456		void *notes_section;
 457		u64 offset, max_sz, sz, real_sz = 0;
 458		if (phdr_ptr->p_type != PT_NOTE)
 459			continue;
 460		max_sz = phdr_ptr->p_memsz;
 461		offset = phdr_ptr->p_offset;
 462		notes_section = kmalloc(max_sz, GFP_KERNEL);
 463		if (!notes_section)
 464			return -ENOMEM;
 465		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
 466		if (rc < 0) {
 467			kfree(notes_section);
 468			return rc;
 469		}
 470		nhdr_ptr = notes_section;
 471		while (nhdr_ptr->n_namesz != 0) {
 472			sz = sizeof(Elf64_Nhdr) +
 473				((nhdr_ptr->n_namesz + 3) & ~3) +
 474				((nhdr_ptr->n_descsz + 3) & ~3);
 475			if ((real_sz + sz) > max_sz) {
 476				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
 477					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
 478				break;
 479			}
 480			real_sz += sz;
 481			nhdr_ptr = (Elf64_Nhdr*)((char*)nhdr_ptr + sz);
 482		}
 483		kfree(notes_section);
 484		phdr_ptr->p_memsz = real_sz;
 485		if (real_sz == 0) {
 486			pr_warn("Warning: Zero PT_NOTE entries found\n");
 487		}
 488	}
 489
 490	return 0;
 491}
 492
 493/**
 494 * get_note_number_and_size_elf64 - get the number of PT_NOTE program
 495 * headers and sum of real size of their ELF note segment headers and
 496 * data.
 497 *
 498 * @ehdr_ptr: ELF header
 499 * @nr_ptnote: buffer for the number of PT_NOTE program headers
 500 * @sz_ptnote: buffer for size of unique PT_NOTE program header
 501 *
 502 * This function is used to merge multiple PT_NOTE program headers
 503 * into a unique single one. The resulting unique entry will have
 504 * @sz_ptnote in its phdr->p_mem.
 505 *
 506 * It is assumed that program headers with PT_NOTE type pointed to by
 507 * @ehdr_ptr has already been updated by update_note_header_size_elf64
 508 * and each of PT_NOTE program headers has actual ELF note segment
 509 * size in its p_memsz member.
 510 */
 511static int __init get_note_number_and_size_elf64(const Elf64_Ehdr *ehdr_ptr,
 512						 int *nr_ptnote, u64 *sz_ptnote)
 513{
 514	int i;
 515	Elf64_Phdr *phdr_ptr;
 516
 517	*nr_ptnote = *sz_ptnote = 0;
 518
 519	phdr_ptr = (Elf64_Phdr *)(ehdr_ptr + 1);
 520	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 521		if (phdr_ptr->p_type != PT_NOTE)
 522			continue;
 523		*nr_ptnote += 1;
 524		*sz_ptnote += phdr_ptr->p_memsz;
 525	}
 526
 527	return 0;
 528}
 529
 530/**
 531 * copy_notes_elf64 - copy ELF note segments in a given buffer
 532 *
 533 * @ehdr_ptr: ELF header
 534 * @notes_buf: buffer into which ELF note segments are copied
 535 *
 536 * This function is used to copy ELF note segment in the 1st kernel
 537 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
 538 * size of the buffer @notes_buf is equal to or larger than sum of the
 539 * real ELF note segment headers and data.
 540 *
 541 * It is assumed that program headers with PT_NOTE type pointed to by
 542 * @ehdr_ptr has already been updated by update_note_header_size_elf64
 543 * and each of PT_NOTE program headers has actual ELF note segment
 544 * size in its p_memsz member.
