kernel/power/swsusp.c

   1 /*
   2  * linux/kernel/power/swsusp.c
   3  *
   4  * This file is to realize architecture-independent
   5  * machine suspend feature using pretty near only high-level routines
   6  *
   7  * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu>
   8  * Copyright (C) 1998,2001-2004 Pavel Machek <pavel@suse.cz>
   9  *
  10  * This file is released under the GPLv2.
  11  *
  12  * I'd like to thank the following people for their work:
  13  *
  14  * Pavel Machek <pavel@ucw.cz>:
  15  * Modifications, defectiveness pointing, being with me at the very beginning,
  16  * suspend to swap space, stop all tasks. Port to 2.4.18-ac and 2.5.17.
  17  *
  18  * Steve Doddi <dirk@loth.demon.co.uk>:
  19  * Support the possibility of hardware state restoring.
  20  *
  21  * Raph <grey.havens@earthling.net>:
  22  * Support for preserving states of network devices and virtual console
  23  * (including X and svgatextmode)
  24  *
  25  * Kurt Garloff <garloff@suse.de>:
  26  * Straightened the critical function in order to prevent compilers from
  27  * playing tricks with local variables.
  28  *
  29  * Andreas Mohr <a.mohr@mailto.de>
  30  *
  31  * Alex Badea <vampire@go.ro>:
  32  * Fixed runaway init
  33  *
  34  * Andreas Steinmetz <ast@domdv.de>:
  35  * Added encrypted suspend option
  36  *
  37  * More state savers are welcome. Especially for the scsi layer...
  38  *
  39  * For TODOs,FIXMEs also look in Documentation/power/swsusp.txt
  40  */
  41
  42 #include <linux/module.h>
  43 #include <linux/mm.h>
  44 #include <linux/suspend.h>
  45 #include <linux/smp_lock.h>
  46 #include <linux/file.h>
  47 #include <linux/utsname.h>
  48 #include <linux/version.h>
  49 #include <linux/delay.h>
  50 #include <linux/reboot.h>
  51 #include <linux/bitops.h>
  52 #include <linux/vt_kern.h>
  53 #include <linux/kbd_kern.h>
  54 #include <linux/keyboard.h>
  55 #include <linux/spinlock.h>
  56 #include <linux/genhd.h>
  57 #include <linux/kernel.h>
  58 #include <linux/major.h>
  59 #include <linux/swap.h>
  60 #include <linux/pm.h>
  61 #include <linux/device.h>
  62 #include <linux/buffer_head.h>
  63 #include <linux/swapops.h>
  64 #include <linux/bootmem.h>
  65 #include <linux/syscalls.h>
  66 #include <linux/console.h>
  67 #include <linux/highmem.h>
  68 #include <linux/bio.h>
  69 #include <linux/mount.h>
  70
  71 #include <asm/uaccess.h>
  72 #include <asm/mmu_context.h>
  73 #include <asm/pgtable.h>
  74 #include <asm/tlbflush.h>
  75 #include <asm/io.h>
  76
  77 #include <linux/random.h>
  78 #include <linux/crypto.h>
  79 #include <asm/scatterlist.h>
  80
  81 #include "power.h"
  82
  83 #define CIPHER "aes"
  84 #define MAXKEY 32
  85 #define MAXIV  32
  86
  87 /* References to section boundaries */
  88 extern const void __nosave_begin, __nosave_end;
  89
  90 /* Variables to be preserved over suspend */
  91 static int nr_copy_pages_check;
  92
  93 extern char resume_file[];
  94
  95 /* Local variables that should not be affected by save */
  96 static unsigned int nr_copy_pages __nosavedata = 0;
  97
  98 /* Suspend pagedir is allocated before final copy, therefore it
  99    must be freed after resume
 100
 101    Warning: this is evil. There are actually two pagedirs at time of
 102    resume. One is "pagedir_save", which is empty frame allocated at
 103    time of suspend, that must be freed. Second is "pagedir_nosave",
 104    allocated at time of resume, that travels through memory not to
 105    collide with anything.
 106
 107    Warning: this is even more evil than it seems. Pagedirs this file
 108    talks about are completely different from page directories used by
 109    MMU hardware.
 110  */
 111 suspend_pagedir_t *pagedir_nosave __nosavedata = NULL;
 112 static suspend_pagedir_t *pagedir_save;
 113
 114 #define SWSUSP_SIG      "S1SUSPEND"
 115
 116 static struct swsusp_header {
 117         char reserved[PAGE_SIZE - 20 - MAXKEY - MAXIV - sizeof(swp_entry_t)];
 118         u8 key_iv[MAXKEY+MAXIV];
 119         swp_entry_t swsusp_info;
 120         char    orig_sig[10];
 121         char    sig[10];
 122 } __attribute__((packed, aligned(PAGE_SIZE))) swsusp_header;
 123
 124 static struct swsusp_info swsusp_info;
 125
 126 /*
 127  * XXX: We try to keep some more pages free so that I/O operations succeed
 128  * without paging. Might this be more?
