kernel/power/snapshot.c

   1 /*
   2  * linux/kernel/power/snapshot.c
   3  *
   4  * This file provide system snapshot/restore functionality.
   5  *
   6  * Copyright (C) 1998-2005 Pavel Machek <pavel@suse.cz>
   7  *
   8  * This file is released under the GPLv2, and is based on swsusp.c.
   9  *
  10  */
  11
  12
  13 #include <linux/version.h>
  14 #include <linux/module.h>
  15 #include <linux/mm.h>
  16 #include <linux/suspend.h>
  17 #include <linux/smp_lock.h>
  18 #include <linux/delay.h>
  19 #include <linux/bitops.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/kernel.h>
  22 #include <linux/pm.h>
  23 #include <linux/device.h>
  24 #include <linux/bootmem.h>
  25 #include <linux/syscalls.h>
  26 #include <linux/console.h>
  27 #include <linux/highmem.h>
  28
  29 #include <asm/uaccess.h>
  30 #include <asm/mmu_context.h>
  31 #include <asm/pgtable.h>
  32 #include <asm/tlbflush.h>
  33 #include <asm/io.h>
  34
  35 #include "power.h"
  36
  37 struct pbe *pagedir_nosave;
  38 static unsigned int nr_copy_pages;
  39 static unsigned int nr_meta_pages;
  40 static unsigned long *buffer;
  41
  42 #ifdef CONFIG_HIGHMEM
  43 unsigned int count_highmem_pages(void)
  44 {
  45         struct zone *zone;
  46         unsigned long zone_pfn;
  47         unsigned int n = 0;
  48
  49         for_each_zone (zone)
  50                 if (is_highmem(zone)) {
  51                         mark_free_pages(zone);
  52                         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; zone_pfn++) {
  53                                 struct page *page;
  54                                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
  55                                 if (!pfn_valid(pfn))
  56                                         continue;
  57                                 page = pfn_to_page(pfn);
  58                                 if (PageReserved(page))
  59                                         continue;
  60                                 if (PageNosaveFree(page))
  61                                         continue;
  62                                 n++;
  63                         }
  64                 }
  65         return n;
  66 }
  67
  68 struct highmem_page {
  69         char *data;
  70         struct page *page;
  71         struct highmem_page *next;
  72 };
  73
  74 static struct highmem_page *highmem_copy;
  75
  76 static int save_highmem_zone(struct zone *zone)
  77 {
  78         unsigned long zone_pfn;
  79         mark_free_pages(zone);
  80         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
  81                 struct page *page;
  82                 struct highmem_page *save;
  83                 void *kaddr;
  84                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
  85
  86                 if (!(pfn%10000))
  87                         printk(".");
  88                 if (!pfn_valid(pfn))
  89                         continue;
  90                 page = pfn_to_page(pfn);
  91                 /*
  92                  * This condition results from rvmalloc() sans vmalloc_32()
  93                  * and architectural memory reservations. This should be
  94                  * corrected eventually when the cases giving rise to this
  95                  * are better understood.
