Linux Audio

Check our new training course

Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * kexec.c - kexec system call core code.
   4 * Copyright (C) 2002-2004 Eric Biederman  <ebiederm@xmission.com>
   5 */
   6
   7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
 
   9#include <linux/capability.h>
  10#include <linux/mm.h>
  11#include <linux/file.h>
  12#include <linux/slab.h>
  13#include <linux/fs.h>
  14#include <linux/kexec.h>
  15#include <linux/mutex.h>
  16#include <linux/list.h>
  17#include <linux/highmem.h>
  18#include <linux/syscalls.h>
  19#include <linux/reboot.h>
  20#include <linux/ioport.h>
  21#include <linux/hardirq.h>
  22#include <linux/elf.h>
  23#include <linux/elfcore.h>
  24#include <linux/utsname.h>
  25#include <linux/numa.h>
  26#include <linux/suspend.h>
  27#include <linux/device.h>
  28#include <linux/freezer.h>
  29#include <linux/panic_notifier.h>
  30#include <linux/pm.h>
  31#include <linux/cpu.h>
  32#include <linux/uaccess.h>
  33#include <linux/io.h>
  34#include <linux/console.h>
  35#include <linux/vmalloc.h>
  36#include <linux/swap.h>
  37#include <linux/syscore_ops.h>
  38#include <linux/compiler.h>
  39#include <linux/hugetlb.h>
  40#include <linux/objtool.h>
  41#include <linux/kmsg_dump.h>
  42
  43#include <asm/page.h>
  44#include <asm/sections.h>
  45
  46#include <crypto/hash.h>
  47#include "kexec_internal.h"
  48
  49atomic_t __kexec_lock = ATOMIC_INIT(0);
  50
  51/* Per cpu memory for storing cpu states in case of system crash. */
  52note_buf_t __percpu *crash_notes;
  53
  54/* Flag to indicate we are going to kexec a new kernel */
  55bool kexec_in_progress = false;
  56
  57
  58/* Location of the reserved area for the crash kernel */
  59struct resource crashk_res = {
  60	.name  = "Crash kernel",
  61	.start = 0,
  62	.end   = 0,
  63	.flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
  64	.desc  = IORES_DESC_CRASH_KERNEL
  65};
  66struct resource crashk_low_res = {
  67	.name  = "Crash kernel",
  68	.start = 0,
  69	.end   = 0,
  70	.flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
  71	.desc  = IORES_DESC_CRASH_KERNEL
  72};
  73
  74int kexec_should_crash(struct task_struct *p)
  75{
  76	/*
  77	 * If crash_kexec_post_notifiers is enabled, don't run
  78	 * crash_kexec() here yet, which must be run after panic
  79	 * notifiers in panic().
  80	 */
  81	if (crash_kexec_post_notifiers)
  82		return 0;
  83	/*
  84	 * There are 4 panic() calls in make_task_dead() path, each of which
  85	 * corresponds to each of these 4 conditions.
  86	 */
  87	if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
  88		return 1;
  89	return 0;
  90}
  91
  92int kexec_crash_loaded(void)
  93{
  94	return !!kexec_crash_image;
  95}
  96EXPORT_SYMBOL_GPL(kexec_crash_loaded);
  97
  98/*
  99 * When kexec transitions to the new kernel there is a one-to-one
 100 * mapping between physical and virtual addresses.  On processors
 101 * where you can disable the MMU this is trivial, and easy.  For
 102 * others it is still a simple predictable page table to setup.
 103 *
 104 * In that environment kexec copies the new kernel to its final
 105 * resting place.  This means I can only support memory whose
 106 * physical address can fit in an unsigned long.  In particular
 107 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
 108 * If the assembly stub has more restrictive requirements
 109 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
 110 * defined more restrictively in <asm/kexec.h>.
 111 *
 112 * The code for the transition from the current kernel to the
 113 * new kernel is placed in the control_code_buffer, whose size
 114 * is given by KEXEC_CONTROL_PAGE_SIZE.  In the best case only a single
 115 * page of memory is necessary, but some architectures require more.
 116 * Because this memory must be identity mapped in the transition from
 117 * virtual to physical addresses it must live in the range
 118 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
 119 * modifiable.
 120 *
 121 * The assembly stub in the control code buffer is passed a linked list
 122 * of descriptor pages detailing the source pages of the new kernel,
 123 * and the destination addresses of those source pages.  As this data
 124 * structure is not used in the context of the current OS, it must
 125 * be self-contained.
 126 *
 127 * The code has been made to work with highmem pages and will use a
 128 * destination page in its final resting place (if it happens
 129 * to allocate it).  The end product of this is that most of the
 130 * physical address space, and most of RAM can be used.
 131 *
 132 * Future directions include:
 133 *  - allocating a page table with the control code buffer identity
 134 *    mapped, to simplify machine_kexec and make kexec_on_panic more
 135 *    reliable.
 136 */
 137
 138/*
 139 * KIMAGE_NO_DEST is an impossible destination address..., for
 140 * allocating pages whose destination address we do not care about.
 141 */
 142#define KIMAGE_NO_DEST (-1UL)
 143#define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
 144
 145static struct page *kimage_alloc_page(struct kimage *image,
 146				       gfp_t gfp_mask,
 147				       unsigned long dest);
 148
 149int sanity_check_segment_list(struct kimage *image)
 150{
 151	int i;
 152	unsigned long nr_segments = image->nr_segments;
 153	unsigned long total_pages = 0;
 154	unsigned long nr_pages = totalram_pages();
 155
 156	/*
 157	 * Verify we have good destination addresses.  The caller is
 158	 * responsible for making certain we don't attempt to load
 159	 * the new image into invalid or reserved areas of RAM.  This
 160	 * just verifies it is an address we can use.
 161	 *
 162	 * Since the kernel does everything in page size chunks ensure
 163	 * the destination addresses are page aligned.  Too many
 164	 * special cases crop of when we don't do this.  The most
 165	 * insidious is getting overlapping destination addresses
 166	 * simply because addresses are changed to page size
 167	 * granularity.
 168	 */
 169	for (i = 0; i < nr_segments; i++) {
 170		unsigned long mstart, mend;
 171
 172		mstart = image->segment[i].mem;
 173		mend   = mstart + image->segment[i].memsz;
 174		if (mstart > mend)
 175			return -EADDRNOTAVAIL;
 176		if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
 177			return -EADDRNOTAVAIL;
 178		if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
 179			return -EADDRNOTAVAIL;
 180	}
 181
 182	/* Verify our destination addresses do not overlap.
 183	 * If we alloed overlapping destination addresses
 184	 * through very weird things can happen with no
 185	 * easy explanation as one segment stops on another.
 186	 */
 187	for (i = 0; i < nr_segments; i++) {
 188		unsigned long mstart, mend;
 189		unsigned long j;
 190
 191		mstart = image->segment[i].mem;
 192		mend   = mstart + image->segment[i].memsz;
 193		for (j = 0; j < i; j++) {
 194			unsigned long pstart, pend;
 195
 196			pstart = image->segment[j].mem;
 197			pend   = pstart + image->segment[j].memsz;
 198			/* Do the segments overlap ? */
 199			if ((mend > pstart) && (mstart < pend))
 200				return -EINVAL;
 201		}
 202	}
 203
 204	/* Ensure our buffer sizes are strictly less than
 205	 * our memory sizes.  This should always be the case,
 206	 * and it is easier to check up front than to be surprised
 207	 * later on.
 208	 */
 209	for (i = 0; i < nr_segments; i++) {
 210		if (image->segment[i].bufsz > image->segment[i].memsz)
 211			return -EINVAL;
 212	}
 213
 214	/*
 215	 * Verify that no more than half of memory will be consumed. If the
 216	 * request from userspace is too large, a large amount of time will be
 217	 * wasted allocating pages, which can cause a soft lockup.
 218	 */
 219	for (i = 0; i < nr_segments; i++) {
 220		if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2)
 221			return -EINVAL;
 222
 223		total_pages += PAGE_COUNT(image->segment[i].memsz);
 224	}
 225
 226	if (total_pages > nr_pages / 2)
 227		return -EINVAL;
 228
 229	/*
 230	 * Verify we have good destination addresses.  Normally
 231	 * the caller is responsible for making certain we don't
 232	 * attempt to load the new image into invalid or reserved
 233	 * areas of RAM.  But crash kernels are preloaded into a
 234	 * reserved area of ram.  We must ensure the addresses
 235	 * are in the reserved area otherwise preloading the
 236	 * kernel could corrupt things.
 237	 */
 238
 239	if (image->type == KEXEC_TYPE_CRASH) {
 240		for (i = 0; i < nr_segments; i++) {
 241			unsigned long mstart, mend;
 242
 243			mstart = image->segment[i].mem;
 244			mend = mstart + image->segment[i].memsz - 1;
 245			/* Ensure we are within the crash kernel limits */
 246			if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
 247			    (mend > phys_to_boot_phys(crashk_res.end)))
 248				return -EADDRNOTAVAIL;
 249		}
 250	}
 251
 252	return 0;
 253}
 254
 255struct kimage *do_kimage_alloc_init(void)
 256{
 257	struct kimage *image;
 258
 259	/* Allocate a controlling structure */
 260	image = kzalloc(sizeof(*image), GFP_KERNEL);
 261	if (!image)
 262		return NULL;
 263
 264	image->head = 0;
 265	image->entry = &image->head;
 266	image->last_entry = &image->head;
 267	image->control_page = ~0; /* By default this does not apply */
 268	image->type = KEXEC_TYPE_DEFAULT;
 269
 270	/* Initialize the list of control pages */
 271	INIT_LIST_HEAD(&image->control_pages);
 272
 273	/* Initialize the list of destination pages */
 274	INIT_LIST_HEAD(&image->dest_pages);
 275
 276	/* Initialize the list of unusable pages */
 277	INIT_LIST_HEAD(&image->unusable_pages);
 278
 
 
 
 
 