 129  */
 130 #define PAGES_FOR_IO    512
 131
 132 /*
 133  * Saving part...
 134  */
 135
 136 /* We memorize in swapfile_used what swap devices are used for suspension */
 137 #define SWAPFILE_UNUSED    0
 138 #define SWAPFILE_SUSPEND   1    /* This is the suspending device */
 139 #define SWAPFILE_IGNORED   2    /* Those are other swap devices ignored for suspension */
 140
 141 static unsigned short swapfile_used[MAX_SWAPFILES];
 142 static unsigned short root_swap;
 143
 144 static int write_page(unsigned long addr, swp_entry_t * loc);
 145 static int bio_read_page(pgoff_t page_off, void * page);
 146
 147 static u8 key_iv[MAXKEY+MAXIV];
 148
 149 #ifdef CONFIG_SWSUSP_ENCRYPT
 150
 151 static int crypto_init(int mode, void **mem)
 152 {
 153         int error = 0;
 154         int len;
 155         char *modemsg;
 156         struct crypto_tfm *tfm;
 157
 158         modemsg = mode ? "suspend not possible" : "resume not possible";
 159
 160         tfm = crypto_alloc_tfm(CIPHER, CRYPTO_TFM_MODE_CBC);
 161         if(!tfm) {
 162                 printk(KERN_ERR "swsusp: no tfm, %s\n", modemsg);
 163                 error = -EINVAL;
 164                 goto out;
 165         }
 166
 167         if(MAXKEY < crypto_tfm_alg_min_keysize(tfm)) {
 168                 printk(KERN_ERR "swsusp: key buffer too small, %s\n", modemsg);
 169                 error = -ENOKEY;
 170                 goto fail;
 171         }
 172
 173         if (mode)
 174                 get_random_bytes(key_iv, MAXKEY+MAXIV);
 175
 176         len = crypto_tfm_alg_max_keysize(tfm);
 177         if (len > MAXKEY)
 178                 len = MAXKEY;
 179
 180         if (crypto_cipher_setkey(tfm, key_iv, len)) {
 181                 printk(KERN_ERR "swsusp: key setup failure, %s\n", modemsg);
 182                 error = -EKEYREJECTED;
 183                 goto fail;
 184         }
 185
 186         len = crypto_tfm_alg_ivsize(tfm);
 187
 188         if (MAXIV < len) {
 189                 printk(KERN_ERR "swsusp: iv buffer too small, %s\n", modemsg);
 190                 error = -EOVERFLOW;
 191                 goto fail;
 192         }
 193
 194         crypto_cipher_set_iv(tfm, key_iv+MAXKEY, len);
 195
 196         *mem=(void *)tfm;
 197
 198         goto out;
 199
 200 fail:   crypto_free_tfm(tfm);
 201 out:    return error;
 202 }
 203
 204 static __inline__ void crypto_exit(void *mem)
 205 {
 206         crypto_free_tfm((struct crypto_tfm *)mem);
 207 }
 208
 209 static __inline__ int crypto_write(struct pbe *p, void *mem)
 210 {
 211         int error = 0;
 212         struct scatterlist src, dst;
 213
 214         src.page   = virt_to_page(p->address);
 215         src.offset = 0;
 216         src.length = PAGE_SIZE;
 217         dst.page   = virt_to_page((void *)&swsusp_header);
 218         dst.offset = 0;
 219         dst.length = PAGE_SIZE;
 220
 221         error = crypto_cipher_encrypt((struct crypto_tfm *)mem, &dst, &src,
 222                                         PAGE_SIZE);
 223
 224         if (!error)
 225                 error = write_page((unsigned long)&swsusp_header,
 226                                 &(p->swap_address));
 227         return error;
 228 }
 229
 230 static __inline__ int crypto_read(struct pbe *p, void *mem)
 231 {
 232         int error = 0;
 233         struct scatterlist src, dst;
 234
 235         error = bio_read_page(swp_offset(p->swap_address), (void *)p->address);
 236         if (!error) {
 237                 src.offset = 0;
 238                 src.length = PAGE_SIZE;
 239                 dst.offset = 0;
 240                 dst.length = PAGE_SIZE;
 241                 src.page = dst.page = virt_to_page((void *)p->address);
 242
 243                 error = crypto_cipher_decrypt((struct crypto_tfm *)mem, &dst,
 244                                                 &src, PAGE_SIZE);
 245         }
 246         return error;
 247 }
 248 #else
 249 static __inline__ int crypto_init(int mode, void *mem)
 250 {
 251         return 0;
 252 }
 253
 254 static __inline__ void crypto_exit(void *mem)
 255 {
 256 }
 257
 258 static __inline__ int crypto_write(struct pbe *p, void *mem)
 259 {
 260         return write_page(p->address, &(p->swap_address));
 261 }
 262
 263 static __inline__ int crypto_read(struct pbe *p, void *mem)
 264 {
 265         return bio_read_page(swp_offset(p->swap_address), (void *)p->address);
