SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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%1000))
  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\n");
 126         for_each_zone (zone) {
 127                 if (is_highmem(zone))
 128                         res = save_highmem_zone(zone);
 129                 if (res)
 130                         return res;
 131         }
 132         return 0;
 133 }
 134
 135 int restore_highmem(void)
 136 {
 137         printk("swsusp: Restoring Highmem\n");
 138         while (highmem_copy) {
 139                 struct highmem_page *save = highmem_copy;
 140                 void *kaddr;
 141                 highmem_copy = save->next;
 142
 143                 kaddr = kmap_atomic(save->page, KM_USER0);
 144                 memcpy(kaddr, save->data, PAGE_SIZE);
 145                 kunmap_atomic(kaddr, KM_USER0);
 146                 free_page((long) save->data);
 147                 kfree(save);
 148         }
 149         return 0;
 150 }
 151 #endif
 152
 153 static int pfn_is_nosave(unsigned long pfn)
 154 {
 155         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 156         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 157         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 158 }
 159
 160 /**
 161  *      saveable - Determine whether a page should be cloned or not.
 162  *      @pfn:   The page
 163  *
 164  *      We save a page if it's Reserved, and not in the range of pages
 165  *      statically defined as 'unsaveable', or if it isn't reserved, and
 166  *      isn't part of a free chunk of pages.
 167  */
 168
 169 static int saveable(struct zone *zone, unsigned long *zone_pfn)
 170 {
 171         unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
 172         struct page *page;
 173
 174         if (!pfn_valid(pfn))
 175                 return 0;
 176
 177         page = pfn_to_page(pfn);
 178         BUG_ON(PageReserved(page) && PageNosave(page));
 179         if (PageNosave(page))
 180                 return 0;
 181         if (PageReserved(page) && pfn_is_nosave(pfn))
 182                 return 0;
 183         if (PageNosaveFree(page))
 184                 return 0;
 185
 186         return 1;
 187 }
 188
 189 unsigned int count_data_pages(void)
 190 {
 191         struct zone *zone;
 192         unsigned long zone_pfn;
 193         unsigned int n = 0;
 194
 195         for_each_zone (zone) {
 196                 if (is_highmem(zone))
 197                         continue;
 198                 mark_free_pages(zone);
 199                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 200                         n += saveable(zone, &zone_pfn);
 201         }
 202         return n;
 203 }
 204
 205 static void copy_data_pages(struct pbe *pblist)
 206 {
 207         struct zone *zone;
 208         unsigned long zone_pfn;
 209         struct pbe *pbe, *p;
 210
 211         pbe = pblist;
 212         for_each_zone (zone) {
 213                 if (is_highmem(zone))
 214                         continue;
 215                 mark_free_pages(zone);
 216                 /* This is necessary for swsusp_free() */
 217                 for_each_pb_page (p, pblist)
 218                         SetPageNosaveFree(virt_to_page(p));
 219                 for_each_pbe (p, pblist)
 220                         SetPageNosaveFree(virt_to_page(p->address));
 221                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 222                         if (saveable(zone, &zone_pfn)) {
 223                                 struct page *page;
 224                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 225                                 BUG_ON(!pbe);
 226                                 pbe->orig_address = (unsigned long)page_address(page);
 227                                 /* copy_page is not usable for copying task structs. */
 228                                 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
 229                                 pbe = pbe->next;
 230                         }
 231                 }
 232         }
 233         BUG_ON(pbe);
 234 }
 235
 236
 237 /**
 238  *      free_pagedir - free pages allocated with alloc_pagedir()
 239  */
 240
 241 static void free_pagedir(struct pbe *pblist)
 242 {
 243         struct pbe *pbe;
 244
 245         while (pblist) {
 246                 pbe = (pblist + PB_PAGE_SKIP)->next;
 247                 ClearPageNosave(virt_to_page(pblist));
 248                 ClearPageNosaveFree(virt_to_page(pblist));
 249                 free_page((unsigned long)pblist);
 250                 pblist = pbe;
 251         }
 252 }
 253
 254 /**
 255  *      fill_pb_page - Create a list of PBEs on a given memory page
 256  */
 257
 258 static inline void fill_pb_page(struct pbe *pbpage)
 259 {
 260         struct pbe *p;
 261
 262         p = pbpage;