 
 279	return image;
 280}
 281
 282int kimage_is_destination_range(struct kimage *image,
 283					unsigned long start,
 284					unsigned long end)
 285{
 286	unsigned long i;
 287
 288	for (i = 0; i < image->nr_segments; i++) {
 289		unsigned long mstart, mend;
 290
 291		mstart = image->segment[i].mem;
 292		mend = mstart + image->segment[i].memsz;
 293		if ((end > mstart) && (start < mend))
 294			return 1;
 295	}
 296
 297	return 0;
 298}
 299
 300static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
 301{
 302	struct page *pages;
 303
 304	if (fatal_signal_pending(current))
 305		return NULL;
 306	pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order);
 307	if (pages) {
 308		unsigned int count, i;
 309
 310		pages->mapping = NULL;
 311		set_page_private(pages, order);
 312		count = 1 << order;
 313		for (i = 0; i < count; i++)
 314			SetPageReserved(pages + i);
 315
 316		arch_kexec_post_alloc_pages(page_address(pages), count,
 317					    gfp_mask);
 318
 319		if (gfp_mask & __GFP_ZERO)
 320			for (i = 0; i < count; i++)
 321				clear_highpage(pages + i);
 322	}
 323
 324	return pages;
 325}
 326
 327static void kimage_free_pages(struct page *page)
 328{
 329	unsigned int order, count, i;
 330
 331	order = page_private(page);
 332	count = 1 << order;
 333
 334	arch_kexec_pre_free_pages(page_address(page), count);
 335
 336	for (i = 0; i < count; i++)
 337		ClearPageReserved(page + i);
 338	__free_pages(page, order);
 339}
 340
 341void kimage_free_page_list(struct list_head *list)
 342{
 343	struct page *page, *next;
 344
 345	list_for_each_entry_safe(page, next, list, lru) {
 346		list_del(&page->lru);
 347		kimage_free_pages(page);
 348	}
 349}
 350
 351static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
 352							unsigned int order)
 353{
 354	/* Control pages are special, they are the intermediaries
 355	 * that are needed while we copy the rest of the pages
 356	 * to their final resting place.  As such they must
 357	 * not conflict with either the destination addresses
 358	 * or memory the kernel is already using.
 359	 *
 360	 * The only case where we really need more than one of
 361	 * these are for architectures where we cannot disable
 362	 * the MMU and must instead generate an identity mapped
 363	 * page table for all of the memory.
 364	 *
 365	 * At worst this runs in O(N) of the image size.
 366	 */
 367	struct list_head extra_pages;
 368	struct page *pages;
 369	unsigned int count;
 370
 371	count = 1 << order;
 372	INIT_LIST_HEAD(&extra_pages);
 373
 374	/* Loop while I can allocate a page and the page allocated
 375	 * is a destination page.
 376	 */
 377	do {
 378		unsigned long pfn, epfn, addr, eaddr;
 379
 380		pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
 381		if (!pages)
 382			break;
 383		pfn   = page_to_boot_pfn(pages);
 384		epfn  = pfn + count;
 385		addr  = pfn << PAGE_SHIFT;
 386		eaddr = epfn << PAGE_SHIFT;
 387		if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
 388			      kimage_is_destination_range(image, addr, eaddr)) {
 389			list_add(&pages->lru, &extra_pages);
 390			pages = NULL;
 391		}
 392	} while (!pages);
 393
 394	if (pages) {
 395		/* Remember the allocated page... */
 396		list_add(&pages->lru, &image->control_pages);
 397
 398		/* Because the page is already in it's destination
 399		 * location we will never allocate another page at
 400		 * that address.  Therefore kimage_alloc_pages
 401		 * will not return it (again) and we don't need
 402		 * to give it an entry in image->segment[].
 403		 */
 404	}
 405	/* Deal with the destination pages I have inadvertently allocated.
 406	 *
 407	 * Ideally I would convert multi-page allocations into single
 408	 * page allocations, and add everything to image->dest_pages.
 409	 *
 410	 * For now it is simpler to just free the pages.
 411	 */
 412	kimage_free_page_list(&extra_pages);
 413
 414	return pages;
 415}
 416
 417static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
 418						      unsigned int order)
 419{
 420	/* Control pages are special, they are the intermediaries
 421	 * that are needed while we copy the rest of the pages
 422	 * to their final resting place.  As such they must
 423	 * not conflict with either the destination addresses
 424	 * or memory the kernel is already using.
 425	 *
 426	 * Control pages are also the only pags we must allocate
 427	 * when loading a crash kernel.  All of the other pages
 428	 * are specified by the segments and we just memcpy
 429	 * into them directly.
 430	 *
 431	 * The only case where we really need more than one of
 432	 * these are for architectures where we cannot disable
 433	 * the MMU and must instead generate an identity mapped
 434	 * page table for all of the memory.
 435	 *
 436	 * Given the low demand this implements a very simple
 437	 * allocator that finds the first hole of the appropriate
 438	 * size in the reserved memory region, and allocates all
 439	 * of the memory up to and including the hole.
 440	 */
 441	unsigned long hole_start, hole_end, size;
 442	struct page *pages;
 443
 444	pages = NULL;
 445	size = (1 << order) << PAGE_SHIFT;
 446	hole_start = (image->control_page + (size - 1)) & ~(size - 1);
 447	hole_end   = hole_start + size - 1;
 448	while (hole_end <= crashk_res.end) {
 449		unsigned long i;
 450
 451		cond_resched();
 452
 453		if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
 454			break;
 455		/* See if I overlap any of the segments */
 456		for (i = 0; i < image->nr_segments; i++) {
 457			unsigned long mstart, mend;
 458
 459			mstart = image->segment[i].mem;
 460			mend   = mstart + image->segment[i].memsz - 1;
 461			if ((hole_end >= mstart) && (hole_start <= mend)) {
 462				/* Advance the hole to the end of the segment */
 463				hole_start = (mend + (size - 1)) & ~(size - 1);
 464				hole_end   = hole_start + size - 1;
 465				break;
 466			}
 467		}
 468		/* If I don't overlap any segments I have found my hole! */
 469		if (i == image->nr_segments) {
 470			pages = pfn_to_page(hole_start >> PAGE_SHIFT);
 471			image->control_page = hole_end;
 472			break;
 473		}
 474	}
 475
 476	/* Ensure that these pages are decrypted if SME is enabled. */
 477	if (pages)
 478		arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0);
 479
 480	return pages;
 481}
 482
 483
 484struct page *kimage_alloc_control_pages(struct kimage *image,
 485					 unsigned int order)
 486{
 487	struct page *pages = NULL;
 488
 489	switch (image->type) {
 490	case KEXEC_TYPE_DEFAULT:
 491		pages = kimage_alloc_normal_control_pages(image, order);
 492		break;
 493	case KEXEC_TYPE_CRASH:
 494		pages = kimage_alloc_crash_control_pages(image, order);
 495		break;
 496	}
 497
 498	return pages;
 499}
 500
 501int kimage_crash_copy_vmcoreinfo(struct kimage *image)
 502{
 503	struct page *vmcoreinfo_page;
 504	void *safecopy;
 505
 506	if (image->type != KEXEC_TYPE_CRASH)
 507		return 0;
 508
 509	/*
 510	 * For kdump, allocate one vmcoreinfo safe copy from the
 511	 * crash memory. as we have arch_kexec_protect_crashkres()
 512	 * after kexec syscall, we naturally protect it from write
 513	 * (even read) access under kernel direct mapping. But on
 514	 * the other hand, we still need to operate it when crash
 515	 * happens to generate vmcoreinfo note, hereby we rely on
 516	 * vmap for this purpose.
 517	 */
 518	vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
 519	if (!vmcoreinfo_page) {
 520		pr_warn("Could not allocate vmcoreinfo buffer\n");
 521		return -ENOMEM;
 522	}
 523	safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
 524	if (!safecopy) {
 525		pr_warn("Could not vmap vmcoreinfo buffer\n");
 526		return -ENOMEM;
 527	}
 528
 529	image->vmcoreinfo_data_copy = safecopy;
 530	crash_update_vmcoreinfo_safecopy(safecopy);
 531
 532	return 0;
 533}
 534
 535static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
 536{
 537	if (*image->entry != 0)
 538		image->entry++;
 539
 540	if (image->entry == image->last_entry) {
 541		kimage_entry_t *ind_page;
 542		struct page *page;
 543
 544		page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
 545		if (!page)
 546			return -ENOMEM;
 547
 548		ind_page = page_address(page);
 549		*image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
 550		image->entry = ind_page;
 551		image->last_entry = ind_page +
 552				      ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
 553	}
 554	*image->entry = entry;
 555	image->entry++;
 556	*image->entry = 0;
 557
 558	return 0;
 559}
 560
 561static int kimage_set_destination(struct kimage *image,
 562				   unsigned long destination)
 563{
 564	destination &= PAGE_MASK;
 565
 566	return kimage_add_entry(image, destination | IND_DESTINATION);
 567}
 568
 569
 570static int kimage_add_page(struct kimage *image, unsigned long page)
 571{
 572	page &= PAGE_MASK;
 573
 574	return kimage_add_entry(image, page | IND_SOURCE);
 575}
 576
 577
 578static void kimage_free_extra_pages(struct kimage *image)
 579{
 580	/* Walk through and free any extra destination pages I may have */
 581	kimage_free_page_list(&image->dest_pages);
 582
 583	/* Walk through and free any unusable pages I have cached */
 584	kimage_free_page_list(&image->unusable_pages);
 585
 586}
 587
 588void kimage_terminate(struct kimage *image)
 589{
 590	if (*image->entry != 0)
