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%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 #else
 154 static inline unsigned int count_highmem_pages(void) {return 0;}
 155 static inline int save_highmem(void) {return 0;}
 156 static inline int restore_highmem(void) {return 0;}
 157 #endif
 158
 159 /**
 160  *      @safe_needed - on resume, for storing the PBE list and the image,
 161  *      we can only use memory pages that do not conflict with the pages
 162  *      used before suspend.
 163  *
 164  *      The unsafe pages are marked with the PG_nosave_free flag
 165  *      and we count them using unsafe_pages
 166  */
 167
 168 static unsigned int unsafe_pages;
 169
 170 static void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
 171 {
 172         void *res;
 173
 174         res = (void *)get_zeroed_page(gfp_mask);
 175         if (safe_needed)
 176                 while (res && PageNosaveFree(virt_to_page(res))) {
 177                         /* The page is unsafe, mark it for swsusp_free() */
 178                         SetPageNosave(virt_to_page(res));
 179                         unsafe_pages++;
 180                         res = (void *)get_zeroed_page(gfp_mask);
 181                 }
 182         if (res) {
 183                 SetPageNosave(virt_to_page(res));
 184                 SetPageNosaveFree(virt_to_page(res));
 185         }
 186         return res;
 187 }
 188
 189 unsigned long get_safe_page(gfp_t gfp_mask)
 190 {
 191         return (unsigned long)alloc_image_page(gfp_mask, 1);
 192 }
 193
 194 /**
 195  *      free_image_page - free page represented by @addr, allocated with
 196  *      alloc_image_page (page flags set by it must be cleared)
 197  */
 198
 199 static inline void free_image_page(void *addr, int clear_nosave_free)
 200 {
 201         ClearPageNosave(virt_to_page(addr));
 202         if (clear_nosave_free)
 203                 ClearPageNosaveFree(virt_to_page(addr));
 204         free_page((unsigned long)addr);
 205 }
 206
 207 /**
 208  *      pfn_is_nosave - check if given pfn is in the 'nosave' section
 209  */
 210
 211 static inline int pfn_is_nosave(unsigned long pfn)
 212 {
 213         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 214         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 215         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 216 }
 217
 218 /**
 219  *      saveable - Determine whether a page should be cloned or not.
 220  *      @pfn:   The page
 221  *
 222  *      We save a page if it isn't Nosave, and is not in the range of pages
 223  *      statically defined as 'unsaveable', and it
 224  *      isn't a part of a free chunk of pages.
 225  */
 226
 227 static struct page *saveable_page(unsigned long pfn)
 228 {
 229         struct page *page;
 230
 231         if (!pfn_valid(pfn))
 232                 return NULL;
 233
 234         page = pfn_to_page(pfn);
 235
 236         if (PageNosave(page))
 237                 return NULL;
 238         if (PageReserved(page) && pfn_is_nosave(pfn))
 239                 return NULL;
 240         if (PageNosaveFree(page))
 241                 return NULL;
 242
 243         return page;
 244 }
 245
 246 unsigned int count_data_pages(void)
 247 {
 248         struct zone *zone;
 249         unsigned long pfn, max_zone_pfn;
 250         unsigned int n = 0;
 251
 252         for_each_zone (zone) {
 253                 if (is_highmem(zone))
 254                         continue;
 255                 mark_free_pages(zone);
 256                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 257                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 258                         n += !!saveable_page(pfn);
 259         }
 260         return n;
 261 }
 262
 263 static inline void copy_data_page(long *dst, long *src)
 264 {
 265         int n;
 266
 267         /* copy_page and memcpy are not usable for copying task structs. */
 268         for (n = PAGE_SIZE / sizeof(long); n; n--)
 269                 *dst++ = *src++;
 270 }
 271
 272 static void copy_data_pages(struct pbe *pblist)