 266 }
 267 #endif
 268
 269 static int mark_swapfiles(swp_entry_t prev)
 270 {
 271         int error;
 272
 273         rw_swap_page_sync(READ,
 274                           swp_entry(root_swap, 0),
 275                           virt_to_page((unsigned long)&swsusp_header));
 276         if (!memcmp("SWAP-SPACE",swsusp_header.sig, 10) ||
 277             !memcmp("SWAPSPACE2",swsusp_header.sig, 10)) {
 278                 memcpy(swsusp_header.orig_sig,swsusp_header.sig, 10);
 279                 memcpy(swsusp_header.sig,SWSUSP_SIG, 10);
 280                 memcpy(swsusp_header.key_iv, key_iv, MAXKEY+MAXIV);
 281                 swsusp_header.swsusp_info = prev;
 282                 error = rw_swap_page_sync(WRITE,
 283                                           swp_entry(root_swap, 0),
 284                                           virt_to_page((unsigned long)
 285                                                        &swsusp_header));
 286         } else {
 287                 pr_debug("swsusp: Partition is not swap space.\n");
 288                 error = -ENODEV;
 289         }
 290         return error;
 291 }
 292
 293 /*
 294  * Check whether the swap device is the specified resume
 295  * device, irrespective of whether they are specified by
 296  * identical names.
 297  *
 298  * (Thus, device inode aliasing is allowed.  You can say /dev/hda4
 299  * instead of /dev/ide/host0/bus0/target0/lun0/part4 [if using devfs]
 300  * and they'll be considered the same device.  This is *necessary* for
 301  * devfs, since the resume code can only recognize the form /dev/hda4,
 302  * but the suspend code would see the long name.)
 303  */
 304 static int is_resume_device(const struct swap_info_struct *swap_info)
 305 {
 306         struct file *file = swap_info->swap_file;
 307         struct inode *inode = file->f_dentry->d_inode;
 308
 309         return S_ISBLK(inode->i_mode) &&
 310                 swsusp_resume_device == MKDEV(imajor(inode), iminor(inode));
 311 }
 312
 313 static int swsusp_swap_check(void) /* This is called before saving image */
 314 {
 315         int i, len;
 316
 317         len=strlen(resume_file);
 318         root_swap = 0xFFFF;
 319
 320         spin_lock(&swap_lock);
 321         for (i=0; i<MAX_SWAPFILES; i++) {
 322                 if (!(swap_info[i].flags & SWP_WRITEOK)) {
 323                         swapfile_used[i]=SWAPFILE_UNUSED;
 324                 } else {
 325                         if (!len) {
 326                                 printk(KERN_WARNING "resume= option should be used to set suspend device" );
 327                                 if (root_swap == 0xFFFF) {
 328                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 329                                         root_swap = i;
 330                                 } else
 331                                         swapfile_used[i] = SWAPFILE_IGNORED;
 332                         } else {
 333                                 /* we ignore all swap devices that are not the resume_file */
 334                                 if (is_resume_device(&swap_info[i])) {
 335                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 336                                         root_swap = i;
 337                                 } else {
 338                                         swapfile_used[i] = SWAPFILE_IGNORED;
 339                                 }
 340                         }
 341                 }
 342         }
 343         spin_unlock(&swap_lock);
 344         return (root_swap != 0xffff) ? 0 : -ENODEV;
 345 }
 346
 347 /**
 348  * This is called after saving image so modification
 349  * will be lost after resume... and that's what we want.
 350  * we make the device unusable. A new call to
 351  * lock_swapdevices can unlock the devices.
 352  */
 353 static void lock_swapdevices(void)
 354 {
 355         int i;
 356
 357         spin_lock(&swap_lock);
 358         for (i = 0; i< MAX_SWAPFILES; i++)
 359                 if (swapfile_used[i] == SWAPFILE_IGNORED) {
 360                         swap_info[i].flags ^= SWP_WRITEOK;
 361                 }
 362         spin_unlock(&swap_lock);
 363 }
 364
 365 /**
 366  *      write_page - Write one page to a fresh swap location.
 367  *      @addr:  Address we're writing.
 368  *      @loc:   Place to store the entry we used.