 591		image->entry++;
 592
 593	*image->entry = IND_DONE;
 594}
 595
 596#define for_each_kimage_entry(image, ptr, entry) \
 597	for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
 598		ptr = (entry & IND_INDIRECTION) ? \
 599			boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
 600
 601static void kimage_free_entry(kimage_entry_t entry)
 602{
 603	struct page *page;
 604
 605	page = boot_pfn_to_page(entry >> PAGE_SHIFT);
 606	kimage_free_pages(page);
 607}
 608
 609void kimage_free(struct kimage *image)
 610{
 611	kimage_entry_t *ptr, entry;
 612	kimage_entry_t ind = 0;
 613
 614	if (!image)
 615		return;
 616
 617	if (image->vmcoreinfo_data_copy) {
 618		crash_update_vmcoreinfo_safecopy(NULL);
 619		vunmap(image->vmcoreinfo_data_copy);
 620	}
 621
 622	kimage_free_extra_pages(image);
 623	for_each_kimage_entry(image, ptr, entry) {
 624		if (entry & IND_INDIRECTION) {
 625			/* Free the previous indirection page */
 626			if (ind & IND_INDIRECTION)
 627				kimage_free_entry(ind);
 628			/* Save this indirection page until we are
 629			 * done with it.
 630			 */
 631			ind = entry;
 632		} else if (entry & IND_SOURCE)
 633			kimage_free_entry(entry);
 634	}
 635	/* Free the final indirection page */
 636	if (ind & IND_INDIRECTION)
 637		kimage_free_entry(ind);
 638
 639	/* Handle any machine specific cleanup */
 640	machine_kexec_cleanup(image);
 641
 642	/* Free the kexec control pages... */
 643	kimage_free_page_list(&image->control_pages);
 644
 645	/*
 646	 * Free up any temporary buffers allocated. This might hit if
 647	 * error occurred much later after buffer allocation.
 648	 */
 649	if (image->file_mode)
 650		kimage_file_post_load_cleanup(image);
 651
 652	kfree(image);
 653}
 654
 655static kimage_entry_t *kimage_dst_used(struct kimage *image,
 656					unsigned long page)
 657{
 658	kimage_entry_t *ptr, entry;
 659	unsigned long destination = 0;
 660
 661	for_each_kimage_entry(image, ptr, entry) {
 662		if (entry & IND_DESTINATION)
 663			destination = entry & PAGE_MASK;
 664		else if (entry & IND_SOURCE) {
 665			if (page == destination)
 666				return ptr;
 667			destination += PAGE_SIZE;
 668		}
 669	}
 670
 671	return NULL;
 672}
 673
 674static struct page *kimage_alloc_page(struct kimage *image,
 675					gfp_t gfp_mask,
 676					unsigned long destination)
 677{
 678	/*
 679	 * Here we implement safeguards to ensure that a source page
 680	 * is not copied to its destination page before the data on
 681	 * the destination page is no longer useful.
 682	 *
 683	 * To do this we maintain the invariant that a source page is
 684	 * either its own destination page, or it is not a
 685	 * destination page at all.
 686	 *
 687	 * That is slightly stronger than required, but the proof
 688	 * that no problems will not occur is trivial, and the
 689	 * implementation is simply to verify.
 690	 *
 691	 * When allocating all pages normally this algorithm will run
 692	 * in O(N) time, but in the worst case it will run in O(N^2)
 693	 * time.   If the runtime is a problem the data structures can
 694	 * be fixed.
 695	 */
 696	struct page *page;
 697	unsigned long addr;
 698
 699	/*
 700	 * Walk through the list of destination pages, and see if I
 701	 * have a match.
 702	 */
 703	list_for_each_entry(page, &image->dest_pages, lru) {
 704		addr = page_to_boot_pfn(page) << PAGE_SHIFT;
 705		if (addr == destination) {
 706			list_del(&page->lru);
 707			return page;
 708		}
 709	}
 710	page = NULL;
 711	while (1) {
 712		kimage_entry_t *old;
 713
 714		/* Allocate a page, if we run out of memory give up */
 715		page = kimage_alloc_pages(gfp_mask, 0);
 716		if (!page)
 717			return NULL;
 718		/* If the page cannot be used file it away */
 719		if (page_to_boot_pfn(page) >
 720				(KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
 721			list_add(&page->lru, &image->unusable_pages);
 722			continue;
 723		}
 724		addr = page_to_boot_pfn(page) << PAGE_SHIFT;
 725
 726		/* If it is the destination page we want use it */
 727		if (addr == destination)
 728			break;
 729
 730		/* If the page is not a destination page use it */
 731		if (!kimage_is_destination_range(image, addr,
 732						  addr + PAGE_SIZE))
 733			break;
 734
 735		/*
 736		 * I know that the page is someones destination page.
 737		 * See if there is already a source page for this
 738		 * destination page.  And if so swap the source pages.
 739		 */
 740		old = kimage_dst_used(image, addr);
 741		if (old) {
 742			/* If so move it */
 743			unsigned long old_addr;
 744			struct page *old_page;
 745
 746			old_addr = *old & PAGE_MASK;
 747			old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
 748			copy_highpage(page, old_page);
 749			*old = addr | (*old & ~PAGE_MASK);
 750
 751			/* The old page I have found cannot be a
 752			 * destination page, so return it if it's
 753			 * gfp_flags honor the ones passed in.
 754			 */
 755			if (!(gfp_mask & __GFP_HIGHMEM) &&
 756			    PageHighMem(old_page)) {
 757				kimage_free_pages(old_page);
 758				continue;
 759			}
 760			page = old_page;
 761			break;
 762		}
 763		/* Place the page on the destination list, to be used later */
 764		list_add(&page->lru, &image->dest_pages);
 765	}
 766
 767	return page;
 768}
 769
 770static int kimage_load_normal_segment(struct kimage *image,
 771					 struct kexec_segment *segment)
 772{
 773	unsigned long maddr;
 774	size_t ubytes, mbytes;
 775	int result;
 776	unsigned char __user *buf = NULL;
 777	unsigned char *kbuf = NULL;
 778
 779	if (image->file_mode)
 780		kbuf = segment->kbuf;
 781	else
 782		buf = segment->buf;
 783	ubytes = segment->bufsz;
 784	mbytes = segment->memsz;
 785	maddr = segment->mem;
 786
 787	result = kimage_set_destination(image, maddr);
 788	if (result < 0)
 789		goto out;
 790
 791	while (mbytes) {
 792		struct page *page;
 793		char *ptr;
 794		size_t uchunk, mchunk;
 795
 796		page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
 797		if (!page) {
 798			result  = -ENOMEM;
 799			goto out;
 800		}
 801		result = kimage_add_page(image, page_to_boot_pfn(page)
 802								<< PAGE_SHIFT);
 803		if (result < 0)
 804			goto out;
 805
 806		ptr = kmap_local_page(page);
 807		/* Start with a clear page */
 808		clear_page(ptr);
 809		ptr += maddr & ~PAGE_MASK;
 810		mchunk = min_t(size_t, mbytes,
 811				PAGE_SIZE - (maddr & ~PAGE_MASK));
 812		uchunk = min(ubytes, mchunk);
 813
 814		/* For file based kexec, source pages are in kernel memory */
 815		if (image->file_mode)
 816			memcpy(ptr, kbuf, uchunk);
 817		else
 818			result = copy_from_user(ptr, buf, uchunk);
 819		kunmap_local(ptr);
 820		if (result) {
 821			result = -EFAULT;
 822			goto out;
 823		}
 824		ubytes -= uchunk;
 825		maddr  += mchunk;
 826		if (image->file_mode)
 827			kbuf += mchunk;
 828		else
 829			buf += mchunk;
 830		mbytes -= mchunk;
 831
 832		cond_resched();
 833	}
 834out:
 835	return result;
 836}
 837
 838static int kimage_load_crash_segment(struct kimage *image,
 839					struct kexec_segment *segment)
 840{
 841	/* For crash dumps kernels we simply copy the data from
 842	 * user space to it's destination.
 843	 * We do things a page at a time for the sake of kmap.
 844	 */
 845	unsigned long maddr;
 846	size_t ubytes, mbytes;
 847	int result;
 848	unsigned char __user *buf = NULL;
 849	unsigned char *kbuf = NULL;
 850
 851	result = 0;
 852	if (image->file_mode)
 853		kbuf = segment->kbuf;
 854	else
 855		buf = segment->buf;
 856	ubytes = segment->bufsz;
 857	mbytes = segment->memsz;
 858	maddr = segment->mem;
 859	while (mbytes) {
 860		struct page *page;
 861		char *ptr;
 862		size_t uchunk, mchunk;
 863
 864		page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
 865		if (!page) {
 866			result  = -ENOMEM;
 867			goto out;
 868		}
 869		arch_kexec_post_alloc_pages(page_address(page), 1, 0);
 870		ptr = kmap_local_page(page);
 871		ptr += maddr & ~PAGE_MASK;
 872		mchunk = min_t(size_t, mbytes,
 873				PAGE_SIZE - (maddr & ~PAGE_MASK));
 874		uchunk = min(ubytes, mchunk);
 875		if (mchunk > uchunk) {
 876			/* Zero the trailing part of the page */
 877			memset(ptr + uchunk, 0, mchunk - uchunk);
 878		}
 879
 880		/* For file based kexec, source pages are in kernel memory */
 881		if (image->file_mode)
 882			memcpy(ptr, kbuf, uchunk);
 883		else
 884			result = copy_from_user(ptr, buf, uchunk);
 885		kexec_flush_icache_page(page);
 886		kunmap_local(ptr);
 887		arch_kexec_pre_free_pages(page_address(page), 1);
 888		if (result) {
 889			result = -EFAULT;
 890			goto out;
 891		}
 892		ubytes -= uchunk;
 893		maddr  += mchunk;
 894		if (image->file_mode)
 895			kbuf += mchunk;
 896		else
 897			buf += mchunk;
 898		mbytes -= mchunk;
 899
 900		cond_resched();
 901	}
 902out:
 903	return result;
 904}
 905
 906int kimage_load_segment(struct kimage *image,
 907				struct kexec_segment *segment)
 908{
 909	int result = -ENOMEM;
 910
 911	switch (image->type) {
 912	case KEXEC_TYPE_DEFAULT:
 913		result = kimage_load_normal_segment(image, segment);
 914		break;
 915	case KEXEC_TYPE_CRASH:
 916		result = kimage_load_crash_segment(image, segment);
 917		break;
 918	}
 919
 920	return result;
 921}
 922
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 923struct kimage *kexec_image;
 924struct kimage *kexec_crash_image;
 925int kexec_load_disabled;
 