 263         pbpage += PB_PAGE_SKIP;
 264         do
 265                 p->next = p + 1;
 266         while (++p < pbpage);
 267 }
 268
 269 /**
 270  *      create_pbe_list - Create a list of PBEs on top of a given chain
 271  *      of memory pages allocated with alloc_pagedir()
 272  */
 273
 274 static inline void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
 275 {
 276         struct pbe *pbpage, *p;
 277         unsigned int num = PBES_PER_PAGE;
 278
 279         for_each_pb_page (pbpage, pblist) {
 280                 if (num >= nr_pages)
 281                         break;
 282
 283                 fill_pb_page(pbpage);
 284                 num += PBES_PER_PAGE;
 285         }
 286         if (pbpage) {
 287                 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
 288                         p->next = p + 1;
 289                 p->next = NULL;
 290         }
 291 }
 292
 293 /**
 294  *      On resume it is necessary to trace and eventually free the unsafe
 295  *      pages that have been allocated, because they are needed for I/O
 296  *      (on x86-64 we likely will "eat" these pages once again while
 297  *      creating the temporary page translation tables)
 298  */
 299
 300 struct eaten_page {
 301         struct eaten_page *next;
 302         char padding[PAGE_SIZE - sizeof(void *)];
 303 };
 304
 305 static struct eaten_page *eaten_pages = NULL;
 306
 307 static void release_eaten_pages(void)
 308 {
 309         struct eaten_page *p, *q;
 310
 311         p = eaten_pages;
 312         while (p) {
 313                 q = p->next;
 314                 /* We don't want swsusp_free() to free this page again */
 315                 ClearPageNosave(virt_to_page(p));
 316                 free_page((unsigned long)p);
 317                 p = q;
 318         }
 319         eaten_pages = NULL;
 320 }
 321
 322 /**
 323  *      @safe_needed - on resume, for storing the PBE list and the image,
 324  *      we can only use memory pages that do not conflict with the pages
 325  *      which had been used before suspend.
 326  *
 327  *      The unsafe pages are marked with the PG_nosave_free flag
 328  *
 329  *      Allocated but unusable (ie eaten) memory pages should be marked
 330  *      so that swsusp_free() can release them
 331  */
 332
 333 static inline void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
 334 {
 335         void *res;
 336
 337         if (safe_needed)
 338                 do {
 339                         res = (void *)get_zeroed_page(gfp_mask);
 340                         if (res && PageNosaveFree(virt_to_page(res))) {
 341                                 /* This is for swsusp_free() */
 342                                 SetPageNosave(virt_to_page(res));
 343                                 ((struct eaten_page *)res)->next = eaten_pages;
 344                                 eaten_pages = res;
 345                         }
 346                 } while (res && PageNosaveFree(virt_to_page(res)));
 347         else
 348                 res = (void *)get_zeroed_page(gfp_mask);
 349         if (res) {
 350                 SetPageNosave(virt_to_page(res));
 351                 SetPageNosaveFree(virt_to_page(res));
 352         }
 353         return res;
 354 }
 355
 356 unsigned long get_safe_page(gfp_t gfp_mask)
 357 {
 358         return (unsigned long)alloc_image_page(gfp_mask, 1);
 359 }
 360
 361 /**
 362  *      alloc_pagedir - Allocate the page directory.
 363  *
 364  *      First, determine exactly how many pages we need and
 365  *      allocate them.
 366  *
 367  *      We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
 368  *      struct pbe elements (pbes) and the last element in the page points
 369  *      to the next page.
 370  *
 371  *      On each page we set up a list of struct_pbe elements.
 372  */
 373
 374 struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask, int safe_needed)
 375 {
 376         unsigned int num;
 377         struct pbe *pblist, *pbe;
 378
 379         if (!nr_pages)
 380                 return NULL;
 381
 382         pblist = alloc_image_page(gfp_mask, safe_needed);
 383         /* FIXME: rewrite this ugly loop */
 384         for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
 385                         pbe = pbe->next, num += PBES_PER_PAGE) {
 386                 pbe += PB_PAGE_SKIP;
 387                 pbe->next = alloc_image_page(gfp_mask, safe_needed);
 388         }
 389         if (!pbe) { /* get_zeroed_page() failed */
 390                 free_pagedir(pblist);
 391                 pblist = NULL;
 392         } else
 393                 create_pbe_list(pblist, nr_pages);
 394         return pblist;
 395 }
 396
 397 /**
 398  * Free pages we allocated for suspend. Suspend pages are alocated
 399  * before atomic copy, so we need to free them after resume.