 369  *
 370  *      Allocate a new swap entry and 'sync' it. Note we discard -EIO
 371  *      errors. That is an artifact left over from swsusp. It did not
 372  *      check the return of rw_swap_page_sync() at all, since most pages
 373  *      written back to swap would return -EIO.
 374  *      This is a partial improvement, since we will at least return other
 375  *      errors, though we need to eventually fix the damn code.
 376  */
 377 static int write_page(unsigned long addr, swp_entry_t * loc)
 378 {
 379         swp_entry_t entry;
 380         int error = 0;
 381
 382         entry = get_swap_page();
 383         if (swp_offset(entry) &&
 384             swapfile_used[swp_type(entry)] == SWAPFILE_SUSPEND) {
 385                 error = rw_swap_page_sync(WRITE, entry,
 386                                           virt_to_page(addr));
 387                 if (error == -EIO)
 388                         error = 0;
 389                 if (!error)
 390                         *loc = entry;
 391         } else
 392                 error = -ENOSPC;
 393         return error;
 394 }
 395
 396 /**
 397  *      data_free - Free the swap entries used by the saved image.
 398  *
 399  *      Walk the list of used swap entries and free each one.
 400  *      This is only used for cleanup when suspend fails.
 401  */
 402 static void data_free(void)
 403 {
 404         swp_entry_t entry;
 405         int i;
 406
 407         for (i = 0; i < nr_copy_pages; i++) {
 408                 entry = (pagedir_nosave + i)->swap_address;
 409                 if (entry.val)
 410                         swap_free(entry);
 411                 else
 412                         break;
 413                 (pagedir_nosave + i)->swap_address = (swp_entry_t){0};
 414         }
 415 }
 416
 417 /**
 418  *      data_write - Write saved image to swap.
 419  *
 420  *      Walk the list of pages in the image and sync each one to swap.
 421  */
 422 static int data_write(void)
 423 {
 424         int error = 0, i = 0;
 425         unsigned int mod = nr_copy_pages / 100;
 426         struct pbe *p;
 427         void *tfm;
 428
 429         if ((error = crypto_init(1, &tfm)))
 430                 return error;
 431
 432         if (!mod)
 433                 mod = 1;
 434
 435         printk( "Writing data to swap (%d pages)...     ", nr_copy_pages );
 436         for_each_pbe (p, pagedir_nosave) {
 437                 if (!(i%mod))
 438                         printk( "\b\b\b\b%3d%%", i / mod );
 439                 if ((error = crypto_write(p, tfm))) {
 440                         crypto_exit(tfm);
 441                         return error;
 442                 }
 443                 i++;
 444         }
 445         printk("\b\b\b\bdone\n");
 446         crypto_exit(tfm);
 447         return error;
 448 }
 449
 450 static void dump_info(void)
 451 {
 452         pr_debug(" swsusp: Version: %u\n",swsusp_info.version_code);
 453         pr_debug(" swsusp: Num Pages: %ld\n",swsusp_info.num_physpages);
 454         pr_debug(" swsusp: UTS Sys: %s\n",swsusp_info.uts.sysname);
 455         pr_debug(" swsusp: UTS Node: %s\n",swsusp_info.uts.nodename);
 456         pr_debug(" swsusp: UTS Release: %s\n",swsusp_info.uts.release);
 457         pr_debug(" swsusp: UTS Version: %s\n",swsusp_info.uts.version);
 458         pr_debug(" swsusp: UTS Machine: %s\n",swsusp_info.uts.machine);
 459         pr_debug(" swsusp: UTS Domain: %s\n",swsusp_info.uts.domainname);
 460         pr_debug(" swsusp: CPUs: %d\n",swsusp_info.cpus);
 461         pr_debug(" swsusp: Image: %ld Pages\n",swsusp_info.image_pages);
 462         pr_debug(" swsusp: Pagedir: %ld Pages\n",swsusp_info.pagedir_pages);
 463 }
 464
 465 static void init_header(void)
 466 {
 467         memset(&swsusp_info, 0, sizeof(swsusp_info));
 468         swsusp_info.version_code = LINUX_VERSION_CODE;
 469         swsusp_info.num_physpages = num_physpages;
 470         memcpy(&swsusp_info.uts, &system_utsname, sizeof(system_utsname));
 471
 472         swsusp_info.suspend_pagedir = pagedir_nosave;
 473         swsusp_info.cpus = num_online_cpus();
 474         swsusp_info.image_pages = nr_copy_pages;
 475 }
 476
 477 static int close_swap(void)
 478 {
 479         swp_entry_t entry;
 480         int error;
 481
 482         dump_info();
 483         error = write_page((unsigned long)&swsusp_info, &entry);
 484         if (!error) {
 485                 printk( "S" );
 486                 error = mark_swapfiles(entry);
 487                 printk( "|\n" );
 488         }
 489         return error;
 490 }
 491
 492 /**
 493  *      free_pagedir_entries - Free pages used by the page directory.