 926#ifdef CONFIG_SYSCTL
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 927static struct ctl_table kexec_core_sysctls[] = {
 928	{
 929		.procname	= "kexec_load_disabled",
 930		.data		= &kexec_load_disabled,
 931		.maxlen		= sizeof(int),
 932		.mode		= 0644,
 933		/* only handle a transition from default "0" to "1" */
 934		.proc_handler	= proc_dointvec_minmax,
 935		.extra1		= SYSCTL_ONE,
 936		.extra2		= SYSCTL_ONE,
 937	},
 
 
 
 
 
 
 
 
 
 
 
 
 938	{ }
 939};
 940
 941static int __init kexec_core_sysctl_init(void)
 942{
 943	register_sysctl_init("kernel", kexec_core_sysctls);
 944	return 0;
 945}
 946late_initcall(kexec_core_sysctl_init);
 947#endif
 948
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 949/*
 950 * No panic_cpu check version of crash_kexec().  This function is called
 951 * only when panic_cpu holds the current CPU number; this is the only CPU
 952 * which processes crash_kexec routines.
 953 */
 954void __noclone __crash_kexec(struct pt_regs *regs)
 955{
 956	/* Take the kexec_lock here to prevent sys_kexec_load
 957	 * running on one cpu from replacing the crash kernel
 958	 * we are using after a panic on a different cpu.
 959	 *
 960	 * If the crash kernel was not located in a fixed area
 961	 * of memory the xchg(&kexec_crash_image) would be
 962	 * sufficient.  But since I reuse the memory...
 963	 */
 964	if (kexec_trylock()) {
 965		if (kexec_crash_image) {
 966			struct pt_regs fixed_regs;
 967
 968			crash_setup_regs(&fixed_regs, regs);
 969			crash_save_vmcoreinfo();
 970			machine_crash_shutdown(&fixed_regs);
 971			machine_kexec(kexec_crash_image);
 972		}
 973		kexec_unlock();
 974	}
 975}
 976STACK_FRAME_NON_STANDARD(__crash_kexec);
 977
 978void crash_kexec(struct pt_regs *regs)
 979{
 980	int old_cpu, this_cpu;
 981
 982	/*
 983	 * Only one CPU is allowed to execute the crash_kexec() code as with
 984	 * panic().  Otherwise parallel calls of panic() and crash_kexec()
 985	 * may stop each other.  To exclude them, we use panic_cpu here too.
 986	 */
 
 987	this_cpu = raw_smp_processor_id();
 988	old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
 989	if (old_cpu == PANIC_CPU_INVALID) {
 990		/* This is the 1st CPU which comes here, so go ahead. */
 991		__crash_kexec(regs);
 992
 993		/*
 994		 * Reset panic_cpu to allow another panic()/crash_kexec()
 995		 * call.
 996		 */
 997		atomic_set(&panic_cpu, PANIC_CPU_INVALID);
 998	}
 999}
1000
 
 
 
 
 
1001ssize_t crash_get_memory_size(void)
1002{
1003	ssize_t size = 0;
1004
1005	if (!kexec_trylock())
1006		return -EBUSY;
1007
1008	if (crashk_res.end != crashk_res.start)
1009		size = resource_size(&crashk_res);
1010
1011	kexec_unlock();
1012	return size;
1013}
1014
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1015int crash_shrink_memory(unsigned long new_size)
1016{
1017	int ret = 0;
1018	unsigned long start, end;
1019	unsigned long old_size;
1020	struct resource *ram_res;
1021
1022	if (!kexec_trylock())
1023		return -EBUSY;
1024
1025	if (kexec_crash_image) {
1026		ret = -ENOENT;
1027		goto unlock;
1028	}
1029	start = crashk_res.start;
1030	end = crashk_res.end;
1031	old_size = (end == 0) ? 0 : end - start + 1;
 
1032	if (new_size >= old_size) {
1033		ret = (new_size == old_size) ? 0 : -EINVAL;
1034		goto unlock;
1035	}
1036
1037	ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1038	if (!ram_res) {
1039		ret = -ENOMEM;
1040		goto unlock;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1041	}
1042
1043	start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1044	end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1045
1046	crash_free_reserved_phys_range(end, crashk_res.end);
1047
1048	if ((start == end) && (crashk_res.parent != NULL))
1049		release_resource(&crashk_res);
1050
1051	ram_res->start = end;
1052	ram_res->end = crashk_res.end;
1053	ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
1054	ram_res->name = "System RAM";
1055
1056	crashk_res.end = end - 1;
1057
1058	insert_resource(&iomem_resource, ram_res);
1059
1060unlock:
1061	kexec_unlock();
1062	return ret;
1063}
1064
1065void crash_save_cpu(struct pt_regs *regs, int cpu)
1066{
1067	struct elf_prstatus prstatus;
1068	u32 *buf;
1069
1070	if ((cpu < 0) || (cpu >= nr_cpu_ids))
1071		return;
1072
1073	/* Using ELF notes here is opportunistic.
1074	 * I need a well defined structure format
1075	 * for the data I pass, and I need tags
1076	 * on the data to indicate what information I have
1077	 * squirrelled away.  ELF notes happen to provide
1078	 * all of that, so there is no need to invent something new.
1079	 */
1080	buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1081	if (!buf)
1082		return;
1083	memset(&prstatus, 0, sizeof(prstatus));
1084	prstatus.common.pr_pid = current->pid;
1085	elf_core_copy_regs(&prstatus.pr_reg, regs);
1086	buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1087			      &prstatus, sizeof(prstatus));
1088	final_note(buf);
1089}
1090
1091static int __init crash_notes_memory_init(void)
1092{
1093	/* Allocate memory for saving cpu registers. */
1094	size_t size, align;
1095
1096	/*
1097	 * crash_notes could be allocated across 2 vmalloc pages when percpu
1098	 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
1099	 * pages are also on 2 continuous physical pages. In this case the
1100	 * 2nd part of crash_notes in 2nd page could be lost since only the
1101	 * starting address and size of crash_notes are exported through sysfs.
1102	 * Here round up the size of crash_notes to the nearest power of two
1103	 * and pass it to __alloc_percpu as align value. This can make sure
1104	 * crash_notes is allocated inside one physical page.
1105	 */
1106	size = sizeof(note_buf_t);
1107	align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
1108
1109	/*
1110	 * Break compile if size is bigger than PAGE_SIZE since crash_notes
1111	 * definitely will be in 2 pages with that.
1112	 */
1113	BUILD_BUG_ON(size > PAGE_SIZE);
1114
1115	crash_notes = __alloc_percpu(size, align);
1116	if (!crash_notes) {
1117		pr_warn("Memory allocation for saving cpu register states failed\n");
1118		return -ENOMEM;
1119	}
1120	return 0;
1121}
1122subsys_initcall(crash_notes_memory_init);
1123
1124
1125/*
1126 * Move into place and start executing a preloaded standalone
1127 * executable.  If nothing was preloaded return an error.
1128 */
1129int kernel_kexec(void)
1130{
1131	int error = 0;
1132
1133	if (!kexec_trylock())
1134		return -EBUSY;
1135	if (!kexec_image) {
1136		error = -EINVAL;
1137		goto Unlock;
1138	}
1139
1140#ifdef CONFIG_KEXEC_JUMP
1141	if (kexec_image->preserve_context) {
1142		pm_prepare_console();
1143		error = freeze_processes();
1144		if (error) {
1145			error = -EBUSY;
1146			goto Restore_console;
1147		}
1148		suspend_console();
1149		error = dpm_suspend_start(PMSG_FREEZE);
1150		if (error)
1151			goto Resume_console;
1152		/* At this point, dpm_suspend_start() has been called,
1153		 * but *not* dpm_suspend_end(). We *must* call
1154		 * dpm_suspend_end() now.  Otherwise, drivers for
1155		 * some devices (e.g. interrupt controllers) become
1156		 * desynchronized with the actual state of the
1157		 * hardware at resume time, and evil weirdness ensues.
1158		 */
1159		error = dpm_suspend_end(PMSG_FREEZE);
1160		if (error)
1161			goto Resume_devices;
1162		error = suspend_disable_secondary_cpus();
1163		if (error)
1164			goto Enable_cpus;
1165		local_irq_disable();
1166		error = syscore_suspend();
1167		if (error)
1168			goto Enable_irqs;
1169	} else
1170#endif
1171	{
1172		kexec_in_progress = true;
1173		kernel_restart_prepare("kexec reboot");
1174		migrate_to_reboot_cpu();
 
1175
1176		/*
1177		 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1178		 * no further code needs to use CPU hotplug (which is true in
1179		 * the reboot case). However, the kexec path depends on using
1180		 * CPU hotplug again; so re-enable it here.
1181		 */
1182		cpu_hotplug_enable();
1183		pr_notice("Starting new kernel\n");
1184		machine_shutdown();
1185	}
1186
1187	kmsg_dump(KMSG_DUMP_SHUTDOWN);
1188	machine_kexec(kexec_image);
1189
1190#ifdef CONFIG_KEXEC_JUMP
1191	if (kexec_image->preserve_context) {
1192		syscore_resume();
1193 Enable_irqs:
1194		local_irq_enable();
1195 Enable_cpus:
1196		suspend_enable_secondary_cpus();
1197		dpm_resume_start(PMSG_RESTORE);
1198 Resume_devices:
1199		dpm_resume_end(PMSG_RESTORE);
1200 Resume_console:
1201		resume_console();
1202		thaw_processes();
1203 Restore_console:
1204		pm_restore_console();
1205	}
1206#endif
1207
1208 Unlock:
1209	kexec_unlock();
1210	return error;
1211}
v6.8
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * kexec.c - kexec system call core code.
   4 * Copyright (C) 2002-2004 Eric Biederman  <ebiederm@xmission.com>
   5 */
   6
   7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9#include <linux/btf.h>
  10#include <linux/capability.h>
  11#include <linux/mm.h>
  12#include <linux/file.h>
  13#include <linux/slab.h>
  14#include <linux/fs.h>
  15#include <linux/kexec.h>
  16#include <linux/mutex.h>
  17#include <linux/list.h>
  18#include <linux/highmem.h>
  19#include <linux/syscalls.h>
  20#include <linux/reboot.h>
  21#include <linux/ioport.h>
  22#include <linux/hardirq.h>
  23#include <linux/elf.h>
  24#include <linux/elfcore.h>
  25#include <linux/utsname.h>
  26#include <linux/numa.h>
  27#include <linux/suspend.h>
  28#include <linux/device.h>
  29#include <linux/freezer.h>
  30#include <linux/panic_notifier.h>
  31#include <linux/pm.h>
  32#include <linux/cpu.h>
  33#include <linux/uaccess.h>
  34#include <linux/io.h>
  35#include <linux/console.h>
  36#include <linux/vmalloc.h>
  37#include <linux/swap.h>
  38#include <linux/syscore_ops.h>
  39#include <linux/compiler.h>
  40#include <linux/hugetlb.h>
  41#include <linux/objtool.h>
  42#include <linux/kmsg_dump.h>
  43
  44#include <asm/page.h>
  45#include <asm/sections.h>
  46
  47#include <crypto/hash.h>
  48#include "kexec_internal.h"
  49
  50atomic_t __kexec_lock = ATOMIC_INIT(0);
  51
 