 545 */
 546static int __init copy_notes_elf64(const Elf64_Ehdr *ehdr_ptr, char *notes_buf)
 547{
 548	int i, rc=0;
 549	Elf64_Phdr *phdr_ptr;
 550
 551	phdr_ptr = (Elf64_Phdr*)(ehdr_ptr + 1);
 552
 553	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 554		u64 offset;
 555		if (phdr_ptr->p_type != PT_NOTE)
 556			continue;
 557		offset = phdr_ptr->p_offset;
 558		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
 559					   &offset);
 560		if (rc < 0)
 561			return rc;
 562		notes_buf += phdr_ptr->p_memsz;
 563	}
 564
 565	return 0;
 566}
 567
 568/* Merges all the PT_NOTE headers into one. */
 569static int __init merge_note_headers_elf64(char *elfptr, size_t *elfsz,
 570					   char **notes_buf, size_t *notes_sz)
 571{
 572	int i, nr_ptnote=0, rc=0;
 573	char *tmp;
 574	Elf64_Ehdr *ehdr_ptr;
 575	Elf64_Phdr phdr;
 576	u64 phdr_sz = 0, note_off;
 577
 578	ehdr_ptr = (Elf64_Ehdr *)elfptr;
 579
 580	rc = update_note_header_size_elf64(ehdr_ptr);
 581	if (rc < 0)
 582		return rc;
 583
 584	rc = get_note_number_and_size_elf64(ehdr_ptr, &nr_ptnote, &phdr_sz);
 585	if (rc < 0)
 586		return rc;
 587
 588	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
 589	*notes_buf = alloc_elfnotes_buf(*notes_sz);
 590	if (!*notes_buf)
 591		return -ENOMEM;
 592
 593	rc = copy_notes_elf64(ehdr_ptr, *notes_buf);
 594	if (rc < 0)
 595		return rc;
 596
 597	/* Prepare merged PT_NOTE program header. */
 598	phdr.p_type    = PT_NOTE;
 599	phdr.p_flags   = 0;
 600	note_off = sizeof(Elf64_Ehdr) +
 601			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf64_Phdr);
 602	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
 603	phdr.p_vaddr   = phdr.p_paddr = 0;
 604	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
 605	phdr.p_align   = 0;
 606
 607	/* Add merged PT_NOTE program header*/
 608	tmp = elfptr + sizeof(Elf64_Ehdr);
 609	memcpy(tmp, &phdr, sizeof(phdr));
 610	tmp += sizeof(phdr);
 611
 612	/* Remove unwanted PT_NOTE program headers. */
 613	i = (nr_ptnote - 1) * sizeof(Elf64_Phdr);
 614	*elfsz = *elfsz - i;
 615	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf64_Ehdr)-sizeof(Elf64_Phdr)));
 616	memset(elfptr + *elfsz, 0, i);
 617	*elfsz = roundup(*elfsz, PAGE_SIZE);
 618
 619	/* Modify e_phnum to reflect merged headers. */
 620	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
 621
 
 
 
 
 
 622	return 0;
 623}
 624
 625/**
 626 * update_note_header_size_elf32 - update p_memsz member of each PT_NOTE entry
 627 *
 628 * @ehdr_ptr: ELF header
 629 *
 630 * This function updates p_memsz member of each PT_NOTE entry in the
 631 * program header table pointed to by @ehdr_ptr to real size of ELF
 632 * note segment.
 633 */
 634static int __init update_note_header_size_elf32(const Elf32_Ehdr *ehdr_ptr)
 635{
 636	int i, rc=0;
 637	Elf32_Phdr *phdr_ptr;
 638	Elf32_Nhdr *nhdr_ptr;
 639
 640	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
 641	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 642		void *notes_section;
 643		u64 offset, max_sz, sz, real_sz = 0;
 644		if (phdr_ptr->p_type != PT_NOTE)
 645			continue;
 646		max_sz = phdr_ptr->p_memsz;
 647		offset = phdr_ptr->p_offset;
 648		notes_section = kmalloc(max_sz, GFP_KERNEL);
 649		if (!notes_section)
 650			return -ENOMEM;
 651		rc = elfcorehdr_read_notes(notes_section, max_sz, &offset);
 652		if (rc < 0) {
 653			kfree(notes_section);
 654			return rc;
 655		}
 656		nhdr_ptr = notes_section;
 657		while (nhdr_ptr->n_namesz != 0) {
 658			sz = sizeof(Elf32_Nhdr) +
 659				((nhdr_ptr->n_namesz + 3) & ~3) +
 660				((nhdr_ptr->n_descsz + 3) & ~3);
 661			if ((real_sz + sz) > max_sz) {
 662				pr_warn("Warning: Exceeded p_memsz, dropping PT_NOTE entry n_namesz=0x%x, n_descsz=0x%x\n",
 663					nhdr_ptr->n_namesz, nhdr_ptr->n_descsz);
 664				break;
 665			}
 666			real_sz += sz;
 667			nhdr_ptr = (Elf32_Nhdr*)((char*)nhdr_ptr + sz);
 668		}
 669		kfree(notes_section);
 670		phdr_ptr->p_memsz = real_sz;
 671		if (real_sz == 0) {
 672			pr_warn("Warning: Zero PT_NOTE entries found\n");
 673		}
 674	}
 675
 676	return 0;
 677}
 678
 679/**
 680 * get_note_number_and_size_elf32 - get the number of PT_NOTE program
 681 * headers and sum of real size of their ELF note segment headers and
 682 * data.