 494  *
 495  *      This is used during suspend for error recovery.
 496  */
 497
 498 static void free_pagedir_entries(void)
 499 {
 500         int i;
 501
 502         for (i = 0; i < swsusp_info.pagedir_pages; i++)
 503                 swap_free(swsusp_info.pagedir[i]);
 504 }
 505
 506
 507 /**
 508  *      write_pagedir - Write the array of pages holding the page directory.
 509  *      @last:  Last swap entry we write (needed for header).
 510  */
 511
 512 static int write_pagedir(void)
 513 {
 514         int error = 0;
 515         unsigned n = 0;
 516         struct pbe * pbe;
 517
 518         printk( "Writing pagedir...");
 519         for_each_pb_page (pbe, pagedir_nosave) {
 520                 if ((error = write_page((unsigned long)pbe, &swsusp_info.pagedir[n++])))
 521                         return error;
 522         }
 523
 524         swsusp_info.pagedir_pages = n;
 525         printk("done (%u pages)\n", n);
 526         return error;
 527 }
 528
 529 /**
 530  *      write_suspend_image - Write entire image and metadata.
 531  *
 532  */
 533 static int write_suspend_image(void)
 534 {
 535         int error;
 536
 537         init_header();
 538         if ((error = data_write()))
 539                 goto FreeData;
 540
 541         if ((error = write_pagedir()))
 542                 goto FreePagedir;
 543
 544         if ((error = close_swap()))
 545                 goto FreePagedir;
 546  Done:
 547         memset(key_iv, 0, MAXKEY+MAXIV);
 548         return error;
 549  FreePagedir:
 550         free_pagedir_entries();
 551  FreeData:
 552         data_free();
 553         goto Done;
 554 }
 555
 556
 557 #ifdef CONFIG_HIGHMEM
 558 struct highmem_page {
 559         char *data;
 560         struct page *page;
 561         struct highmem_page *next;
 562 };
 563
 564 static struct highmem_page *highmem_copy;
 565
 566 static int save_highmem_zone(struct zone *zone)
 567 {
 568         unsigned long zone_pfn;
 569         mark_free_pages(zone);
 570         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 571                 struct page *page;
 572                 struct highmem_page *save;
 573                 void *kaddr;
 574                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
 575
 576                 if (!(pfn%1000))
 577                         printk(".");
 578                 if (!pfn_valid(pfn))
 579                         continue;
 580                 page = pfn_to_page(pfn);
 581                 /*
 582                  * This condition results from rvmalloc() sans vmalloc_32()
 583                  * and architectural memory reservations. This should be
 584                  * corrected eventually when the cases giving rise to this
 585                  * are better understood.
 586                  */
 587                 if (PageReserved(page)) {
 588                         printk("highmem reserved page?!\n");
 589                         continue;
 590                 }
 591                 BUG_ON(PageNosave(page));
 592                 if (PageNosaveFree(page))
 593                         continue;
 594                 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
 595                 if (!save)
 596                         return -ENOMEM;
 597                 save->next = highmem_copy;
 598                 save->page = page;
 599                 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
 600                 if (!save->data) {
 601                         kfree(save);
 602                         return -ENOMEM;
 603                 }
 604                 kaddr = kmap_atomic(page, KM_USER0);
 605                 memcpy(save->data, kaddr, PAGE_SIZE);
 606                 kunmap_atomic(kaddr, KM_USER0);
 607                 highmem_copy = save;
 608         }
 609         return 0;
 610 }
 611 #endif /* CONFIG_HIGHMEM */
 612
 613
 614 static int save_highmem(void)
 615 {
 616 #ifdef CONFIG_HIGHMEM
 617         struct zone *zone;
 618         int res = 0;
 619
 620         pr_debug("swsusp: Saving Highmem\n");
 621         for_each_zone (zone) {
 622                 if (is_highmem(zone))
 623                         res = save_highmem_zone(zone);
 624                 if (res)
 625                         return res;
 626         }
 627 #endif
 628         return 0;
 629 }
 630
 631 static int restore_highmem(void)
 632 {
 633 #ifdef CONFIG_HIGHMEM
 634         printk("swsusp: Restoring Highmem\n");
 635         while (highmem_copy) {
 636                 struct highmem_page *save = highmem_copy;
 637                 void *kaddr;
 638                 highmem_copy = save->next;
 639
 640                 kaddr = kmap_atomic(save->page, KM_USER0);
 641                 memcpy(kaddr, save->data, PAGE_SIZE);
 642                 kunmap_atomic(kaddr, KM_USER0);
 643                 free_page((long) save->data);
 644                 kfree(save);
 645         }
 646 #endif
 647         return 0;
 648 }
 649
 650
 651 static int pfn_is_nosave(unsigned long pfn)
 652 {
 653         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 654         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 655         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 656 }
 657
 658 /**
 659  *      saveable - Determine whether a page should be cloned or not.