 
 
  52/* Flag to indicate we are going to kexec a new kernel */
  53bool kexec_in_progress = false;
  54
  55bool kexec_file_dbg_print;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  56
  57int kexec_should_crash(struct task_struct *p)
  58{
  59	/*
  60	 * If crash_kexec_post_notifiers is enabled, don't run
  61	 * crash_kexec() here yet, which must be run after panic
  62	 * notifiers in panic().
  63	 */
  64	if (crash_kexec_post_notifiers)
  65		return 0;
  66	/*
  67	 * There are 4 panic() calls in make_task_dead() path, each of which
  68	 * corresponds to each of these 4 conditions.
  69	 */
  70	if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
  71		return 1;
  72	return 0;
  73}
  74
  75int kexec_crash_loaded(void)
  76{
  77	return !!kexec_crash_image;
  78}
  79EXPORT_SYMBOL_GPL(kexec_crash_loaded);
  80
  81/*
  82 * When kexec transitions to the new kernel there is a one-to-one
  83 * mapping between physical and virtual addresses.  On processors
  84 * where you can disable the MMU this is trivial, and easy.  For
  85 * others it is still a simple predictable page table to setup.
  86 *
  87 * In that environment kexec copies the new kernel to its final
  88 * resting place.  This means I can only support memory whose
  89 * physical address can fit in an unsigned long.  In particular
  90 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
  91 * If the assembly stub has more restrictive requirements
  92 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
  93 * defined more restrictively in <asm/kexec.h>.
  94 *
  95 * The code for the transition from the current kernel to the
  96 * new kernel is placed in the control_code_buffer, whose size
  97 * is given by KEXEC_CONTROL_PAGE_SIZE.  In the best case only a single
  98 * page of memory is necessary, but some architectures require more.
  99 * Because this memory must be identity mapped in the transition from
 100 * virtual to physical addresses it must live in the range
 101 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
 102 * modifiable.
 103 *
 104 * The assembly stub in the control code buffer is passed a linked list
 105 * of descriptor pages detailing the source pages of the new kernel,
 106 * and the destination addresses of those source pages.  As this data
 107 * structure is not used in the context of the current OS, it must
 108 * be self-contained.
 109 *
 110 * The code has been made to work with highmem pages and will use a
 111 * destination page in its final resting place (if it happens
 112 * to allocate it).  The end product of this is that most of the
 113 * physical address space, and most of RAM can be used.
 114 *
 115 * Future directions include:
 116 *  - allocating a page table with the control code buffer identity
 117 *    mapped, to simplify machine_kexec and make kexec_on_panic more
 118 *    reliable.
 119 */
 120
 121/*
 122 * KIMAGE_NO_DEST is an impossible destination address..., for
 123 * allocating pages whose destination address we do not care about.
 124 */
 125#define KIMAGE_NO_DEST (-1UL)
 126#define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
 127
 128static struct page *kimage_alloc_page(struct kimage *image,
 129				       gfp_t gfp_mask,
 130				       unsigned long dest);
 131
 132int sanity_check_segment_list(struct kimage *image)
 133{
 134	int i;
 135	unsigned long nr_segments = image->nr_segments;
 136	unsigned long total_pages = 0;
 137	unsigned long nr_pages = totalram_pages();
 138
 139	/*
 140	 * Verify we have good destination addresses.  The caller is
 141	 * responsible for making certain we don't attempt to load
 142	 * the new image into invalid or reserved areas of RAM.  This
 143	 * just verifies it is an address we can use.
 144	 *
 145	 * Since the kernel does everything in page size chunks ensure
 146	 * the destination addresses are page aligned.  Too many
 147	 * special cases crop of when we don't do this.  The most
 148	 * insidious is getting overlapping destination addresses
 149	 * simply because addresses are changed to page size
 150	 * granularity.
 151	 */
 152	for (i = 0; i < nr_segments; i++) {
 153		unsigned long mstart, mend;
 154
 155		mstart = image->segment[i].mem;
 156		mend   = mstart + image->segment[i].memsz;
 157		if (mstart > mend)
 158			return -EADDRNOTAVAIL;
 159		if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
 160			return -EADDRNOTAVAIL;
 161		if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
 162			return -EADDRNOTAVAIL;
 163	}
 164
 165	/* Verify our destination addresses do not overlap.
 166	 * If we alloed overlapping destination addresses
 167	 * through very weird things can happen with no
 168	 * easy explanation as one segment stops on another.
 169	 */
 170	for (i = 0; i < nr_segments; i++) {
 171		unsigned long mstart, mend;
 172		unsigned long j;
 173
 174		mstart = image->segment[i].mem;
 175		mend   = mstart + image->segment[i].memsz;
 176		for (j = 0; j < i; j++) {
 177			unsigned long pstart, pend;
 178
 179			pstart = image->segment[j].mem;
 180			pend   = pstart + image->segment[j].memsz;
 181			/* Do the segments overlap ? */
 182			if ((mend > pstart) && (mstart < pend))
 183				return -EINVAL;
 184		}
 185	}
 186
 187	/* Ensure our buffer sizes are strictly less than
 188	 * our memory sizes.  This should always be the case,
 189	 * and it is easier to check up front than to be surprised
 190	 * later on.
 191	 */
 192	for (i = 0; i < nr_segments; i++) {
 193		if (image->segment[i].bufsz > image->segment[i].memsz)
 194			return -EINVAL;
 195	}
 196
 197	/*
 198	 * Verify that no more than half of memory will be consumed. If the
 199	 * request from userspace is too large, a large amount of time will be
 200	 * wasted allocating pages, which can cause a soft lockup.
 201	 */
 202	for (i = 0; i < nr_segments; i++) {
 203		if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2)
 204			return -EINVAL;
 205
 206		total_pages += PAGE_COUNT(image->segment[i].memsz);
 207	}
 208
 209	if (total_pages > nr_pages / 2)
 210		return -EINVAL;
 211
 212	/*
 213	 * Verify we have good destination addresses.  Normally
 214	 * the caller is responsible for making certain we don't
 215	 * attempt to load the new image into invalid or reserved
 216	 * areas of RAM.  But crash kernels are preloaded into a
 217	 * reserved area of ram.  We must ensure the addresses
 218	 * are in the reserved area otherwise preloading the
 219	 * kernel could corrupt things.
 220	 */
 221
 222	if (image->type == KEXEC_TYPE_CRASH) {
 223		for (i = 0; i < nr_segments; i++) {
 224			unsigned long mstart, mend;
 225
 226			mstart = image->segment[i].mem;
 227			mend = mstart + image->segment[i].memsz - 1;
 228			/* Ensure we are within the crash kernel limits */
 229			if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
 230			    (mend > phys_to_boot_phys(crashk_res.end)))
 231				return -EADDRNOTAVAIL;
 232		}
 233	}
 234
 235	return 0;
 236}
 237
 238struct kimage *do_kimage_alloc_init(void)
 239{
 240	struct kimage *image;
 241
 242	/* Allocate a controlling structure */
 243	image = kzalloc(sizeof(*image), GFP_KERNEL);
 244	if (!image)
 245		return NULL;
 246
 247	image->head = 0;
 248	image->entry = &image->head;
 249	image->last_entry = &image->head;
 250	image->control_page = ~0; /* By default this does not apply */
 251	image->type = KEXEC_TYPE_DEFAULT;
 252
 253	/* Initialize the list of control pages */
 254	INIT_LIST_HEAD(&image->control_pages);
 255
 256	/* Initialize the list of destination pages */
 257	INIT_LIST_HEAD(&image->dest_pages);
 258
 259	/* Initialize the list of unusable pages */
 260	INIT_LIST_HEAD(&image->unusable_pages);
 261
 262#ifdef CONFIG_CRASH_HOTPLUG
 263	image->hp_action = KEXEC_CRASH_HP_NONE;
 264	image->elfcorehdr_index = -1;
 265	image->elfcorehdr_updated = false;
 266#endif
 267
 268	return image;
 269}
 270
 271int kimage_is_destination_range(struct kimage *image,
 272					unsigned long start,
 273					unsigned long end)
 274{
 275	unsigned long i;
 276
 277	for (i = 0; i < image->nr_segments; i++) {
 278		unsigned long mstart, mend;
 279
 280		mstart = image->segment[i].mem;
 281		mend = mstart + image->segment[i].memsz - 1;
 282		if ((end >= mstart) && (start <= mend))
 283			return 1;
 284	}
 285
 286	return 0;
 287}
 288
 289static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
 290{
 291	struct page *pages;
 292
 293	if (fatal_signal_pending(current))
 294		return NULL;
 295	pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order);
 296	if (pages) {
 297		unsigned int count, i;
 298
 299		pages->mapping = NULL;
 300		set_page_private(pages, order);
 301		count = 1 << order;
 302		for (i = 0; i < count; i++)
 303			SetPageReserved(pages + i);
 304
 305		arch_kexec_post_alloc_pages(page_address(pages), count,
 306					    gfp_mask);
 307
 308		if (gfp_mask & __GFP_ZERO)
 309			for (i = 0; i < count; i++)
 310				clear_highpage(pages + i);
 311	}
 312
 313	return pages;
 314}
 315
 316static void kimage_free_pages(struct page *page)
 317{
 318	unsigned int order, count, i;
 319
 320	order = page_private(page);
 321	count = 1 << order;
 322
 323	arch_kexec_pre_free_pages(page_address(page), count);
 324
 325	for (i = 0; i < count; i++)
 326		ClearPageReserved(page + i);
 327	__free_pages(page, order);
 328}
 329
 330void kimage_free_page_list(struct list_head *list)
 331{
 332	struct page *page, *next;
 333
 334	list_for_each_entry_safe(page, next, list, lru) {
 335		list_del(&page->lru);
 336		kimage_free_pages(page);
 337	}
 338}
 339
 340static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
 341							unsigned int order)
 342{
 343	/* Control pages are special, they are the intermediaries
 344	 * that are needed while we copy the rest of the pages
 345	 * to their final resting place.  As such they must
 346	 * not conflict with either the destination addresses
 347	 * or memory the kernel is already using.
 348	 *
 349	 * The only case where we really need more than one of
 350	 * these are for architectures where we cannot disable
 351	 * the MMU and must instead generate an identity mapped
 352	 * page table for all of the memory.
 353	 *
 354	 * At worst this runs in O(N) of the image size.
 355	 */
 356	struct list_head extra_pages;
 357	struct page *pages;
 358	unsigned int count;
 359
 360	count = 1 << order;
 361	INIT_LIST_HEAD(&extra_pages);
 362
 363	/* Loop while I can allocate a page and the page allocated
 364	 * is a destination page.
 365	 */
 366	do {
 367		unsigned long pfn, epfn, addr, eaddr;
 368
 369		pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
 370		if (!pages)
 371			break;
 372		pfn   = page_to_boot_pfn(pages);
 373		epfn  = pfn + count;
 374		addr  = pfn << PAGE_SHIFT;
 375		eaddr = (epfn << PAGE_SHIFT) - 1;