 683 *
 684 * @ehdr_ptr: ELF header
 685 * @nr_ptnote: buffer for the number of PT_NOTE program headers
 686 * @sz_ptnote: buffer for size of unique PT_NOTE program header
 687 *
 688 * This function is used to merge multiple PT_NOTE program headers
 689 * into a unique single one. The resulting unique entry will have
 690 * @sz_ptnote in its phdr->p_mem.
 691 *
 692 * It is assumed that program headers with PT_NOTE type pointed to by
 693 * @ehdr_ptr has already been updated by update_note_header_size_elf32
 694 * and each of PT_NOTE program headers has actual ELF note segment
 695 * size in its p_memsz member.
 696 */
 697static int __init get_note_number_and_size_elf32(const Elf32_Ehdr *ehdr_ptr,
 698						 int *nr_ptnote, u64 *sz_ptnote)
 699{
 700	int i;
 701	Elf32_Phdr *phdr_ptr;
 702
 703	*nr_ptnote = *sz_ptnote = 0;
 704
 705	phdr_ptr = (Elf32_Phdr *)(ehdr_ptr + 1);
 706	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 707		if (phdr_ptr->p_type != PT_NOTE)
 708			continue;
 709		*nr_ptnote += 1;
 710		*sz_ptnote += phdr_ptr->p_memsz;
 711	}
 712
 713	return 0;
 714}
 715
 716/**
 717 * copy_notes_elf32 - copy ELF note segments in a given buffer
 718 *
 719 * @ehdr_ptr: ELF header
 720 * @notes_buf: buffer into which ELF note segments are copied
 721 *
 722 * This function is used to copy ELF note segment in the 1st kernel
 723 * into the buffer @notes_buf in the 2nd kernel. It is assumed that
 724 * size of the buffer @notes_buf is equal to or larger than sum of the
 725 * real ELF note segment headers and data.
 726 *
 727 * It is assumed that program headers with PT_NOTE type pointed to by
 728 * @ehdr_ptr has already been updated by update_note_header_size_elf32
 729 * and each of PT_NOTE program headers has actual ELF note segment
 730 * size in its p_memsz member.
 731 */
 732static int __init copy_notes_elf32(const Elf32_Ehdr *ehdr_ptr, char *notes_buf)
 733{
 734	int i, rc=0;
 735	Elf32_Phdr *phdr_ptr;
 736
 737	phdr_ptr = (Elf32_Phdr*)(ehdr_ptr + 1);
 738
 739	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 740		u64 offset;
 741		if (phdr_ptr->p_type != PT_NOTE)
 742			continue;
 743		offset = phdr_ptr->p_offset;
 744		rc = elfcorehdr_read_notes(notes_buf, phdr_ptr->p_memsz,
 745					   &offset);
 746		if (rc < 0)
 747			return rc;
 748		notes_buf += phdr_ptr->p_memsz;
 749	}
 750
 751	return 0;
 752}
 753
 754/* Merges all the PT_NOTE headers into one. */
 755static int __init merge_note_headers_elf32(char *elfptr, size_t *elfsz,
 756					   char **notes_buf, size_t *notes_sz)
 757{
 758	int i, nr_ptnote=0, rc=0;
 759	char *tmp;
 760	Elf32_Ehdr *ehdr_ptr;
 761	Elf32_Phdr phdr;
 762	u64 phdr_sz = 0, note_off;
 763
 764	ehdr_ptr = (Elf32_Ehdr *)elfptr;
 765
 766	rc = update_note_header_size_elf32(ehdr_ptr);
 767	if (rc < 0)
 768		return rc;
 769
 770	rc = get_note_number_and_size_elf32(ehdr_ptr, &nr_ptnote, &phdr_sz);
 771	if (rc < 0)
 772		return rc;
 773
 774	*notes_sz = roundup(phdr_sz, PAGE_SIZE);
 775	*notes_buf = alloc_elfnotes_buf(*notes_sz);
 776	if (!*notes_buf)
 777		return -ENOMEM;
 778
 779	rc = copy_notes_elf32(ehdr_ptr, *notes_buf);
 780	if (rc < 0)
 781		return rc;
 782
 783	/* Prepare merged PT_NOTE program header. */
 784	phdr.p_type    = PT_NOTE;
 785	phdr.p_flags   = 0;
 786	note_off = sizeof(Elf32_Ehdr) +