  96                  */
  97                 if (PageReserved(page))
  98                         continue;
  99                 BUG_ON(PageNosave(page));
 100                 if (PageNosaveFree(page))
 101                         continue;
 102                 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
 103                 if (!save)
 104                         return -ENOMEM;
 105                 save->next = highmem_copy;
 106                 save->page = page;
 107                 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
 108                 if (!save->data) {
 109                         kfree(save);
 110                         return -ENOMEM;
 111                 }
 112                 kaddr = kmap_atomic(page, KM_USER0);
 113                 memcpy(save->data, kaddr, PAGE_SIZE);
 114                 kunmap_atomic(kaddr, KM_USER0);
 115                 highmem_copy = save;
 116         }
 117         return 0;
 118 }
 119
 120 int save_highmem(void)
 121 {
 122         struct zone *zone;
 123         int res = 0;
 124
 125         pr_debug("swsusp: Saving Highmem");
 126         drain_local_pages();
 127         for_each_zone (zone) {
 128                 if (is_highmem(zone))
 129                         res = save_highmem_zone(zone);
 130                 if (res)
 131                         return res;
 132         }
 133         printk("\n");
 134         return 0;
 135 }
 136
 137 int restore_highmem(void)
 138 {
 139         printk("swsusp: Restoring Highmem\n");
 140         while (highmem_copy) {
 141                 struct highmem_page *save = highmem_copy;
 142                 void *kaddr;
 143                 highmem_copy = save->next;
 144
 145                 kaddr = kmap_atomic(save->page, KM_USER0);
 146                 memcpy(kaddr, save->data, PAGE_SIZE);
 147                 kunmap_atomic(kaddr, KM_USER0);
 148                 free_page((long) save->data);
 149                 kfree(save);
 150         }
 151         return 0;
 152 }
 153 #endif
 154
 155 static int pfn_is_nosave(unsigned long pfn)
 156 {
 157         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 158         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 159         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 160 }
 161
 162 /**
 163  *      saveable - Determine whether a page should be cloned or not.
 164  *      @pfn:   The page
 165  *
 166  *      We save a page if it's Reserved, and not in the range of pages
 167  *      statically defined as 'unsaveable', or if it isn't reserved, and
 168  *      isn't part of a free chunk of pages.
 169  */
 170
 171 static int saveable(struct zone *zone, unsigned long *zone_pfn)
 172 {
 173         unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
 174         struct page *page;
 175
 176         if (!pfn_valid(pfn))
 177                 return 0;
 178
 179         page = pfn_to_page(pfn);
 180         BUG_ON(PageReserved(page) && PageNosave(page));
 181         if (PageNosave(page))
 182                 return 0;
 183         if (PageReserved(page) && pfn_is_nosave(pfn))
 184                 return 0;
 185         if (PageNosaveFree(page))
 186                 return 0;
 187
 188         return 1;
 189 }
 190
 191 unsigned int count_data_pages(void)
 192 {
 193         struct zone *zone;
 194         unsigned long zone_pfn;
 195         unsigned int n = 0;
 196
 197         for_each_zone (zone) {
 198                 if (is_highmem(zone))
 199                         continue;
 200                 mark_free_pages(zone);
 201                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 202                         n += saveable(zone, &zone_pfn);
 203         }
 204         return n;
 205 }
 206
 207 static void copy_data_pages(struct pbe *pblist)
 208 {
 209         struct zone *zone;
 210         unsigned long zone_pfn;
 211         struct pbe *pbe, *p;
 212
 213         pbe = pblist;
 214         for_each_zone (zone) {
 215                 if (is_highmem(zone))
 216                         continue;
 217                 mark_free_pages(zone);
 218                 /* This is necessary for swsusp_free() */
 219                 for_each_pb_page (p, pblist)
 220                         SetPageNosaveFree(virt_to_page(p));
 221                 for_each_pbe (p, pblist)
 222                         SetPageNosaveFree(virt_to_page(p->address));
 223                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 224                         if (saveable(zone, &zone_pfn)) {
 225                                 struct page *page;
 226                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 227                                 BUG_ON(!pbe);
 228                                 pbe->orig_address = (unsigned long)page_address(page);