 400  */
 401
 402 void swsusp_free(void)
 403 {
 404         struct zone *zone;
 405         unsigned long zone_pfn;
 406
 407         for_each_zone(zone) {
 408                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 409                         if (pfn_valid(zone_pfn + zone->zone_start_pfn)) {
 410                                 struct page *page;
 411                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 412                                 if (PageNosave(page) && PageNosaveFree(page)) {
 413                                         ClearPageNosave(page);
 414                                         ClearPageNosaveFree(page);
 415                                         free_page((long) page_address(page));
 416                                 }
 417                         }
 418         }
 419         nr_copy_pages = 0;
 420         nr_meta_pages = 0;
 421         pagedir_nosave = NULL;
 422         buffer = NULL;
 423 }
 424
 425
 426 /**
 427  *      enough_free_mem - Make sure we enough free memory to snapshot.
 428  *
 429  *      Returns TRUE or FALSE after checking the number of available
 430  *      free pages.
 431  */
 432
 433 static int enough_free_mem(unsigned int nr_pages)
 434 {
 435         struct zone *zone;
 436         unsigned int n = 0;
 437
 438         for_each_zone (zone)
 439                 if (!is_highmem(zone))
 440                         n += zone->free_pages;
 441         pr_debug("swsusp: available memory: %u pages\n", n);
 442         return n > (nr_pages + PAGES_FOR_IO +
 443                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 444 }
 445
 446 static int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
 447 {
 448         struct pbe *p;
 449
 450         for_each_pbe (p, pblist) {
 451                 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
 452                 if (!p->address)
 453                         return -ENOMEM;
 454         }
 455         return 0;
 456 }
 457
 458 static struct pbe *swsusp_alloc(unsigned int nr_pages)
 459 {
 460         struct pbe *pblist;
 461
 462         if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
 463                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 464                 return NULL;
 465         }
 466
 467         if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
 468                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 469                 swsusp_free();
 470                 return NULL;
 471         }
 472
 473         return pblist;
 474 }
 475
 476 asmlinkage int swsusp_save(void)
 477 {
 478         unsigned int nr_pages;
 479
 480         pr_debug("swsusp: critical section: \n");
 481
 482         drain_local_pages();
 483         nr_pages = count_data_pages();
 484         printk("swsusp: Need to copy %u pages\n", nr_pages);
 485
 486         pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
 487                  nr_pages,
 488                  (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
 489                  PAGES_FOR_IO, nr_free_pages());
 490
 491         if (!enough_free_mem(nr_pages)) {
 492                 printk(KERN_ERR "swsusp: Not enough free memory\n");
 493                 return -ENOMEM;
 494         }
 495
 496         pagedir_nosave = swsusp_alloc(nr_pages);
 497         if (!pagedir_nosave)
 498                 return -ENOMEM;
 499
 500         /* During allocating of suspend pagedir, new cold pages may appear.
 501          * Kill them.
 502          */
 503         drain_local_pages();
 504         copy_data_pages(pagedir_nosave);
 505
 506         /*
 507          * End of critical section. From now on, we can write to memory,
 508          * but we should not touch disk. This specially means we must _not_
 509          * touch swap space! Except we must write out our image of course.
 510          */
 511
 512         nr_copy_pages = nr_pages;
 513         nr_meta_pages = (nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 514
 515         printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
 516         return 0;
 517 }
 518
 519 static void init_header(struct swsusp_info *info)
 520 {
 521         memset(info, 0, sizeof(struct swsusp_info));
 522         info->version_code = LINUX_VERSION_CODE;
 523         info->num_physpages = num_physpages;
 524         memcpy(&info->uts, &system_utsname, sizeof(system_utsname));
 525         info->cpus = num_online_cpus();
 526         info->image_pages = nr_copy_pages;
 527         info->pages = nr_copy_pages + nr_meta_pages + 1;
 528         info->size = info->pages;
 529         info->size <<= PAGE_SHIFT;
 530 }
 531
 532 /**
 533  *      pack_orig_addresses - the .orig_address fields of the PBEs from the
 534  *      list starting at @pbe are stored in the array @buf[] (1 page)
 535  */
 536
 537 static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pbe)
 538 {
 539         int j;
 540
 541         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 542                 buf[j] = pbe->orig_address;
 543                 pbe = pbe->next;
 544         }
 545         if (!pbe)
 546                 for (; j < PAGE_SIZE / sizeof(long); j++)
 547                         buf[j] = 0;
 548         return pbe;
 549 }
 550
 551 /**
 552  *      snapshot_read_next - used for reading the system memory snapshot.
 553  *
 554  *      On the first call to it @handle should point to a zeroed
 555  *      snapshot_handle structure.  The structure gets updated and a pointer
 556  *      to it should be passed to this function every next time.
 557  *
 558  *      The @count parameter should contain the number of bytes the caller
 559  *      wants to read from the snapshot.  It must not be zero.