 787			(ehdr_ptr->e_phnum - nr_ptnote +1) * sizeof(Elf32_Phdr);
 788	phdr.p_offset  = roundup(note_off, PAGE_SIZE);
 789	phdr.p_vaddr   = phdr.p_paddr = 0;
 790	phdr.p_filesz  = phdr.p_memsz = phdr_sz;
 791	phdr.p_align   = 0;
 792
 793	/* Add merged PT_NOTE program header*/
 794	tmp = elfptr + sizeof(Elf32_Ehdr);
 795	memcpy(tmp, &phdr, sizeof(phdr));
 796	tmp += sizeof(phdr);
 797
 798	/* Remove unwanted PT_NOTE program headers. */
 799	i = (nr_ptnote - 1) * sizeof(Elf32_Phdr);
 800	*elfsz = *elfsz - i;
 801	memmove(tmp, tmp+i, ((*elfsz)-sizeof(Elf32_Ehdr)-sizeof(Elf32_Phdr)));
 802	memset(elfptr + *elfsz, 0, i);
 803	*elfsz = roundup(*elfsz, PAGE_SIZE);
 804
 805	/* Modify e_phnum to reflect merged headers. */
 806	ehdr_ptr->e_phnum = ehdr_ptr->e_phnum - nr_ptnote + 1;
 807
 
 
 
 
 
 808	return 0;
 809}
 810
 811/* Add memory chunks represented by program headers to vmcore list. Also update
 812 * the new offset fields of exported program headers. */
 813static int __init process_ptload_program_headers_elf64(char *elfptr,
 814						size_t elfsz,
 815						size_t elfnotes_sz,
 816						struct list_head *vc_list)
 817{
 818	int i;
 819	Elf64_Ehdr *ehdr_ptr;
 820	Elf64_Phdr *phdr_ptr;
 821	loff_t vmcore_off;
 822	struct vmcore *new;
 823
 824	ehdr_ptr = (Elf64_Ehdr *)elfptr;
 825	phdr_ptr = (Elf64_Phdr*)(elfptr + sizeof(Elf64_Ehdr)); /* PT_NOTE hdr */
 826
 827	/* Skip Elf header, program headers and Elf note segment. */
 828	vmcore_off = elfsz + elfnotes_sz;
 829
 830	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 831		u64 paddr, start, end, size;
 832
 833		if (phdr_ptr->p_type != PT_LOAD)
 834			continue;
 835
 836		paddr = phdr_ptr->p_offset;
 837		start = rounddown(paddr, PAGE_SIZE);
 838		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
 839		size = end - start;
 840
 841		/* Add this contiguous chunk of memory to vmcore list.*/
 842		new = get_new_element();
 843		if (!new)
 844			return -ENOMEM;
 845		new->paddr = start;
 846		new->size = size;
 847		list_add_tail(&new->list, vc_list);
 848
 849		/* Update the program header offset. */
 850		phdr_ptr->p_offset = vmcore_off + (paddr - start);
 851		vmcore_off = vmcore_off + size;
 852	}
 853	return 0;
 854}
 855
 856static int __init process_ptload_program_headers_elf32(char *elfptr,
 857						size_t elfsz,
 858						size_t elfnotes_sz,
 859						struct list_head *vc_list)
 860{
 861	int i;
 862	Elf32_Ehdr *ehdr_ptr;
 863	Elf32_Phdr *phdr_ptr;
 864	loff_t vmcore_off;
 865	struct vmcore *new;
 866
 867	ehdr_ptr = (Elf32_Ehdr *)elfptr;
 868	phdr_ptr = (Elf32_Phdr*)(elfptr + sizeof(Elf32_Ehdr)); /* PT_NOTE hdr */
 869
 870	/* Skip Elf header, program headers and Elf note segment. */
 871	vmcore_off = elfsz + elfnotes_sz;
 872
 873	for (i = 0; i < ehdr_ptr->e_phnum; i++, phdr_ptr++) {
 874		u64 paddr, start, end, size;
 875
 876		if (phdr_ptr->p_type != PT_LOAD)
 877			continue;
 878
 879		paddr = phdr_ptr->p_offset;
 880		start = rounddown(paddr, PAGE_SIZE);
 881		end = roundup(paddr + phdr_ptr->p_memsz, PAGE_SIZE);
 882		size = end - start;
 883
 884		/* Add this contiguous chunk of memory to vmcore list.*/
 885		new = get_new_element();
 886		if (!new)
 887			return -ENOMEM;
 888		new->paddr = start;
 889		new->size = size;
 890		list_add_tail(&new->list, vc_list);