 273 {
 274         struct zone *zone;
 275         unsigned long pfn, max_zone_pfn;
 276         struct pbe *pbe;
 277
 278         pbe = pblist;
 279         for_each_zone (zone) {
 280                 if (is_highmem(zone))
 281                         continue;
 282                 mark_free_pages(zone);
 283                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 284                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) {
 285                         struct page *page = saveable_page(pfn);
 286
 287                         if (page) {
 288                                 void *ptr = page_address(page);
 289
 290                                 BUG_ON(!pbe);
 291                                 copy_data_page((void *)pbe->address, ptr);
 292                                 pbe->orig_address = (unsigned long)ptr;
 293                                 pbe = pbe->next;
 294                         }
 295                 }
 296         }
 297         BUG_ON(pbe);
 298 }
 299
 300 /**
 301  *      free_pagedir - free pages allocated with alloc_pagedir()
 302  */
 303
 304 static void free_pagedir(struct pbe *pblist, int clear_nosave_free)
 305 {
 306         struct pbe *pbe;
 307
 308         while (pblist) {
 309                 pbe = (pblist + PB_PAGE_SKIP)->next;
 310                 free_image_page(pblist, clear_nosave_free);
 311                 pblist = pbe;
 312         }
 313 }
 314
 315 /**
 316  *      fill_pb_page - Create a list of PBEs on a given memory page
 317  */
 318
 319 static inline void fill_pb_page(struct pbe *pbpage, unsigned int n)
 320 {
 321         struct pbe *p;
 322
 323         p = pbpage;
 324         pbpage += n - 1;
 325         do
 326                 p->next = p + 1;
 327         while (++p < pbpage);
 328 }
 329
 330 /**
 331  *      create_pbe_list - Create a list of PBEs on top of a given chain
 332  *      of memory pages allocated with alloc_pagedir()
 333  *
 334  *      This function assumes that pages allocated by alloc_image_page() will
 335  *      always be zeroed.
 336  */
 337
 338 static inline void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
 339 {
 340         struct pbe *pbpage;
 341         unsigned int num = PBES_PER_PAGE;
 342
 343         for_each_pb_page (pbpage, pblist) {
 344                 if (num >= nr_pages)
 345                         break;
 346
 347                 fill_pb_page(pbpage, PBES_PER_PAGE);
 348                 num += PBES_PER_PAGE;
 349         }
 350         if (pbpage) {
 351                 num -= PBES_PER_PAGE;
 352                 fill_pb_page(pbpage, nr_pages - num);
 353         }
 354 }
 355
 356 /**
 357  *      alloc_pagedir - Allocate the page directory.
 358  *
 359  *      First, determine exactly how many pages we need and
 360  *      allocate them.
 361  *
 362  *      We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
 363  *      struct pbe elements (pbes) and the last element in the page points
 364  *      to the next page.
 365  *
 366  *      On each page we set up a list of struct_pbe elements.
 367  */
 368
 369 static struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask,
 370                                  int safe_needed)
 371 {
 372         unsigned int num;
 373         struct pbe *pblist, *pbe;
 374
 375         if (!nr_pages)
 376                 return NULL;
 377
 378         pblist = alloc_image_page(gfp_mask, safe_needed);
 379         pbe = pblist;
 380         for (num = PBES_PER_PAGE; num < nr_pages; num += PBES_PER_PAGE) {
 381                 if (!pbe) {
 382                         free_pagedir(pblist, 1);
 383                         return NULL;
 384                 }
 385                 pbe += PB_PAGE_SKIP;
 386                 pbe->next = alloc_image_page(gfp_mask, safe_needed);
 387                 pbe = pbe->next;
 388         }
 389         create_pbe_list(pblist, nr_pages);
 390         return pblist;
 391 }
 392
 393 /**
 394  * Free pages we allocated for suspend. Suspend pages are alocated
 395  * before atomic copy, so we need to free them after resume.