 660  *      @pfn:   The page
 661  *
 662  *      We save a page if it's Reserved, and not in the range of pages
 663  *      statically defined as 'unsaveable', or if it isn't reserved, and
 664  *      isn't part of a free chunk of pages.
 665  */
 666
 667 static int saveable(struct zone * zone, unsigned long * zone_pfn)
 668 {
 669         unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
 670         struct page * page;
 671
 672         if (!pfn_valid(pfn))
 673                 return 0;
 674
 675         page = pfn_to_page(pfn);
 676         BUG_ON(PageReserved(page) && PageNosave(page));
 677         if (PageNosave(page))
 678                 return 0;
 679         if (PageReserved(page) && pfn_is_nosave(pfn)) {
 680                 pr_debug("[nosave pfn 0x%lx]", pfn);
 681                 return 0;
 682         }
 683         if (PageNosaveFree(page))
 684                 return 0;
 685
 686         return 1;
 687 }
 688
 689 static void count_data_pages(void)
 690 {
 691         struct zone *zone;
 692         unsigned long zone_pfn;
 693
 694         nr_copy_pages = 0;
 695
 696         for_each_zone (zone) {
 697                 if (is_highmem(zone))
 698                         continue;
 699                 mark_free_pages(zone);
 700                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 701                         nr_copy_pages += saveable(zone, &zone_pfn);
 702         }
 703 }
 704
 705
 706 static void copy_data_pages(void)
 707 {
 708         struct zone *zone;
 709         unsigned long zone_pfn;
 710         struct pbe * pbe = pagedir_nosave;
 711
 712         pr_debug("copy_data_pages(): pages to copy: %d\n", nr_copy_pages);
 713         for_each_zone (zone) {
 714                 if (is_highmem(zone))
 715                         continue;
 716                 mark_free_pages(zone);
 717                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 718                         if (saveable(zone, &zone_pfn)) {
 719                                 struct page * page;
 720                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 721                                 BUG_ON(!pbe);
 722                                 pbe->orig_address = (long) page_address(page);
 723                                 /* copy_page is not usable for copying task structs. */
 724                                 memcpy((void *)pbe->address, (void *)pbe->orig_address, PAGE_SIZE);
 725                                 pbe = pbe->next;
 726                         }
 727                 }
 728         }
 729         BUG_ON(pbe);
 730 }
 731
 732
 733 /**
 734  *      calc_nr - Determine the number of pages needed for a pbe list.
 735  */
 736
 737 static int calc_nr(int nr_copy)
 738 {
 739         return nr_copy + (nr_copy+PBES_PER_PAGE-2)/(PBES_PER_PAGE-1);
 740 }
 741
 742 /**
 743  *      free_pagedir - free pages allocated with alloc_pagedir()
 744  */
 745
 746 static inline void free_pagedir(struct pbe *pblist)
 747 {
 748         struct pbe *pbe;
 749
 750         while (pblist) {
 751                 pbe = (pblist + PB_PAGE_SKIP)->next;
 752                 free_page((unsigned long)pblist);
 753                 pblist = pbe;
 754         }
 755 }
 756
 757 /**
 758  *      fill_pb_page - Create a list of PBEs on a given memory page
 759  */
 760
 761 static inline void fill_pb_page(struct pbe *pbpage)
 762 {
 763         struct pbe *p;
 764
 765         p = pbpage;
 766         pbpage += PB_PAGE_SKIP;
 767         do
 768                 p->next = p + 1;
 769         while (++p < pbpage);
 770 }
 771
 772 /**
 773  *      create_pbe_list - Create a list of PBEs on top of a given chain
 774  *      of memory pages allocated with alloc_pagedir()
 775  */
 776
 777 static void create_pbe_list(struct pbe *pblist, unsigned nr_pages)
 778 {
 779         struct pbe *pbpage, *p;
 780         unsigned num = PBES_PER_PAGE;
 781
 782         for_each_pb_page (pbpage, pblist) {
 783                 if (num >= nr_pages)
 784                         break;
 785
 786                 fill_pb_page(pbpage);
 787                 num += PBES_PER_PAGE;
 788         }
 789         if (pbpage) {
 790                 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
 791                         p->next = p + 1;
 792                 p->next = NULL;
 793         }
 794         pr_debug("create_pbe_list(): initialized %d PBEs\n", num);
 795 }
 796
 797 /**
 798  *      alloc_pagedir - Allocate the page directory.
 799  *
 800  *      First, determine exactly how many pages we need and
 801  *      allocate them.
 802  *
 803  *      We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
 804  *      struct pbe elements (pbes) and the last element in the page points
 805  *      to the next page.