 560  *
 561  *      On success the function returns a positive number.  Then, the caller
 562  *      is allowed to read up to the returned number of bytes from the memory
 563  *      location computed by the data_of() macro.  The number returned
 564  *      may be smaller than @count, but this only happens if the read would
 565  *      cross a page boundary otherwise.
 566  *
 567  *      The function returns 0 to indicate the end of data stream condition,
 568  *      and a negative number is returned on error.  In such cases the
 569  *      structure pointed to by @handle is not updated and should not be used
 570  *      any more.
 571  */
 572
 573 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
 574 {
 575         if (handle->page > nr_meta_pages + nr_copy_pages)
 576                 return 0;
 577         if (!buffer) {
 578                 /* This makes the buffer be freed by swsusp_free() */
 579                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 580                 if (!buffer)
 581                         return -ENOMEM;
 582         }
 583         if (!handle->offset) {
 584                 init_header((struct swsusp_info *)buffer);
 585                 handle->buffer = buffer;
 586                 handle->pbe = pagedir_nosave;
 587         }
 588         if (handle->prev < handle->page) {
 589                 if (handle->page <= nr_meta_pages) {
 590                         handle->pbe = pack_orig_addresses(buffer, handle->pbe);
 591                         if (!handle->pbe)
 592                                 handle->pbe = pagedir_nosave;
 593                 } else {
 594                         handle->buffer = (void *)handle->pbe->address;
 595                         handle->pbe = handle->pbe->next;
 596                 }
 597                 handle->prev = handle->page;
 598         }
 599         handle->buf_offset = handle->page_offset;
 600         if (handle->page_offset + count >= PAGE_SIZE) {
 601                 count = PAGE_SIZE - handle->page_offset;
 602                 handle->page_offset = 0;
 603                 handle->page++;
 604         } else {
 605                 handle->page_offset += count;
 606         }
 607         handle->offset += count;
 608         return count;
 609 }
 610
 611 /**
 612  *      mark_unsafe_pages - mark the pages that cannot be used for storing
 613  *      the image during resume, because they conflict with the pages that
 614  *      had been used before suspend
 615  */
 616
 617 static int mark_unsafe_pages(struct pbe *pblist)
 618 {
 619         struct zone *zone;
 620         unsigned long zone_pfn;
 621         struct pbe *p;
 622
 623         if (!pblist) /* a sanity check */
 624                 return -EINVAL;
 625
 626         /* Clear page flags */
 627         for_each_zone (zone) {
 628                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 629                         if (pfn_valid(zone_pfn + zone->zone_start_pfn))
 630                                 ClearPageNosaveFree(pfn_to_page(zone_pfn +
 631                                         zone->zone_start_pfn));
 632         }
 633
 634         /* Mark orig addresses */
 635         for_each_pbe (p, pblist) {
 636                 if (virt_addr_valid(p->orig_address))
 637                         SetPageNosaveFree(virt_to_page(p->orig_address));
 638                 else
 639                         return -EFAULT;
 640         }
 641
 642         return 0;
 643 }
 644
 645 static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
 646 {
 647         /* We assume both lists contain the same number of elements */
 648         while (src) {
 649                 dst->orig_address = src->orig_address;
 650                 dst = dst->next;
 651                 src = src->next;
 652         }
 653 }
 654
 655 static int check_header(struct swsusp_info *info)
 656 {
 657         char *reason = NULL;
 658
 659         if (info->version_code != LINUX_VERSION_CODE)
 660                 reason = "kernel version";
 661         if (info->num_physpages != num_physpages)
 662                 reason = "memory size";
 663         if (strcmp(info->uts.sysname,system_utsname.sysname))
 664                 reason = "system type";
 665         if (strcmp(info->uts.release,system_utsname.release))
 666                 reason = "kernel release";
 667         if (strcmp(info->uts.version,system_utsname.version))
 668                 reason = "version";
 669         if (strcmp(info->uts.machine,system_utsname.machine))
 670                 reason = "machine";
 671         if (reason) {
 672                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
 673                 return -EPERM;
 674         }
 675         return 0;
 676 }
 677
 678 /**
 679  *      load header - check the image header and copy data from it
 680  */
 681
 682 static int load_header(struct snapshot_handle *handle,
 683                               struct swsusp_info *info)
 684 {
 685         int error;
 686         struct pbe *pblist;
 687
 688         error = check_header(info);
 689         if (!error) {
 690                 pblist = alloc_pagedir(info->image_pages, GFP_ATOMIC, 0);
 691                 if (!pblist)
 692                         return -ENOMEM;
 693                 pagedir_nosave = pblist;
 694                 handle->pbe = pblist;
 695                 nr_copy_pages = info->image_pages;