 229                                 /* copy_page is not usable for copying task structs. */
 230                                 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
 231                                 pbe = pbe->next;
 232                         }
 233                 }
 234         }
 235         BUG_ON(pbe);
 236 }
 237
 238
 239 /**
 240  *      free_pagedir - free pages allocated with alloc_pagedir()
 241  */
 242
 243 static void free_pagedir(struct pbe *pblist, int clear_nosave_free)
 244 {
 245         struct pbe *pbe;
 246
 247         while (pblist) {
 248                 pbe = (pblist + PB_PAGE_SKIP)->next;
 249                 ClearPageNosave(virt_to_page(pblist));
 250                 if (clear_nosave_free)
 251                         ClearPageNosaveFree(virt_to_page(pblist));
 252                 free_page((unsigned long)pblist);
 253                 pblist = pbe;
 254         }
 255 }
 256
 257 /**
 258  *      fill_pb_page - Create a list of PBEs on a given memory page
 259  */
 260
 261 static inline void fill_pb_page(struct pbe *pbpage)
 262 {
 263         struct pbe *p;
 264
 265         p = pbpage;
 266         pbpage += PB_PAGE_SKIP;
 267         do
 268                 p->next = p + 1;
 269         while (++p < pbpage);
 270 }
 271
 272 /**
 273  *      create_pbe_list - Create a list of PBEs on top of a given chain
 274  *      of memory pages allocated with alloc_pagedir()
 275  */
 276
 277 static inline void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
 278 {
 279         struct pbe *pbpage, *p;
 280         unsigned int num = PBES_PER_PAGE;
 281
 282         for_each_pb_page (pbpage, pblist) {
 283                 if (num >= nr_pages)
 284                         break;
 285
 286                 fill_pb_page(pbpage);
 287                 num += PBES_PER_PAGE;
 288         }
 289         if (pbpage) {
 290                 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
 291                         p->next = p + 1;
 292                 p->next = NULL;
 293         }
 294 }
 295
 296 /**
 297  *      On resume it is necessary to trace and eventually free the unsafe
 298  *      pages that have been allocated, because they are needed for I/O
 299  *      (on x86-64 we likely will "eat" these pages once again while
 300  *      creating the temporary page translation tables)
 301  */
 302
 303 struct eaten_page {
 304         struct eaten_page *next;
 305         char padding[PAGE_SIZE - sizeof(void *)];
 306 };
 307
 308 static struct eaten_page *eaten_pages = NULL;
 309
 310 static void release_eaten_pages(void)
 311 {
 312         struct eaten_page *p, *q;
 313
 314         p = eaten_pages;
 315         while (p) {
 316                 q = p->next;
 317                 /* We don't want swsusp_free() to free this page again */
 318                 ClearPageNosave(virt_to_page(p));
 319                 free_page((unsigned long)p);
 320                 p = q;
 321         }
 322         eaten_pages = NULL;
 323 }
 324
 325 /**
 326  *      @safe_needed - on resume, for storing the PBE list and the image,
 327  *      we can only use memory pages that do not conflict with the pages
 328  *      which had been used before suspend.
 329  *
 330  *      The unsafe pages are marked with the PG_nosave_free flag
 331  *
 332  *      Allocated but unusable (ie eaten) memory pages should be marked
 333  *      so that swsusp_free() can release them
 334  */
 335
 336 static inline void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
 337 {
 338         void *res;
 339
 340         if (safe_needed)
 341                 do {
 342                         res = (void *)get_zeroed_page(gfp_mask);
 343                         if (res && PageNosaveFree(virt_to_page(res))) {
 344                                 /* This is for swsusp_free() */
 345                                 SetPageNosave(virt_to_page(res));
 346                                 ((struct eaten_page *)res)->next = eaten_pages;
 347                                 eaten_pages = res;
 348                         }
 349                 } while (res && PageNosaveFree(virt_to_page(res)));
 350         else
 351                 res = (void *)get_zeroed_page(gfp_mask);
 352         if (res) {
 353                 SetPageNosave(virt_to_page(res));
 354                 SetPageNosaveFree(virt_to_page(res));
 355         }
 356         return res;
 357 }
 358
 359 unsigned long get_safe_page(gfp_t gfp_mask)
 360 {
 361         return (unsigned long)alloc_image_page(gfp_mask, 1);
 362 }
 363
 364 /**
 365  *      alloc_pagedir - Allocate the page directory.
 366  *
 367  *      First, determine exactly how many pages we need and
 368  *      allocate them.