 806  *
 807  *      On each page we set up a list of struct_pbe elements.
 808  */
 809
 810 static struct pbe * alloc_pagedir(unsigned nr_pages)
 811 {
 812         unsigned num;
 813         struct pbe *pblist, *pbe;
 814
 815         if (!nr_pages)
 816                 return NULL;
 817
 818         pr_debug("alloc_pagedir(): nr_pages = %d\n", nr_pages);
 819         pblist = (struct pbe *)get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
 820         for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
 821                         pbe = pbe->next, num += PBES_PER_PAGE) {
 822                 pbe += PB_PAGE_SKIP;
 823                 pbe->next = (struct pbe *)get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
 824         }
 825         if (!pbe) { /* get_zeroed_page() failed */
 826                 free_pagedir(pblist);
 827                 pblist = NULL;
 828         }
 829         return pblist;
 830 }
 831
 832 /**
 833  *      free_image_pages - Free pages allocated for snapshot
 834  */
 835
 836 static void free_image_pages(void)
 837 {
 838         struct pbe * p;
 839
 840         for_each_pbe (p, pagedir_save) {
 841                 if (p->address) {
 842                         ClearPageNosave(virt_to_page(p->address));
 843                         free_page(p->address);
 844                         p->address = 0;
 845                 }
 846         }
 847 }
 848
 849 /**
 850  *      alloc_image_pages - Allocate pages for the snapshot.
 851  */
 852
 853 static int alloc_image_pages(void)
 854 {
 855         struct pbe * p;
 856
 857         for_each_pbe (p, pagedir_save) {
 858                 p->address = get_zeroed_page(GFP_ATOMIC | __GFP_COLD);
 859                 if (!p->address)
 860                         return -ENOMEM;
 861                 SetPageNosave(virt_to_page(p->address));
 862         }
 863         return 0;
 864 }
 865
 866 /* Free pages we allocated for suspend. Suspend pages are alocated
 867  * before atomic copy, so we need to free them after resume.
 868  */
 869 void swsusp_free(void)
 870 {
 871         BUG_ON(PageNosave(virt_to_page(pagedir_save)));
 872         BUG_ON(PageNosaveFree(virt_to_page(pagedir_save)));
 873         free_image_pages();
 874         free_pagedir(pagedir_save);
 875 }
 876
 877
 878 /**
 879  *      enough_free_mem - Make sure we enough free memory to snapshot.
 880  *
 881  *      Returns TRUE or FALSE after checking the number of available
 882  *      free pages.
 883  */
 884
 885 static int enough_free_mem(void)
 886 {
 887         if (nr_free_pages() < (nr_copy_pages + PAGES_FOR_IO)) {
 888                 pr_debug("swsusp: Not enough free pages: Have %d\n",
 889                          nr_free_pages());
 890                 return 0;
 891         }
 892         return 1;
 893 }
 894
 895
 896 /**
 897  *      enough_swap - Make sure we have enough swap to save the image.
 898  *
 899  *      Returns TRUE or FALSE after checking the total amount of swap
 900  *      space avaiable.
 901  *
 902  *      FIXME: si_swapinfo(&i) returns all swap devices information.
 903  *      We should only consider resume_device.
 904  */
 905
 906 static int enough_swap(void)
 907 {
 908         struct sysinfo i;
 909
 910         si_swapinfo(&i);
 911         if (i.freeswap < (nr_copy_pages + PAGES_FOR_IO))  {
 912                 pr_debug("swsusp: Not enough swap. Need %ld\n",i.freeswap);
 913                 return 0;
 914         }
 915         return 1;
 916 }
 917
 918 static int swsusp_alloc(void)
 919 {
 920         int error;
 921
 922         pagedir_nosave = NULL;
 923         nr_copy_pages = calc_nr(nr_copy_pages);
 924         nr_copy_pages_check = nr_copy_pages;
 925
 926         pr_debug("suspend: (pages needed: %d + %d free: %d)\n",
 927                  nr_copy_pages, PAGES_FOR_IO, nr_free_pages());
 928
 929         if (!enough_free_mem())
 930                 return -ENOMEM;
 931
 932         if (!enough_swap())
 933                 return -ENOSPC;
 934
 935         if (!(pagedir_save = alloc_pagedir(nr_copy_pages))) {
 936                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 937                 return -ENOMEM;
 938         }
 939         create_pbe_list(pagedir_save, nr_copy_pages);
 940         pagedir_nosave = pagedir_save;
 941         if ((error = alloc_image_pages())) {
 942                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 943                 swsusp_free();
 944                 return error;
 945         }
 946
 947         return 0;
 948 }
 949
 950 static int suspend_prepare_image(void)
 951 {
 952         int error;
 953
 954         pr_debug("swsusp: critical section: \n");
 955         if (save_highmem()) {
 956                 printk(KERN_CRIT "Suspend machine: Not enough free pages for highmem\n");
 957                 restore_highmem();
 958                 return -ENOMEM;
 959         }
 960
 961         drain_local_pages();
 962         count_data_pages();
 963         printk("swsusp: Need to copy %u pages\n", nr_copy_pages);
 964
 965         error = swsusp_alloc();
 966         if (error)
 967                 return error;
 968
 969         /* During allocating of suspend pagedir, new cold pages may appear.