 891
 892		/* Update the program header offset */
 893		phdr_ptr->p_offset = vmcore_off + (paddr - start);
 894		vmcore_off = vmcore_off + size;
 895	}
 896	return 0;
 897}
 898
 899/* Sets offset fields of vmcore elements. */
 900static void __init set_vmcore_list_offsets(size_t elfsz, size_t elfnotes_sz,
 901					   struct list_head *vc_list)
 902{
 903	loff_t vmcore_off;
 904	struct vmcore *m;
 905
 906	/* Skip Elf header, program headers and Elf note segment. */
 907	vmcore_off = elfsz + elfnotes_sz;
 908
 909	list_for_each_entry(m, vc_list, list) {
 910		m->offset = vmcore_off;
 911		vmcore_off += m->size;
 912	}
 913}
 914
 915static void free_elfcorebuf(void)
 916{
 917	free_pages((unsigned long)elfcorebuf, get_order(elfcorebuf_sz_orig));
 918	elfcorebuf = NULL;
 919	vfree(elfnotes_buf);
 920	elfnotes_buf = NULL;
 921}
 922
 923static int __init parse_crash_elf64_headers(void)
 924{
 925	int rc=0;
 926	Elf64_Ehdr ehdr;
 927	u64 addr;
 928
 929	addr = elfcorehdr_addr;
 930
 931	/* Read Elf header */
 932	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf64_Ehdr), &addr);
 933	if (rc < 0)
 934		return rc;
 935
 936	/* Do some basic Verification. */
 937	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
 938		(ehdr.e_type != ET_CORE) ||
 939		!vmcore_elf64_check_arch(&ehdr) ||
 940		ehdr.e_ident[EI_CLASS] != ELFCLASS64 ||
 941		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
 942		ehdr.e_version != EV_CURRENT ||
 943		ehdr.e_ehsize != sizeof(Elf64_Ehdr) ||
 944		ehdr.e_phentsize != sizeof(Elf64_Phdr) ||
 945		ehdr.e_phnum == 0) {
 946		pr_warn("Warning: Core image elf header is not sane\n");
 947		return -EINVAL;
 948	}
 949
 950	/* Read in all elf headers. */
 951	elfcorebuf_sz_orig = sizeof(Elf64_Ehdr) +
 952				ehdr.e_phnum * sizeof(Elf64_Phdr);
 953	elfcorebuf_sz = elfcorebuf_sz_orig;
 954	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 955					      get_order(elfcorebuf_sz_orig));
 956	if (!elfcorebuf)
 957		return -ENOMEM;
 958	addr = elfcorehdr_addr;
 959	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
 960	if (rc < 0)
 961		goto fail;
 962
 963	/* Merge all PT_NOTE headers into one. */
 964	rc = merge_note_headers_elf64(elfcorebuf, &elfcorebuf_sz,
 965				      &elfnotes_buf, &elfnotes_sz);
 966	if (rc)
 967		goto fail;
 968	rc = process_ptload_program_headers_elf64(elfcorebuf, elfcorebuf_sz,
 969						  elfnotes_sz, &vmcore_list);
 970	if (rc)
 971		goto fail;
 972	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
 973	return 0;
 974fail:
 975	free_elfcorebuf();
 976	return rc;
 977}
 978
 979static int __init parse_crash_elf32_headers(void)
 980{
 981	int rc=0;
 982	Elf32_Ehdr ehdr;
 983	u64 addr;
 984
 985	addr = elfcorehdr_addr;
 986
 987	/* Read Elf header */
 988	rc = elfcorehdr_read((char *)&ehdr, sizeof(Elf32_Ehdr), &addr);
 989	if (rc < 0)
 990		return rc;
 991
 992	/* Do some basic Verification. */
 993	if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
 994		(ehdr.e_type != ET_CORE) ||
 995		!elf_check_arch(&ehdr) ||
 996		ehdr.e_ident[EI_CLASS] != ELFCLASS32||
 997		ehdr.e_ident[EI_VERSION] != EV_CURRENT ||
 998		ehdr.e_version != EV_CURRENT ||
 999		ehdr.e_ehsize != sizeof(Elf32_Ehdr) ||
1000		ehdr.e_phentsize != sizeof(Elf32_Phdr) ||
1001		ehdr.e_phnum == 0) {