 369  *
 370  *      We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
 371  *      struct pbe elements (pbes) and the last element in the page points
 372  *      to the next page.
 373  *
 374  *      On each page we set up a list of struct_pbe elements.
 375  */
 376
 377 struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask, int safe_needed)
 378 {
 379         unsigned int num;
 380         struct pbe *pblist, *pbe;
 381
 382         if (!nr_pages)
 383                 return NULL;
 384
 385         pblist = alloc_image_page(gfp_mask, safe_needed);
 386         /* FIXME: rewrite this ugly loop */
 387         for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
 388                         pbe = pbe->next, num += PBES_PER_PAGE) {
 389                 pbe += PB_PAGE_SKIP;
 390                 pbe->next = alloc_image_page(gfp_mask, safe_needed);
 391         }
 392         if (!pbe) { /* get_zeroed_page() failed */
 393                 free_pagedir(pblist, 1);
 394                 pblist = NULL;
 395         } else
 396                 create_pbe_list(pblist, nr_pages);
 397         return pblist;
 398 }
 399
 400 /**
 401  * Free pages we allocated for suspend. Suspend pages are alocated
 402  * before atomic copy, so we need to free them after resume.
 403  */
 404
 405 void swsusp_free(void)
 406 {
 407         struct zone *zone;
 408         unsigned long zone_pfn;
 409
 410         for_each_zone(zone) {
 411                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 412                         if (pfn_valid(zone_pfn + zone->zone_start_pfn)) {
 413                                 struct page *page;
 414                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 415                                 if (PageNosave(page) && PageNosaveFree(page)) {
 416                                         ClearPageNosave(page);
 417                                         ClearPageNosaveFree(page);
 418                                         free_page((long) page_address(page));
 419                                 }
 420                         }
 421         }
 422         nr_copy_pages = 0;
 423         nr_meta_pages = 0;
 424         pagedir_nosave = NULL;
 425         buffer = NULL;
 426 }
 427
 428
 429 /**
 430  *      enough_free_mem - Make sure we enough free memory to snapshot.
 431  *
 432  *      Returns TRUE or FALSE after checking the number of available
 433  *      free pages.
 434  */
 435
 436 static int enough_free_mem(unsigned int nr_pages)
 437 {
 438         struct zone *zone;
 439         unsigned int n = 0;
 440
 441         for_each_zone (zone)
 442                 if (!is_highmem(zone))
 443                         n += zone->free_pages;
 444         pr_debug("swsusp: available memory: %u pages\n", n);
 445         return n > (nr_pages + PAGES_FOR_IO +
 446                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 447 }
 448
 449 static int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
 450 {
 451         struct pbe *p;
 452
 453         for_each_pbe (p, pblist) {
 454                 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
 455                 if (!p->address)
 456                         return -ENOMEM;
 457         }
 458         return 0;
 459 }
 460
 461 static struct pbe *swsusp_alloc(unsigned int nr_pages)
 462 {
 463         struct pbe *pblist;
 464
 465         if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
 466                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 467                 return NULL;
 468         }
 469
 470         if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
 471                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 472                 swsusp_free();
 473                 return NULL;
 474         }
 475
 476         return pblist;
 477 }
 478
 479 asmlinkage int swsusp_save(void)
 480 {
 481         unsigned int nr_pages;
 482
 483         pr_debug("swsusp: critical section: \n");
 484
 485         drain_local_pages();
 486         nr_pages = count_data_pages();
 487         printk("swsusp: Need to copy %u pages\n", nr_pages);
 488
 489         pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
 490                  nr_pages,
 491                  (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
 492                  PAGES_FOR_IO, nr_free_pages());
 493
 494         if (!enough_free_mem(nr_pages)) {
 495                 printk(KERN_ERR "swsusp: Not enough free memory\n");
 496                 return -ENOMEM;
 497         }
 498
 499         pagedir_nosave = swsusp_alloc(nr_pages);
 500         if (!pagedir_nosave)
 501                 return -ENOMEM;
 502
 503         /* During allocating of suspend pagedir, new cold pages may appear.