 396  */
 397
 398 void swsusp_free(void)
 399 {
 400         struct zone *zone;
 401         unsigned long pfn, max_zone_pfn;
 402
 403         for_each_zone(zone) {
 404                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 405                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 406                         if (pfn_valid(pfn)) {
 407                                 struct page *page = pfn_to_page(pfn);
 408
 409                                 if (PageNosave(page) && PageNosaveFree(page)) {
 410                                         ClearPageNosave(page);
 411                                         ClearPageNosaveFree(page);
 412                                         free_page((long) page_address(page));
 413                                 }
 414                         }
 415         }
 416         nr_copy_pages = 0;
 417         nr_meta_pages = 0;
 418         pagedir_nosave = NULL;
 419         buffer = NULL;
 420 }
 421
 422
 423 /**
 424  *      enough_free_mem - Make sure we enough free memory to snapshot.
 425  *
 426  *      Returns TRUE or FALSE after checking the number of available
 427  *      free pages.
 428  */
 429
 430 static int enough_free_mem(unsigned int nr_pages)
 431 {
 432         struct zone *zone;
 433         unsigned int n = 0;
 434
 435         for_each_zone (zone)
 436                 if (!is_highmem(zone))
 437                         n += zone->free_pages;
 438         pr_debug("swsusp: available memory: %u pages\n", n);
 439         return n > (nr_pages + PAGES_FOR_IO +
 440                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 441 }
 442
 443 static int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
 444 {
 445         struct pbe *p;
 446
 447         for_each_pbe (p, pblist) {
 448                 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
 449                 if (!p->address)
 450                         return -ENOMEM;
 451         }
 452         return 0;
 453 }
 454
 455 static struct pbe *swsusp_alloc(unsigned int nr_pages)
 456 {
 457         struct pbe *pblist;
 458
 459         if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
 460                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 461                 return NULL;
 462         }
 463
 464         if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
 465                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 466                 swsusp_free();
 467                 return NULL;
 468         }
 469
 470         return pblist;
 471 }
 472
 473 asmlinkage int swsusp_save(void)
 474 {
 475         unsigned int nr_pages;
 476
 477         pr_debug("swsusp: critical section: \n");
 478
 479         drain_local_pages();
 480         nr_pages = count_data_pages();
 481         printk("swsusp: Need to copy %u pages\n", nr_pages);
 482
 483         pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
 484                  nr_pages,
 485                  (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
 486                  PAGES_FOR_IO, nr_free_pages());
 487
 488         if (!enough_free_mem(nr_pages)) {
 489                 printk(KERN_ERR "swsusp: Not enough free memory\n");
 490                 return -ENOMEM;
 491         }
 492
 493         pagedir_nosave = swsusp_alloc(nr_pages);
 494         if (!pagedir_nosave)
 495                 return -ENOMEM;
 496
 497         /* During allocating of suspend pagedir, new cold pages may appear.
 498          * Kill them.
 499          */
 500         drain_local_pages();
 501         copy_data_pages(pagedir_nosave);
 502
 503         /*
 504          * End of critical section. From now on, we can write to memory,
 505          * but we should not touch disk. This specially means we must _not_
 506          * touch swap space! Except we must write out our image of course.
 507          */
 508
 509         nr_copy_pages = nr_pages;
 510         nr_meta_pages = (nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 511
 512         printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
 513         return 0;
 514 }
 515
 516 static void init_header(struct swsusp_info *info)
 517 {
 518         memset(info, 0, sizeof(struct swsusp_info));
 519         info->version_code = LINUX_VERSION_CODE;
 520         info->num_physpages = num_physpages;
 521         memcpy(&info->uts, &system_utsname, sizeof(system_utsname));
 522         info->cpus = num_online_cpus();
 523         info->image_pages = nr_copy_pages;
 524         info->pages = nr_copy_pages + nr_meta_pages + 1;
 525         info->size = info->pages;
 526         info->size <<= PAGE_SHIFT;
 527 }
 528
 529 /**
 530  *      pack_orig_addresses - the .orig_address fields of the PBEs from the
 531  *      list starting at @pbe are stored in the array @buf[] (1 page)
 532  */
 533
 534 static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pbe)
 535 {
 536         int j;
 537
 538         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 539                 buf[j] = pbe->orig_address;
 540                 pbe = pbe->next;
 541         }
 542         if (!pbe)
 543                 for (; j < PAGE_SIZE / sizeof(long); j++)
 544                         buf[j] = 0;
 545         return pbe;
 546 }
 547
 548 /**
 549  *      snapshot_read_next - used for reading the system memory snapshot.