1002		pr_warn("Warning: Core image elf header is not sane\n");
1003		return -EINVAL;
1004	}
1005
1006	/* Read in all elf headers. */
1007	elfcorebuf_sz_orig = sizeof(Elf32_Ehdr) + ehdr.e_phnum * sizeof(Elf32_Phdr);
1008	elfcorebuf_sz = elfcorebuf_sz_orig;
1009	elfcorebuf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1010					      get_order(elfcorebuf_sz_orig));
1011	if (!elfcorebuf)
1012		return -ENOMEM;
1013	addr = elfcorehdr_addr;
1014	rc = elfcorehdr_read(elfcorebuf, elfcorebuf_sz_orig, &addr);
1015	if (rc < 0)
1016		goto fail;
1017
1018	/* Merge all PT_NOTE headers into one. */
1019	rc = merge_note_headers_elf32(elfcorebuf, &elfcorebuf_sz,
1020				      &elfnotes_buf, &elfnotes_sz);
1021	if (rc)
1022		goto fail;
1023	rc = process_ptload_program_headers_elf32(elfcorebuf, elfcorebuf_sz,
1024						  elfnotes_sz, &vmcore_list);
1025	if (rc)
1026		goto fail;
1027	set_vmcore_list_offsets(elfcorebuf_sz, elfnotes_sz, &vmcore_list);
1028	return 0;
1029fail:
1030	free_elfcorebuf();
1031	return rc;
1032}
1033
1034static int __init parse_crash_elf_headers(void)
1035{
1036	unsigned char e_ident[EI_NIDENT];
1037	u64 addr;
1038	int rc=0;
1039
1040	addr = elfcorehdr_addr;
1041	rc = elfcorehdr_read(e_ident, EI_NIDENT, &addr);
1042	if (rc < 0)
1043		return rc;
1044	if (memcmp(e_ident, ELFMAG, SELFMAG) != 0) {
1045		pr_warn("Warning: Core image elf header not found\n");
1046		return -EINVAL;
1047	}
1048
1049	if (e_ident[EI_CLASS] == ELFCLASS64) {
1050		rc = parse_crash_elf64_headers();
1051		if (rc)
1052			return rc;
1053	} else if (e_ident[EI_CLASS] == ELFCLASS32) {
1054		rc = parse_crash_elf32_headers();
1055		if (rc)
1056			return rc;
1057	} else {
1058		pr_warn("Warning: Core image elf header is not sane\n");
1059		return -EINVAL;
1060	}
1061
1062	/* Determine vmcore size. */
1063	vmcore_size = get_vmcore_size(elfcorebuf_sz, elfnotes_sz,
1064				      &vmcore_list);
1065
1066	return 0;
1067}
1068
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1069/* Init function for vmcore module. */
1070static int __init vmcore_init(void)
1071{
1072	int rc = 0;
1073
1074	/* Allow architectures to allocate ELF header in 2nd kernel */
1075	rc = elfcorehdr_alloc(&elfcorehdr_addr, &elfcorehdr_size);
1076	if (rc)
1077		return rc;
1078	/*
1079	 * If elfcorehdr= has been passed in cmdline or created in 2nd kernel,
1080	 * then capture the dump.
1081	 */
1082	if (!(is_vmcore_usable()))
1083		return rc;
1084	rc = parse_crash_elf_headers();
1085	if (rc) {
 
1086		pr_warn("Kdump: vmcore not initialized\n");
1087		return rc;
1088	}
1089	elfcorehdr_free(elfcorehdr_addr);
1090	elfcorehdr_addr = ELFCORE_ADDR_ERR;
1091
1092	proc_vmcore = proc_create("vmcore", S_IRUSR, NULL, &proc_vmcore_operations);
1093	if (proc_vmcore)
1094		proc_vmcore->size = vmcore_size;
1095	return 0;
1096}
1097fs_initcall(vmcore_init);
1098
1099/* Cleanup function for vmcore module. */
1100void vmcore_cleanup(void)
1101{
1102	struct list_head *pos, *next;
1103
1104	if (proc_vmcore) {
1105		proc_remove(proc_vmcore);
1106		proc_vmcore = NULL;
1107	}
1108
1109	/* clear the vmcore list. */
1110	list_for_each_safe(pos, next, &vmcore_list) {
1111		struct vmcore *m;
1112
1113		m = list_entry(pos, struct vmcore, list);
1114		list_del(&m->list);
1115		kfree(m);
1116	}
1117	free_elfcorebuf();
 
 
 
1118}