 504          * Kill them.
 505          */
 506         drain_local_pages();
 507         copy_data_pages(pagedir_nosave);
 508
 509         /*
 510          * End of critical section. From now on, we can write to memory,
 511          * but we should not touch disk. This specially means we must _not_
 512          * touch swap space! Except we must write out our image of course.
 513          */
 514
 515         nr_copy_pages = nr_pages;
 516         nr_meta_pages = (nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 517
 518         printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
 519         return 0;
 520 }
 521
 522 static void init_header(struct swsusp_info *info)
 523 {
 524         memset(info, 0, sizeof(struct swsusp_info));
 525         info->version_code = LINUX_VERSION_CODE;
 526         info->num_physpages = num_physpages;
 527         memcpy(&info->uts, &system_utsname, sizeof(system_utsname));
 528         info->cpus = num_online_cpus();
 529         info->image_pages = nr_copy_pages;
 530         info->pages = nr_copy_pages + nr_meta_pages + 1;
 531         info->size = info->pages;
 532         info->size <<= PAGE_SHIFT;
 533 }
 534
 535 /**
 536  *      pack_orig_addresses - the .orig_address fields of the PBEs from the
 537  *      list starting at @pbe are stored in the array @buf[] (1 page)
 538  */
 539
 540 static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pbe)
 541 {
 542         int j;
 543
 544         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 545                 buf[j] = pbe->orig_address;
 546                 pbe = pbe->next;
 547         }
 548         if (!pbe)
 549                 for (; j < PAGE_SIZE / sizeof(long); j++)
 550                         buf[j] = 0;
 551         return pbe;
 552 }
 553
 554 /**
 555  *      snapshot_read_next - used for reading the system memory snapshot.
 556  *
 557  *      On the first call to it @handle should point to a zeroed
 558  *      snapshot_handle structure.  The structure gets updated and a pointer
 559  *      to it should be passed to this function every next time.
 560  *
 561  *      The @count parameter should contain the number of bytes the caller
 562  *      wants to read from the snapshot.  It must not be zero.
 563  *
 564  *      On success the function returns a positive number.  Then, the caller
 565  *      is allowed to read up to the returned number of bytes from the memory
 566  *      location computed by the data_of() macro.  The number returned
 567  *      may be smaller than @count, but this only happens if the read would
 568  *      cross a page boundary otherwise.
 569  *
 570  *      The function returns 0 to indicate the end of data stream condition,
 571  *      and a negative number is returned on error.  In such cases the
 572  *      structure pointed to by @handle is not updated and should not be used
 573  *      any more.