 376		if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
 377			      kimage_is_destination_range(image, addr, eaddr)) {
 378			list_add(&pages->lru, &extra_pages);
 379			pages = NULL;
 380		}
 381	} while (!pages);
 382
 383	if (pages) {
 384		/* Remember the allocated page... */
 385		list_add(&pages->lru, &image->control_pages);
 386
 387		/* Because the page is already in it's destination
 388		 * location we will never allocate another page at
 389		 * that address.  Therefore kimage_alloc_pages
 390		 * will not return it (again) and we don't need
 391		 * to give it an entry in image->segment[].
 392		 */
 393	}
 394	/* Deal with the destination pages I have inadvertently allocated.
 395	 *
 396	 * Ideally I would convert multi-page allocations into single
 397	 * page allocations, and add everything to image->dest_pages.
 398	 *
 399	 * For now it is simpler to just free the pages.
 400	 */
 401	kimage_free_page_list(&extra_pages);
 402
 403	return pages;
 404}
 405
 406static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
 407						      unsigned int order)
 408{
 409	/* Control pages are special, they are the intermediaries
 410	 * that are needed while we copy the rest of the pages
 411	 * to their final resting place.  As such they must
 412	 * not conflict with either the destination addresses
 413	 * or memory the kernel is already using.
 414	 *
 415	 * Control pages are also the only pags we must allocate
 416	 * when loading a crash kernel.  All of the other pages
 417	 * are specified by the segments and we just memcpy
 418	 * into them directly.
 419	 *
 420	 * The only case where we really need more than one of
 421	 * these are for architectures where we cannot disable
 422	 * the MMU and must instead generate an identity mapped
 423	 * page table for all of the memory.
 424	 *
 425	 * Given the low demand this implements a very simple
 426	 * allocator that finds the first hole of the appropriate
 427	 * size in the reserved memory region, and allocates all
 428	 * of the memory up to and including the hole.
 429	 */
 430	unsigned long hole_start, hole_end, size;
 431	struct page *pages;
 432
 433	pages = NULL;
 434	size = (1 << order) << PAGE_SHIFT;
 435	hole_start = ALIGN(image->control_page, size);
 436	hole_end   = hole_start + size - 1;
 437	while (hole_end <= crashk_res.end) {
 438		unsigned long i;
 439
 440		cond_resched();
 441
 442		if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
 443			break;
 444		/* See if I overlap any of the segments */
 445		for (i = 0; i < image->nr_segments; i++) {
 446			unsigned long mstart, mend;
 447
 448			mstart = image->segment[i].mem;
 449			mend   = mstart + image->segment[i].memsz - 1;
 450			if ((hole_end >= mstart) && (hole_start <= mend)) {
 451				/* Advance the hole to the end of the segment */
 452				hole_start = ALIGN(mend, size);
 453				hole_end   = hole_start + size - 1;
 454				break;
 455			}
 456		}
 457		/* If I don't overlap any segments I have found my hole! */
 458		if (i == image->nr_segments) {
 459			pages = pfn_to_page(hole_start >> PAGE_SHIFT);
 460			image->control_page = hole_end + 1;
 461			break;
 462		}
 463	}
 464
 465	/* Ensure that these pages are decrypted if SME is enabled. */
 466	if (pages)
 467		arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0);
 468
 469	return pages;
 470}
 471
 472
 473struct page *kimage_alloc_control_pages(struct kimage *image,
 474					 unsigned int order)
 475{
 476	struct page *pages = NULL;
 477
 478	switch (image->type) {
 479	case KEXEC_TYPE_DEFAULT:
 480		pages = kimage_alloc_normal_control_pages(image, order);
 481		break;
 482	case KEXEC_TYPE_CRASH:
 483		pages = kimage_alloc_crash_control_pages(image, order);
 484		break;
 485	}
 486
 487	return pages;
 488}
 489
 490int kimage_crash_copy_vmcoreinfo(struct kimage *image)
 491{
 492	struct page *vmcoreinfo_page;
 493	void *safecopy;
 494
 495	if (image->type != KEXEC_TYPE_CRASH)
 496		return 0;
 497
 498	/*
 499	 * For kdump, allocate one vmcoreinfo safe copy from the
 500	 * crash memory. as we have arch_kexec_protect_crashkres()
 501	 * after kexec syscall, we naturally protect it from write
 502	 * (even read) access under kernel direct mapping. But on
 503	 * the other hand, we still need to operate it when crash
 504	 * happens to generate vmcoreinfo note, hereby we rely on
 505	 * vmap for this purpose.
 506	 */
 507	vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
 508	if (!vmcoreinfo_page) {
 509		pr_warn("Could not allocate vmcoreinfo buffer\n");
 510		return -ENOMEM;
 511	}
 512	safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
 513	if (!safecopy) {
 514		pr_warn("Could not vmap vmcoreinfo buffer\n");
 515		return -ENOMEM;
 516	}
 517
 518	image->vmcoreinfo_data_copy = safecopy;
 519	crash_update_vmcoreinfo_safecopy(safecopy);
 520
 521	return 0;
 522}
 523
 524static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
 525{
 526	if (*image->entry != 0)
 527		image->entry++;
 528
 529	if (image->entry == image->last_entry) {
 530		kimage_entry_t *ind_page;
 531		struct page *page;
 532
 533		page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
 534		if (!page)
 535			return -ENOMEM;
 536
 537		ind_page = page_address(page);
 538		*image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
 539		image->entry = ind_page;
 540		image->last_entry = ind_page +
 541				      ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
 542	}
 543	*image->entry = entry;
 544	image->entry++;
 545	*image->entry = 0;
 546
 547	return 0;
 548}
 549
 550static int kimage_set_destination(struct kimage *image,
 551				   unsigned long destination)
 552{
 553	destination &= PAGE_MASK;
 554
 555	return kimage_add_entry(image, destination | IND_DESTINATION);
 556}
 557
 558
 559static int kimage_add_page(struct kimage *image, unsigned long page)
 560{
 561	page &= PAGE_MASK;
 562
 563	return kimage_add_entry(image, page | IND_SOURCE);
 564}
 565
 566
 567static void kimage_free_extra_pages(struct kimage *image)
 568{
 569	/* Walk through and free any extra destination pages I may have */
 570	kimage_free_page_list(&image->dest_pages);
 571
 572	/* Walk through and free any unusable pages I have cached */
 573	kimage_free_page_list(&image->unusable_pages);
 574
 575}
 576
 577void kimage_terminate(struct kimage *image)
 578{
 579	if (*image->entry != 0)
 580		image->entry++;
 581
 582	*image->entry = IND_DONE;
 583}
 584
 585#define for_each_kimage_entry(image, ptr, entry) \
 586	for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
 587		ptr = (entry & IND_INDIRECTION) ? \
 588			boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
 589
 590static void kimage_free_entry(kimage_entry_t entry)
 591{
 592	struct page *page;
 593
 594	page = boot_pfn_to_page(entry >> PAGE_SHIFT);
 595	kimage_free_pages(page);
 596}
 597
 598void kimage_free(struct kimage *image)
 599{
 600	kimage_entry_t *ptr, entry;
 601	kimage_entry_t ind = 0;
 602
 603	if (!image)
 604		return;
 605
 606	if (image->vmcoreinfo_data_copy) {
 607		crash_update_vmcoreinfo_safecopy(NULL);
 608		vunmap(image->vmcoreinfo_data_copy);
 609	}
 610
 611	kimage_free_extra_pages(image);
 612	for_each_kimage_entry(image, ptr, entry) {
 613		if (entry & IND_INDIRECTION) {
 614			/* Free the previous indirection page */
 615			if (ind & IND_INDIRECTION)
 616				kimage_free_entry(ind);
 617			/* Save this indirection page until we are
 618			 * done with it.
 619			 */
 620			ind = entry;
 621		} else if (entry & IND_SOURCE)
 622			kimage_free_entry(entry);
 623	}
 624	/* Free the final indirection page */
 625	if (ind & IND_INDIRECTION)
 626		kimage_free_entry(ind);
 627
 628	/* Handle any machine specific cleanup */
 629	machine_kexec_cleanup(image);
 630
 631	/* Free the kexec control pages... */
 632	kimage_free_page_list(&image->control_pages);
 633
 634	/*
 635	 * Free up any temporary buffers allocated. This might hit if
 636	 * error occurred much later after buffer allocation.
 637	 */
 638	if (image->file_mode)
 639		kimage_file_post_load_cleanup(image);
 640
 641	kfree(image);
 642}
 643
 644static kimage_entry_t *kimage_dst_used(struct kimage *image,
 645					unsigned long page)
 646{
 647	kimage_entry_t *ptr, entry;
 648	unsigned long destination = 0;
 649
 650	for_each_kimage_entry(image, ptr, entry) {
 651		if (entry & IND_DESTINATION)
 652			destination = entry & PAGE_MASK;
 653		else if (entry & IND_SOURCE) {
 654			if (page == destination)
 655				return ptr;
 656			destination += PAGE_SIZE;
 657		}
 658	}
 659
 660	return NULL;
 661}
 662
 663static struct page *kimage_alloc_page(struct kimage *image,
 664					gfp_t gfp_mask,
 665					unsigned long destination)
 666{
 667	/*
 668	 * Here we implement safeguards to ensure that a source page
 669	 * is not copied to its destination page before the data on
 670	 * the destination page is no longer useful.
 671	 *
 672	 * To do this we maintain the invariant that a source page is
 673	 * either its own destination page, or it is not a
 674	 * destination page at all.
 675	 *
 676	 * That is slightly stronger than required, but the proof
 677	 * that no problems will not occur is trivial, and the
 678	 * implementation is simply to verify.
 679	 *
 680	 * When allocating all pages normally this algorithm will run
 681	 * in O(N) time, but in the worst case it will run in O(N^2)
 682	 * time.   If the runtime is a problem the data structures can
 683	 * be fixed.
 684	 */
 685	struct page *page;
 686	unsigned long addr;
 687
 688	/*
 689	 * Walk through the list of destination pages, and see if I
 690	 * have a match.
 691	 */
 692	list_for_each_entry(page, &image->dest_pages, lru) {
 693		addr = page_to_boot_pfn(page) << PAGE_SHIFT;
 694		if (addr == destination) {
 695			list_del(&page->lru);
 696			return page;
 697		}
 698	}
 699	page = NULL;
 700	while (1) {
 701		kimage_entry_t *old;
 702
 703		/* Allocate a page, if we run out of memory give up */
 704		page = kimage_alloc_pages(gfp_mask, 0);
 705		if (!page)
 706			return NULL;
 707		/* If the page cannot be used file it away */
 708		if (page_to_boot_pfn(page) >
 709				(KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
 710			list_add(&page->lru, &image->unusable_pages);
 711			continue;
 712		}
 713		addr = page_to_boot_pfn(page) << PAGE_SHIFT;
 714
 715		/* If it is the destination page we want use it */
 716		if (addr == destination)
 717			break;
 718
 719		/* If the page is not a destination page use it */
 720		if (!kimage_is_destination_range(image, addr,
 721						  addr + PAGE_SIZE - 1))
 722			break;
 723
 724		/*
 725		 * I know that the page is someones destination page.
 726		 * See if there is already a source page for this
 727		 * destination page.  And if so swap the source pages.