 550  *
 551  *      On the first call to it @handle should point to a zeroed
 552  *      snapshot_handle structure.  The structure gets updated and a pointer
 553  *      to it should be passed to this function every next time.
 554  *
 555  *      The @count parameter should contain the number of bytes the caller
 556  *      wants to read from the snapshot.  It must not be zero.
 557  *
 558  *      On success the function returns a positive number.  Then, the caller
 559  *      is allowed to read up to the returned number of bytes from the memory
 560  *      location computed by the data_of() macro.  The number returned
 561  *      may be smaller than @count, but this only happens if the read would
 562  *      cross a page boundary otherwise.
 563  *
 564  *      The function returns 0 to indicate the end of data stream condition,
 565  *      and a negative number is returned on error.  In such cases the
 566  *      structure pointed to by @handle is not updated and should not be used
 567  *      any more.
 568  */
 569
 570 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
 571 {
 572         if (handle->cur > nr_meta_pages + nr_copy_pages)
 573                 return 0;
 574         if (!buffer) {
 575                 /* This makes the buffer be freed by swsusp_free() */
 576                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 577                 if (!buffer)
 578                         return -ENOMEM;
 579         }
 580         if (!handle->offset) {
 581                 init_header((struct swsusp_info *)buffer);
 582                 handle->buffer = buffer;
 583                 handle->pbe = pagedir_nosave;
 584         }
 585         if (handle->prev < handle->cur) {
 586                 if (handle->cur <= nr_meta_pages) {
 587                         handle->pbe = pack_orig_addresses(buffer, handle->pbe);
 588                         if (!handle->pbe)
 589                                 handle->pbe = pagedir_nosave;
 590                 } else {
 591                         handle->buffer = (void *)handle->pbe->address;
 592                         handle->pbe = handle->pbe->next;
 593                 }
 594                 handle->prev = handle->cur;
 595         }
 596         handle->buf_offset = handle->cur_offset;
 597         if (handle->cur_offset + count >= PAGE_SIZE) {
 598                 count = PAGE_SIZE - handle->cur_offset;
 599                 handle->cur_offset = 0;
 600                 handle->cur++;
 601         } else {
 602                 handle->cur_offset += count;
 603         }
 604         handle->offset += count;
 605         return count;
 606 }
 607
 608 /**
 609  *      mark_unsafe_pages - mark the pages that cannot be used for storing
 610  *      the image during resume, because they conflict with the pages that
 611  *      had been used before suspend
 612  */
 613
 614 static int mark_unsafe_pages(struct pbe *pblist)
 615 {
 616         struct zone *zone;
 617         unsigned long pfn, max_zone_pfn;
 618         struct pbe *p;
 619
 620         if (!pblist) /* a sanity check */
 621                 return -EINVAL;
 622
 623         /* Clear page flags */
 624         for_each_zone (zone) {
 625                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 626                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 627                         if (pfn_valid(pfn))
 628                                 ClearPageNosaveFree(pfn_to_page(pfn));
 629         }
 630
 631         /* Mark orig addresses */
 632         for_each_pbe (p, pblist) {
 633                 if (virt_addr_valid(p->orig_address))
 634                         SetPageNosaveFree(virt_to_page(p->orig_address));
 635                 else
 636                         return -EFAULT;
 637         }
 638
 639         unsafe_pages = 0;
 640
 641         return 0;
 642 }
 643
 644 static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
 645 {
 646         /* We assume both lists contain the same number of elements */
 647         while (src) {
 648                 dst->orig_address = src->orig_address;
 649                 dst = dst->next;
 650                 src = src->next;
 651         }
 652 }
 653
 654 static int check_header(struct swsusp_info *info)
 655 {
 656         char *reason = NULL;
 657
 658         if (info->version_code != LINUX_VERSION_CODE)
 659                 reason = "kernel version";
 660         if (info->num_physpages != num_physpages)
 661                 reason = "memory size";
 662         if (strcmp(info->uts.sysname,system_utsname.sysname))
 663                 reason = "system type";
 664         if (strcmp(info->uts.release,system_utsname.release))
 665                 reason = "kernel release";
 666         if (strcmp(info->uts.version,system_utsname.version))
 667                 reason = "version";
 668         if (strcmp(info->uts.machine,system_utsname.machine))
 669                 reason = "machine";
 670         if (reason) {
 671                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
 672                 return -EPERM;
 673         }
 674         return 0;
 675 }
 676