 574  */
 575
 576 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
 577 {
 578         if (handle->page > nr_meta_pages + nr_copy_pages)
 579                 return 0;
 580         if (!buffer) {
 581                 /* This makes the buffer be freed by swsusp_free() */
 582                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 583                 if (!buffer)
 584                         return -ENOMEM;
 585         }
 586         if (!handle->offset) {
 587                 init_header((struct swsusp_info *)buffer);
 588                 handle->buffer = buffer;
 589                 handle->pbe = pagedir_nosave;
 590         }
 591         if (handle->prev < handle->page) {
 592                 if (handle->page <= nr_meta_pages) {
 593                         handle->pbe = pack_orig_addresses(buffer, handle->pbe);
 594                         if (!handle->pbe)
 595                                 handle->pbe = pagedir_nosave;
 596                 } else {
 597                         handle->buffer = (void *)handle->pbe->address;
 598                         handle->pbe = handle->pbe->next;
 599                 }
 600                 handle->prev = handle->page;
 601         }
 602         handle->buf_offset = handle->page_offset;
 603         if (handle->page_offset + count >= PAGE_SIZE) {
 604                 count = PAGE_SIZE - handle->page_offset;
 605                 handle->page_offset = 0;
 606                 handle->page++;
 607         } else {
 608                 handle->page_offset += count;
 609         }
 610         handle->offset += count;
 611         return count;
 612 }
 613
 614 /**
 615  *      mark_unsafe_pages - mark the pages that cannot be used for storing
 616  *      the image during resume, because they conflict with the pages that
 617  *      had been used before suspend
 618  */
 619
 620 static int mark_unsafe_pages(struct pbe *pblist)
 621 {
 622         struct zone *zone;
 623         unsigned long zone_pfn;
 624         struct pbe *p;
 625
 626         if (!pblist) /* a sanity check */
 627                 return -EINVAL;
 628
 629         /* Clear page flags */
 630         for_each_zone (zone) {
 631                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 632                         if (pfn_valid(zone_pfn + zone->zone_start_pfn))
 633                                 ClearPageNosaveFree(pfn_to_page(zone_pfn +
 634                                         zone->zone_start_pfn));
 635         }
 636
 637         /* Mark orig addresses */
 638         for_each_pbe (p, pblist) {
 639                 if (virt_addr_valid(p->orig_address))
 640                         SetPageNosaveFree(virt_to_page(p->orig_address));
 641                 else
 642                         return -EFAULT;
 643         }
 644
 645         return 0;
 646 }
 647
 648 static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
 649 {
 650         /* We assume both lists contain the same number of elements */
 651         while (src) {
 652                 dst->orig_address = src->orig_address;
 653                 dst = dst->next;
 654                 src = src->next;
 655         }
 656 }
 657
 658 static int check_header(struct swsusp_info *info)
 659 {
 660         char *reason = NULL;
 661
 662         if (info->version_code != LINUX_VERSION_CODE)
 663                 reason = "kernel version";
 664         if (info->num_physpages != num_physpages)
 665                 reason = "memory size";
 666         if (strcmp(info->uts.sysname,system_utsname.sysname))
 667                 reason = "system type";
 668         if (strcmp(info->uts.release,system_utsname.release))
 669                 reason = "kernel release";
 670         if (strcmp(info->uts.version,system_utsname.version))
 671                 reason = "version";
 672         if (strcmp(info->uts.machine,system_utsname.machine))
 673                 reason = "machine";
 674         if (reason) {
 675                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
 676                 return -EPERM;
 677         }
 678         return 0;
 679 }
 680
 681 /**
 682  *      load header - check the image header and copy data from it
 683  */
 684
 685 static int load_header(struct snapshot_handle *handle,
 686                               struct swsusp_info *info)
 687 {
 688         int error;
 689         struct pbe *pblist;
 690
 691         error = check_header(info);
 692         if (!error) {
 693                 pblist = alloc_pagedir(info->image_pages, GFP_ATOMIC, 0);
 694                 if (!pblist)
 695                         return -ENOMEM;
 696                 pagedir_nosave = pblist;
 697                 handle->pbe = pblist;
 698                 nr_copy_pages = info->image_pages;
 699                 nr_meta_pages = info->pages - info->image_pages - 1;
 700         }
 701         return error;
 702 }
 703
 704 /**
 705  *      unpack_orig_addresses - copy the elements of @buf[] (1 page) to
 706  *      the PBEs in the list starting at @pbe
 707  */
 708
 709 static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
 710                                                 struct pbe *pbe)
 711 {
 712         int j;
 713
 714         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 715                 pbe->orig_address = buf[j];
 716                 pbe = pbe->next;
 717         }
 718         return pbe;
 719 }
 720
 721 /**
 722  *      create_image - use metadata contained in the PBE list
 723  *      pointed to by pagedir_nosave to mark the pages that will
 724  *      be overwritten in the process of restoring the system
 725  *      memory state from the image and allocate memory for
 726  *      the image avoiding these pages
 727  */
 728
 729 static int create_image(struct snapshot_handle *handle)
 730 {
 731         int error = 0;
 732         struct pbe *p, *pblist;
 733
 734         p = pagedir_nosave;
 735         error = mark_unsafe_pages(p);
 736         if (!error) {
 737                 pblist = alloc_pagedir(nr_copy_pages, GFP_ATOMIC, 1);
 738                 if (pblist)
 739                         copy_page_backup_list(pblist, p);
 740                 free_pagedir(p, 0);
 741                 if (!pblist)
 742                         error = -ENOMEM;
 743         }
 744         if (!error)
 745                 error = alloc_data_pages(pblist, GFP_ATOMIC, 1);
 746         if (!error) {
 747                 release_eaten_pages();
 748                 pagedir_nosave = pblist;
 749         } else {
 750                 pagedir_nosave = NULL;
 751                 handle->pbe = NULL;
 752                 nr_copy_pages = 0;
 753                 nr_meta_pages = 0;
 754         }
 755         return error;
 756 }
 757
 758 /**
 759  *      snapshot_write_next - used for writing the system memory snapshot.