 728		 */
 729		old = kimage_dst_used(image, addr);
 730		if (old) {
 731			/* If so move it */
 732			unsigned long old_addr;
 733			struct page *old_page;
 734
 735			old_addr = *old & PAGE_MASK;
 736			old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
 737			copy_highpage(page, old_page);
 738			*old = addr | (*old & ~PAGE_MASK);
 739
 740			/* The old page I have found cannot be a
 741			 * destination page, so return it if it's
 742			 * gfp_flags honor the ones passed in.
 743			 */
 744			if (!(gfp_mask & __GFP_HIGHMEM) &&
 745			    PageHighMem(old_page)) {
 746				kimage_free_pages(old_page);
 747				continue;
 748			}
 749			page = old_page;
 750			break;
 751		}
 752		/* Place the page on the destination list, to be used later */
 753		list_add(&page->lru, &image->dest_pages);
 754	}
 755
 756	return page;
 757}
 758
 759static int kimage_load_normal_segment(struct kimage *image,
 760					 struct kexec_segment *segment)
 761{
 762	unsigned long maddr;
 763	size_t ubytes, mbytes;
 764	int result;
 765	unsigned char __user *buf = NULL;
 766	unsigned char *kbuf = NULL;
 767
 768	if (image->file_mode)
 769		kbuf = segment->kbuf;
 770	else
 771		buf = segment->buf;
 772	ubytes = segment->bufsz;
 773	mbytes = segment->memsz;
 774	maddr = segment->mem;
 775
 776	result = kimage_set_destination(image, maddr);
 777	if (result < 0)
 778		goto out;
 779
 780	while (mbytes) {
 781		struct page *page;
 782		char *ptr;
 783		size_t uchunk, mchunk;
 784
 785		page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
 786		if (!page) {
 787			result  = -ENOMEM;
 788			goto out;
 789		}
 790		result = kimage_add_page(image, page_to_boot_pfn(page)
 791								<< PAGE_SHIFT);
 792		if (result < 0)
 793			goto out;
 794
 795		ptr = kmap_local_page(page);
 796		/* Start with a clear page */
 797		clear_page(ptr);
 798		ptr += maddr & ~PAGE_MASK;
 799		mchunk = min_t(size_t, mbytes,
 800				PAGE_SIZE - (maddr & ~PAGE_MASK));
 801		uchunk = min(ubytes, mchunk);
 802
 803		/* For file based kexec, source pages are in kernel memory */
 804		if (image->file_mode)
 805			memcpy(ptr, kbuf, uchunk);
 806		else
 807			result = copy_from_user(ptr, buf, uchunk);
 808		kunmap_local(ptr);
 809		if (result) {
 810			result = -EFAULT;
 811			goto out;
 812		}
 813		ubytes -= uchunk;
 814		maddr  += mchunk;
 815		if (image->file_mode)
 816			kbuf += mchunk;
 817		else
 818			buf += mchunk;
 819		mbytes -= mchunk;
 820
 821		cond_resched();
 822	}
 823out:
 824	return result;
 825}
 826
 827static int kimage_load_crash_segment(struct kimage *image,
 828					struct kexec_segment *segment)
 829{
 830	/* For crash dumps kernels we simply copy the data from
 831	 * user space to it's destination.
 832	 * We do things a page at a time for the sake of kmap.
 833	 */
 834	unsigned long maddr;
 835	size_t ubytes, mbytes;
 836	int result;
 837	unsigned char __user *buf = NULL;
 838	unsigned char *kbuf = NULL;
 839
 840	result = 0;
 841	if (image->file_mode)
 842		kbuf = segment->kbuf;
 843	else
 844		buf = segment->buf;
 845	ubytes = segment->bufsz;
 846	mbytes = segment->memsz;
 847	maddr = segment->mem;
 848	while (mbytes) {
 849		struct page *page;
 850		char *ptr;
 851		size_t uchunk, mchunk;
 852
 853		page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
 854		if (!page) {
 855			result  = -ENOMEM;
 856			goto out;
 857		}
 858		arch_kexec_post_alloc_pages(page_address(page), 1, 0);
 859		ptr = kmap_local_page(page);
 860		ptr += maddr & ~PAGE_MASK;
 861		mchunk = min_t(size_t, mbytes,
 862				PAGE_SIZE - (maddr & ~PAGE_MASK));
 863		uchunk = min(ubytes, mchunk);
 864		if (mchunk > uchunk) {
 865			/* Zero the trailing part of the page */
 866			memset(ptr + uchunk, 0, mchunk - uchunk);
 867		}
 868
 869		/* For file based kexec, source pages are in kernel memory */
 870		if (image->file_mode)
 871			memcpy(ptr, kbuf, uchunk);
 872		else
 873			result = copy_from_user(ptr, buf, uchunk);
 874		kexec_flush_icache_page(page);
 875		kunmap_local(ptr);
 876		arch_kexec_pre_free_pages(page_address(page), 1);
 877		if (result) {
 878			result = -EFAULT;
 879			goto out;
 880		}
 881		ubytes -= uchunk;
 882		maddr  += mchunk;
 883		if (image->file_mode)
 884			kbuf += mchunk;
 885		else
 886			buf += mchunk;
 887		mbytes -= mchunk;
 888
 889		cond_resched();
 890	}
 891out:
 892	return result;
 893}
 894
 895int kimage_load_segment(struct kimage *image,
 896				struct kexec_segment *segment)
 897{
 898	int result = -ENOMEM;
 899
 900	switch (image->type) {
 901	case KEXEC_TYPE_DEFAULT:
 902		result = kimage_load_normal_segment(image, segment);
 903		break;
 904	case KEXEC_TYPE_CRASH:
 905		result = kimage_load_crash_segment(image, segment);
 906		break;
 907	}
 908
 909	return result;
 910}
 911
 912struct kexec_load_limit {
 913	/* Mutex protects the limit count. */
 914	struct mutex mutex;
 915	int limit;
 916};
 917
 918static struct kexec_load_limit load_limit_reboot = {
 919	.mutex = __MUTEX_INITIALIZER(load_limit_reboot.mutex),
 920	.limit = -1,
 921};
 922
 923static struct kexec_load_limit load_limit_panic = {
 924	.mutex = __MUTEX_INITIALIZER(load_limit_panic.mutex),
 925	.limit = -1,
 926};
 927
 928struct kimage *kexec_image;
 929struct kimage *kexec_crash_image;
 930static int kexec_load_disabled;
 931
 932#ifdef CONFIG_SYSCTL
 933static int kexec_limit_handler(struct ctl_table *table, int write,
 934			       void *buffer, size_t *lenp, loff_t *ppos)
 935{
 936	struct kexec_load_limit *limit = table->data;
 937	int val;
 938	struct ctl_table tmp = {
 939		.data = &val,
 940		.maxlen = sizeof(val),
 941		.mode = table->mode,
 942	};
 943	int ret;
 944
 945	if (write) {
 946		ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
 947		if (ret)
 948			return ret;
 949
 950		if (val < 0)
 951			return -EINVAL;
 952
 953		mutex_lock(&limit->mutex);
 954		if (limit->limit != -1 && val >= limit->limit)
 955			ret = -EINVAL;
 956		else
 957			limit->limit = val;
 958		mutex_unlock(&limit->mutex);
 959
 960		return ret;
 961	}
 962
 963	mutex_lock(&limit->mutex);
 964	val = limit->limit;
 965	mutex_unlock(&limit->mutex);
 966
 967	return proc_dointvec(&tmp, write, buffer, lenp, ppos);
 968}
 969
 970static struct ctl_table kexec_core_sysctls[] = {
 971	{
 972		.procname	= "kexec_load_disabled",
 973		.data		= &kexec_load_disabled,
 974		.maxlen		= sizeof(int),
 975		.mode		= 0644,
 976		/* only handle a transition from default "0" to "1" */
 977		.proc_handler	= proc_dointvec_minmax,
 978		.extra1		= SYSCTL_ONE,
 979		.extra2		= SYSCTL_ONE,
 980	},
 981	{
 982		.procname	= "kexec_load_limit_panic",
 983		.data		= &load_limit_panic,
 984		.mode		= 0644,
 985		.proc_handler	= kexec_limit_handler,
 986	},
 987	{
 988		.procname	= "kexec_load_limit_reboot",
 989		.data		= &load_limit_reboot,
 990		.mode		= 0644,
 991		.proc_handler	= kexec_limit_handler,
 992	},
 993	{ }
 994};
 995
 996static int __init kexec_core_sysctl_init(void)
 997{
 998	register_sysctl_init("kernel", kexec_core_sysctls);
 999	return 0;
1000}
1001late_initcall(kexec_core_sysctl_init);
1002#endif
1003
1004bool kexec_load_permitted(int kexec_image_type)
1005{
1006	struct kexec_load_limit *limit;
1007
1008	/*
1009	 * Only the superuser can use the kexec syscall and if it has not
1010	 * been disabled.
1011	 */
1012	if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
1013		return false;
1014
1015	/* Check limit counter and decrease it.*/
1016	limit = (kexec_image_type == KEXEC_TYPE_CRASH) ?
1017		&load_limit_panic : &load_limit_reboot;
1018	mutex_lock(&limit->mutex);
1019	if (!limit->limit) {
1020		mutex_unlock(&limit->mutex);
1021		return false;
1022	}
1023	if (limit->limit != -1)
1024		limit->limit--;
1025	mutex_unlock(&limit->mutex);
1026
1027	return true;
1028}
1029
1030/*
1031 * No panic_cpu check version of crash_kexec().  This function is called
1032 * only when panic_cpu holds the current CPU number; this is the only CPU
1033 * which processes crash_kexec routines.
1034 */
1035void __noclone __crash_kexec(struct pt_regs *regs)
1036{
1037	/* Take the kexec_lock here to prevent sys_kexec_load
1038	 * running on one cpu from replacing the crash kernel
1039	 * we are using after a panic on a different cpu.
1040	 *
1041	 * If the crash kernel was not located in a fixed area
1042	 * of memory the xchg(&kexec_crash_image) would be
1043	 * sufficient.  But since I reuse the memory...
1044	 */
1045	if (kexec_trylock()) {
1046		if (kexec_crash_image) {
1047			struct pt_regs fixed_regs;
1048
1049			crash_setup_regs(&fixed_regs, regs);
1050			crash_save_vmcoreinfo();
1051			machine_crash_shutdown(&fixed_regs);
1052			machine_kexec(kexec_crash_image);
1053		}
1054		kexec_unlock();
1055	}
1056}
1057STACK_FRAME_NON_STANDARD(__crash_kexec);
1058
1059__bpf_kfunc void crash_kexec(struct pt_regs *regs)
1060{
1061	int old_cpu, this_cpu;
1062
1063	/*
1064	 * Only one CPU is allowed to execute the crash_kexec() code as with
1065	 * panic().  Otherwise parallel calls of panic() and crash_kexec()
1066	 * may stop each other.  To exclude them, we use panic_cpu here too.
1067	 */
1068	old_cpu = PANIC_CPU_INVALID;
1069	this_cpu = raw_smp_processor_id();
1070
1071	if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
1072		/* This is the 1st CPU which comes here, so go ahead. */
1073		__crash_kexec(regs);
1074
1075		/*
1076		 * Reset panic_cpu to allow another panic()/crash_kexec()
1077		 * call.
1078		 */
1079		atomic_set(&panic_cpu, PANIC_CPU_INVALID);
1080	}
1081}
1082
1083static inline resource_size_t crash_resource_size(const struct resource *res)
1084{
1085	return !res->end ? 0 : resource_size(res);
1086}
1087
1088ssize_t crash_get_memory_size(void)
1089{
1090	ssize_t size = 0;
1091
1092	if (!kexec_trylock())
1093		return -EBUSY;
1094
1095	size += crash_resource_size(&crashk_res);
1096	size += crash_resource_size(&crashk_low_res);
1097
1098	kexec_unlock();
1099	return size;
1100}
1101
1102static int __crash_shrink_memory(struct resource *old_res,
1103				 unsigned long new_size)
1104{
1105	struct resource *ram_res;
1106
1107	ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1108	if (!ram_res)
1109		return -ENOMEM;
1110
1111	ram_res->start = old_res->start + new_size;
1112	ram_res->end   = old_res->end;
1113	ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
1114	ram_res->name  = "System RAM";
1115
1116	if (!new_size) {
1117		release_resource(old_res);
1118		old_res->start = 0;
1119		old_res->end   = 0;
1120	} else {
1121		crashk_res.end = ram_res->start - 1;
1122	}
1123
1124	crash_free_reserved_phys_range(ram_res->start, ram_res->end);
1125	insert_resource(&iomem_resource, ram_res);
1126
1127	return 0;
1128}
1129
1130int crash_shrink_memory(unsigned long new_size)
1131{
1132	int ret = 0;
1133	unsigned long old_size, low_size;
 