 677 /**
 678  *      load header - check the image header and copy data from it
 679  */
 680
 681 static int load_header(struct snapshot_handle *handle,
 682                               struct swsusp_info *info)
 683 {
 684         int error;
 685         struct pbe *pblist;
 686
 687         error = check_header(info);
 688         if (!error) {
 689                 pblist = alloc_pagedir(info->image_pages, GFP_ATOMIC, 0);
 690                 if (!pblist)
 691                         return -ENOMEM;
 692                 pagedir_nosave = pblist;
 693                 handle->pbe = pblist;
 694                 nr_copy_pages = info->image_pages;
 695                 nr_meta_pages = info->pages - info->image_pages - 1;
 696         }
 697         return error;
 698 }
 699
 700 /**
 701  *      unpack_orig_addresses - copy the elements of @buf[] (1 page) to
 702  *      the PBEs in the list starting at @pbe
 703  */
 704
 705 static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
 706                                                 struct pbe *pbe)
 707 {
 708         int j;
 709
 710         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 711                 pbe->orig_address = buf[j];
 712                 pbe = pbe->next;
 713         }
 714         return pbe;
 715 }
 716
 717 /**
 718  *      prepare_image - use metadata contained in the PBE list
 719  *      pointed to by pagedir_nosave to mark the pages that will
 720  *      be overwritten in the process of restoring the system
 721  *      memory state from the image ("unsafe" pages) and allocate
 722  *      memory for the image
 723  *
 724  *      The idea is to allocate the PBE list first and then
 725  *      allocate as many pages as it's needed for the image data,
 726  *      but not to assign these pages to the PBEs initially.
 727  *      Instead, we just mark them as allocated and create a list
 728  *      of "safe" which will be used later
 729  */
 730
 731 struct safe_page {
 732         struct safe_page *next;
 733         char padding[PAGE_SIZE - sizeof(void *)];
 734 };
 735
 736 static struct safe_page *safe_pages;
 737
 738 static int prepare_image(struct snapshot_handle *handle)
 739 {
 740         int error = 0;
 741         unsigned int nr_pages = nr_copy_pages;
 742         struct pbe *p, *pblist = NULL;
 743
 744         p = pagedir_nosave;
 745         error = mark_unsafe_pages(p);
 746         if (!error) {
 747                 pblist = alloc_pagedir(nr_pages, GFP_ATOMIC, 1);
 748                 if (pblist)
 749                         copy_page_backup_list(pblist, p);
 750                 free_pagedir(p, 0);
 751                 if (!pblist)
 752                         error = -ENOMEM;
 753         }
 754         safe_pages = NULL;
 755         if (!error && nr_pages > unsafe_pages) {
 756                 nr_pages -= unsafe_pages;
 757                 while (nr_pages--) {
 758                         struct safe_page *ptr;
 759
 760                         ptr = (struct safe_page *)get_zeroed_page(GFP_ATOMIC);
 761                         if (!ptr) {
 762                                 error = -ENOMEM;
 763                                 break;
 764                         }
 765                         if (!PageNosaveFree(virt_to_page(ptr))) {
 766                                 /* The page is "safe", add it to the list */
 767                                 ptr->next = safe_pages;
 768                                 safe_pages = ptr;
 769                         }
 770                         /* Mark the page as allocated */
 771                         SetPageNosave(virt_to_page(ptr));
 772                         SetPageNosaveFree(virt_to_page(ptr));
 773                 }
 774         }
 775         if (!error) {
 776                 pagedir_nosave = pblist;
 777         } else {
 778                 handle->pbe = NULL;
 779                 swsusp_free();
 780         }
 781         return error;
 782 }
 783
 784 static void *get_buffer(struct snapshot_handle *handle)
 785 {
 786         struct pbe *pbe = handle->pbe, *last = handle->last_pbe;
 787         struct page *page = virt_to_page(pbe->orig_address);
 788
 789         if (PageNosave(page) && PageNosaveFree(page)) {
 790                 /*
 791                  * We have allocated the "original" page frame and we can
 792                  * use it directly to store the read page
 793                  */
 794                 pbe->address = 0;
 795                 if (last && last->next)
 796                         last->next = NULL;
 797                 return (void *)pbe->orig_address;
 798         }
 799         /*
 800          * The "original" page frame has not been allocated and we have to
 801          * use a "safe" page frame to store the read page
 802          */
 803         pbe->address = (unsigned long)safe_pages;
 804         safe_pages = safe_pages->next;
 805         if (last)
 806                 last->next = pbe;
 807         handle->last_pbe = pbe;
 808         return (void *)pbe->address;
 809 }
 810
 811 /**
 812  *      snapshot_write_next - used for writing the system memory snapshot.