 760  *
 761  *      On the first call to it @handle should point to a zeroed
 762  *      snapshot_handle structure.  The structure gets updated and a pointer
 763  *      to it should be passed to this function every next time.
 764  *
 765  *      The @count parameter should contain the number of bytes the caller
 766  *      wants to write to the image.  It must not be zero.
 767  *
 768  *      On success the function returns a positive number.  Then, the caller
 769  *      is allowed to write up to the returned number of bytes to the memory
 770  *      location computed by the data_of() macro.  The number returned
 771  *      may be smaller than @count, but this only happens if the write would
 772  *      cross a page boundary otherwise.
 773  *
 774  *      The function returns 0 to indicate the "end of file" condition,
 775  *      and a negative number is returned on error.  In such cases the
 776  *      structure pointed to by @handle is not updated and should not be used
 777  *      any more.
 778  */
 779
 780 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
 781 {
 782         int error = 0;
 783
 784         if (handle->prev && handle->page > nr_meta_pages + nr_copy_pages)
 785                 return 0;
 786         if (!buffer) {
 787                 /* This makes the buffer be freed by swsusp_free() */
 788                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 789                 if (!buffer)
 790                         return -ENOMEM;
 791         }
 792         if (!handle->offset)
 793                 handle->buffer = buffer;
 794         if (handle->prev < handle->page) {
 795                 if (!handle->prev) {
 796                         error = load_header(handle, (struct swsusp_info *)buffer);
 797                         if (error)
 798                                 return error;
 799                 } else if (handle->prev <= nr_meta_pages) {
 800                         handle->pbe = unpack_orig_addresses(buffer, handle->pbe);
 801                         if (!handle->pbe) {
 802                                 error = create_image(handle);
 803                                 if (error)
 804                                         return error;
 805                                 handle->pbe = pagedir_nosave;
 806                                 handle->buffer = (void *)handle->pbe->address;
 807                         }
 808                 } else {
 809                         handle->pbe = handle->pbe->next;
 810                         handle->buffer = (void *)handle->pbe->address;
 811                 }
 812                 handle->prev = handle->page;
 813         }
 814         handle->buf_offset = handle->page_offset;
 815         if (handle->page_offset + count >= PAGE_SIZE) {
 816                 count = PAGE_SIZE - handle->page_offset;
 817                 handle->page_offset = 0;
 818                 handle->page++;
 819         } else {
 820                 handle->page_offset += count;
 821         }
 822         handle->offset += count;
 823         return count;
 824 }
 825
 826 int snapshot_image_loaded(struct snapshot_handle *handle)
 827 {
 828         return !(!handle->pbe || handle->pbe->next || !nr_copy_pages ||
 829                 handle->page <= nr_meta_pages + nr_copy_pages);
 830 }