 696                 nr_meta_pages = info->pages - info->image_pages - 1;
 697         }
 698         return error;
 699 }
 700
 701 /**
 702  *      unpack_orig_addresses - copy the elements of @buf[] (1 page) to
 703  *      the PBEs in the list starting at @pbe
 704  */
 705
 706 static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
 707                                                 struct pbe *pbe)
 708 {
 709         int j;
 710
 711         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 712                 pbe->orig_address = buf[j];
 713                 pbe = pbe->next;
 714         }
 715         return pbe;
 716 }
 717
 718 /**
 719  *      create_image - use metadata contained in the PBE list
 720  *      pointed to by pagedir_nosave to mark the pages that will
 721  *      be overwritten in the process of restoring the system
 722  *      memory state from the image and allocate memory for
 723  *      the image avoiding these pages
 724  */
 725
 726 static int create_image(struct snapshot_handle *handle)
 727 {
 728         int error = 0;
 729         struct pbe *p, *pblist;
 730
 731         p = pagedir_nosave;
 732         error = mark_unsafe_pages(p);
 733         if (!error) {
 734                 pblist = alloc_pagedir(nr_copy_pages, GFP_ATOMIC, 1);
 735                 if (pblist)
 736                         copy_page_backup_list(pblist, p);
 737                 free_pagedir(p);
 738                 if (!pblist)
 739                         error = -ENOMEM;
 740         }
 741         if (!error)
 742                 error = alloc_data_pages(pblist, GFP_ATOMIC, 1);
 743         if (!error) {
 744                 release_eaten_pages();
 745                 pagedir_nosave = pblist;
 746         } else {
 747                 pagedir_nosave = NULL;
 748                 handle->pbe = NULL;
 749                 nr_copy_pages = 0;
 750                 nr_meta_pages = 0;
 751         }
 752         return error;
 753 }
 754
 755 /**
 756  *      snapshot_write_next - used for writing the system memory snapshot.
 757  *
 758  *      On the first call to it @handle should point to a zeroed
 759  *      snapshot_handle structure.  The structure gets updated and a pointer
 760  *      to it should be passed to this function every next time.
 761  *
 762  *      The @count parameter should contain the number of bytes the caller
 763  *      wants to write to the image.  It must not be zero.
 764  *
 765  *      On success the function returns a positive number.  Then, the caller
 766  *      is allowed to write up to the returned number of bytes to the memory
 767  *      location computed by the data_of() macro.  The number returned
 768  *      may be smaller than @count, but this only happens if the write would
 769  *      cross a page boundary otherwise.
 770  *
 771  *      The function returns 0 to indicate the "end of file" condition,
 772  *      and a negative number is returned on error.  In such cases the
 773  *      structure pointed to by @handle is not updated and should not be used
 774  *      any more.
 775  */
 776
 777 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
 778 {
 779         int error = 0;
 780
 781         if (handle->prev && handle->page > nr_meta_pages + nr_copy_pages)
 782                 return 0;
 783         if (!buffer) {
 784                 /* This makes the buffer be freed by swsusp_free() */
 785                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 786                 if (!buffer)
 787                         return -ENOMEM;
 788         }
 789         if (!handle->offset)
 790                 handle->buffer = buffer;
 791         if (handle->prev < handle->page) {
 792                 if (!handle->prev) {
 793                         error = load_header(handle, (struct swsusp_info *)buffer);
 794                         if (error)
 795                                 return error;
 796                 } else if (handle->prev <= nr_meta_pages) {
 797                         handle->pbe = unpack_orig_addresses(buffer, handle->pbe);
 798                         if (!handle->pbe) {
 799                                 error = create_image(handle);
 800                                 if (error)
 801                                         return error;
 802                                 handle->pbe = pagedir_nosave;
 803                                 handle->buffer = (void *)handle->pbe->address;
 804                         }
 805                 } else {
 806                         handle->pbe = handle->pbe->next;
 807                         handle->buffer = (void *)handle->pbe->address;
 808                 }
 809                 handle->prev = handle->page;
 810         }
 811         handle->buf_offset = handle->page_offset;
 812         if (handle->page_offset + count >= PAGE_SIZE) {
 813                 count = PAGE_SIZE - handle->page_offset;
 814                 handle->page_offset = 0;
 815                 handle->page++;
 816         } else {
 817                 handle->page_offset += count;
 818         }
 819         handle->offset += count;
 820         return count;
 821 }
 822
 823 int snapshot_image_loaded(struct snapshot_handle *handle)
 824 {
 825         return !(!handle->pbe || handle->pbe->next || !nr_copy_pages ||
 826                 handle->page <= nr_meta_pages + nr_copy_pages);
 827 }