 813  *
 814  *      On the first call to it @handle should point to a zeroed
 815  *      snapshot_handle structure.  The structure gets updated and a pointer
 816  *      to it should be passed to this function every next time.
 817  *
 818  *      The @count parameter should contain the number of bytes the caller
 819  *      wants to write to the image.  It must not be zero.
 820  *
 821  *      On success the function returns a positive number.  Then, the caller
 822  *      is allowed to write up to the returned number of bytes to the memory
 823  *      location computed by the data_of() macro.  The number returned
 824  *      may be smaller than @count, but this only happens if the write would
 825  *      cross a page boundary otherwise.
 826  *
 827  *      The function returns 0 to indicate the "end of file" condition,
 828  *      and a negative number is returned on error.  In such cases the
 829  *      structure pointed to by @handle is not updated and should not be used
 830  *      any more.
 831  */
 832
 833 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
 834 {
 835         int error = 0;
 836
 837         if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages)
 838                 return 0;
 839         if (!buffer) {
 840                 /* This makes the buffer be freed by swsusp_free() */
 841                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 842                 if (!buffer)
 843                         return -ENOMEM;
 844         }
 845         if (!handle->offset)
 846                 handle->buffer = buffer;
 847         handle->sync_read = 1;
 848         if (handle->prev < handle->cur) {
 849                 if (!handle->prev) {
 850                         error = load_header(handle,
 851                                         (struct swsusp_info *)buffer);
 852                         if (error)
 853                                 return error;
 854                 } else if (handle->prev <= nr_meta_pages) {
 855                         handle->pbe = unpack_orig_addresses(buffer,
 856                                                         handle->pbe);
 857                         if (!handle->pbe) {
 858                                 error = prepare_image(handle);
 859                                 if (error)
 860                                         return error;
 861                                 handle->pbe = pagedir_nosave;
 862                                 handle->last_pbe = NULL;
 863                                 handle->buffer = get_buffer(handle);
 864                                 handle->sync_read = 0;
 865                         }
 866                 } else {
 867                         handle->pbe = handle->pbe->next;
 868                         handle->buffer = get_buffer(handle);
 869                         handle->sync_read = 0;
 870                 }
 871                 handle->prev = handle->cur;
 872         }
 873         handle->buf_offset = handle->cur_offset;
 874         if (handle->cur_offset + count >= PAGE_SIZE) {
 875                 count = PAGE_SIZE - handle->cur_offset;
 876                 handle->cur_offset = 0;
 877                 handle->cur++;
 878         } else {
 879                 handle->cur_offset += count;
 880         }
 881         handle->offset += count;
 882         return count;
 883 }
 884
 885 int snapshot_image_loaded(struct snapshot_handle *handle)
 886 {
 887         return !(!handle->pbe || handle->pbe->next || !nr_copy_pages ||
 888                 handle->cur <= nr_meta_pages + nr_copy_pages);
 889 }