 
1134
1135	if (!kexec_trylock())
1136		return -EBUSY;
1137
1138	if (kexec_crash_image) {
1139		ret = -ENOENT;
1140		goto unlock;
1141	}
1142
1143	low_size = crash_resource_size(&crashk_low_res);
1144	old_size = crash_resource_size(&crashk_res) + low_size;
1145	new_size = roundup(new_size, KEXEC_CRASH_MEM_ALIGN);
1146	if (new_size >= old_size) {
1147		ret = (new_size == old_size) ? 0 : -EINVAL;
1148		goto unlock;
1149	}
1150
1151	/*
1152	 * (low_size > new_size) implies that low_size is greater than zero.
1153	 * This also means that if low_size is zero, the else branch is taken.
1154	 *
1155	 * If low_size is greater than 0, (low_size > new_size) indicates that
1156	 * crashk_low_res also needs to be shrunken. Otherwise, only crashk_res
1157	 * needs to be shrunken.
1158	 */
1159	if (low_size > new_size) {
1160		ret = __crash_shrink_memory(&crashk_res, 0);
1161		if (ret)
1162			goto unlock;
1163
1164		ret = __crash_shrink_memory(&crashk_low_res, new_size);
1165	} else {
1166		ret = __crash_shrink_memory(&crashk_res, new_size - low_size);
1167	}
1168
1169	/* Swap crashk_res and crashk_low_res if needed */
1170	if (!crashk_res.end && crashk_low_res.end) {
1171		crashk_res.start = crashk_low_res.start;
1172		crashk_res.end   = crashk_low_res.end;
1173		release_resource(&crashk_low_res);
1174		crashk_low_res.start = 0;
1175		crashk_low_res.end   = 0;
1176		insert_resource(&iomem_resource, &crashk_res);
1177	}
1178
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1179unlock:
1180	kexec_unlock();
1181	return ret;
1182}
1183
1184void crash_save_cpu(struct pt_regs *regs, int cpu)
1185{
1186	struct elf_prstatus prstatus;
1187	u32 *buf;
1188
1189	if ((cpu < 0) || (cpu >= nr_cpu_ids))
1190		return;
1191
1192	/* Using ELF notes here is opportunistic.
1193	 * I need a well defined structure format
1194	 * for the data I pass, and I need tags
1195	 * on the data to indicate what information I have
1196	 * squirrelled away.  ELF notes happen to provide
1197	 * all of that, so there is no need to invent something new.
1198	 */
1199	buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1200	if (!buf)
1201		return;
1202	memset(&prstatus, 0, sizeof(prstatus));
1203	prstatus.common.pr_pid = current->pid;
1204	elf_core_copy_regs(&prstatus.pr_reg, regs);
1205	buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1206			      &prstatus, sizeof(prstatus));
1207	final_note(buf);
1208}
1209
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1210/*
1211 * Move into place and start executing a preloaded standalone
1212 * executable.  If nothing was preloaded return an error.
1213 */
1214int kernel_kexec(void)
1215{
1216	int error = 0;
1217
1218	if (!kexec_trylock())
1219		return -EBUSY;
1220	if (!kexec_image) {
1221		error = -EINVAL;
1222		goto Unlock;
1223	}
1224
1225#ifdef CONFIG_KEXEC_JUMP
1226	if (kexec_image->preserve_context) {
1227		pm_prepare_console();
1228		error = freeze_processes();
1229		if (error) {
1230			error = -EBUSY;
1231			goto Restore_console;
1232		}
1233		suspend_console();
1234		error = dpm_suspend_start(PMSG_FREEZE);
1235		if (error)
1236			goto Resume_console;
1237		/* At this point, dpm_suspend_start() has been called,
1238		 * but *not* dpm_suspend_end(). We *must* call
1239		 * dpm_suspend_end() now.  Otherwise, drivers for
1240		 * some devices (e.g. interrupt controllers) become
1241		 * desynchronized with the actual state of the
1242		 * hardware at resume time, and evil weirdness ensues.
1243		 */
1244		error = dpm_suspend_end(PMSG_FREEZE);
1245		if (error)
1246			goto Resume_devices;
1247		error = suspend_disable_secondary_cpus();
1248		if (error)
1249			goto Enable_cpus;
1250		local_irq_disable();
1251		error = syscore_suspend();
1252		if (error)
1253			goto Enable_irqs;
1254	} else
1255#endif
1256	{
1257		kexec_in_progress = true;
1258		kernel_restart_prepare("kexec reboot");
1259		migrate_to_reboot_cpu();
1260		syscore_shutdown();
1261
1262		/*
1263		 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1264		 * no further code needs to use CPU hotplug (which is true in
1265		 * the reboot case). However, the kexec path depends on using
1266		 * CPU hotplug again; so re-enable it here.
1267		 */
1268		cpu_hotplug_enable();
1269		pr_notice("Starting new kernel\n");
1270		machine_shutdown();
1271	}
1272
1273	kmsg_dump(KMSG_DUMP_SHUTDOWN);
1274	machine_kexec(kexec_image);
1275
1276#ifdef CONFIG_KEXEC_JUMP
1277	if (kexec_image->preserve_context) {
1278		syscore_resume();
1279 Enable_irqs:
1280		local_irq_enable();
1281 Enable_cpus:
1282		suspend_enable_secondary_cpus();
1283		dpm_resume_start(PMSG_RESTORE);
1284 Resume_devices:
1285		dpm_resume_end(PMSG_RESTORE);
1286 Resume_console:
1287		resume_console();
1288		thaw_processes();
1289 Restore_console:
1290		pm_restore_console();
1291	}
1292#endif
1293
1294 Unlock:
1295	kexec_unlock();
1296	return error;
1297}