 970          * Kill them.
 971          */
 972         drain_local_pages();
 973         copy_data_pages();
 974
 975         /*
 976          * End of critical section. From now on, we can write to memory,
 977          * but we should not touch disk. This specially means we must _not_
 978          * touch swap space! Except we must write out our image of course.
 979          */
 980
 981         printk("swsusp: critical section/: done (%d pages copied)\n", nr_copy_pages );
 982         return 0;
 983 }
 984
 985
 986 /* It is important _NOT_ to umount filesystems at this point. We want
 987  * them synced (in case something goes wrong) but we DO not want to mark
 988  * filesystem clean: it is not. (And it does not matter, if we resume
 989  * correctly, we'll mark system clean, anyway.)
 990  */
 991 int swsusp_write(void)
 992 {
 993         int error;
 994         device_resume();
 995         lock_swapdevices();
 996         error = write_suspend_image();
 997         /* This will unlock ignored swap devices since writing is finished */
 998         lock_swapdevices();
 999         return error;
1000
1001 }
1002
1003
1004 extern asmlinkage int swsusp_arch_suspend(void);
1005 extern asmlinkage int swsusp_arch_resume(void);
1006
1007
1008 asmlinkage int swsusp_save(void)
1009 {
1010         return suspend_prepare_image();
1011 }
1012
1013 int swsusp_suspend(void)
1014 {
1015         int error;
1016         if ((error = arch_prepare_suspend()))
1017                 return error;
1018         local_irq_disable();
1019         /* At this point, device_suspend() has been called, but *not*
1020          * device_power_down(). We *must* device_power_down() now.
1021          * Otherwise, drivers for some devices (e.g. interrupt controllers)
1022          * become desynchronized with the actual state of the hardware
1023          * at resume time, and evil weirdness ensues.
1024          */
1025         if ((error = device_power_down(PMSG_FREEZE))) {
1026                 printk(KERN_ERR "Some devices failed to power down, aborting suspend\n");
1027                 local_irq_enable();
1028                 return error;
1029         }
1030
1031         if ((error = swsusp_swap_check())) {
1032                 printk(KERN_ERR "swsusp: cannot find swap device, try swapon -a.\n");
1033                 device_power_up();
1034                 local_irq_enable();
1035                 return error;
1036         }
1037
1038         save_processor_state();
1039         if ((error = swsusp_arch_suspend()))
1040                 printk(KERN_ERR "Error %d suspending\n", error);
1041         /* Restore control flow magically appears here */
1042         restore_processor_state();
1043         BUG_ON (nr_copy_pages_check != nr_copy_pages);
1044         restore_highmem();
1045         device_power_up();
1046         local_irq_enable();
1047         return error;
1048 }
1049
1050 int swsusp_resume(void)
1051 {
1052         int error;
1053         local_irq_disable();
1054         if (device_power_down(PMSG_FREEZE))
1055                 printk(KERN_ERR "Some devices failed to power down, very bad\n");
1056         /* We'll ignore saved state, but this gets preempt count (etc) right */
1057         save_processor_state();
1058         error = swsusp_arch_resume();
1059         /* Code below is only ever reached in case of failure. Otherwise
1060          * execution continues at place where swsusp_arch_suspend was called
1061          */
1062         BUG_ON(!error);
1063         restore_processor_state();
1064         restore_highmem();
1065         touch_softlockup_watchdog();
1066         device_power_up();
1067         local_irq_enable();
1068         return error;
1069 }
1070
1071 /**
1072  *      On resume, for storing the PBE list and the image,
1073  *      we can only use memory pages that do not conflict with the pages
1074  *      which had been used before suspend.
1075  *
1076  *      We don't know which pages are usable until we allocate them.
1077  *
1078  *      Allocated but unusable (ie eaten) memory pages are linked together
1079  *      to create a list, so that we can free them easily
1080  *
1081  *      We could have used a type other than (void *)
1082  *      for this purpose, but ...
1083  */
1084 static void **eaten_memory = NULL;
1085
1086 static inline void eat_page(void *page)
1087 {
1088         void **c;
1089
1090         c = eaten_memory;
1091         eaten_memory = page;
1092         *eaten_memory = c;
1093 }
1094
1095 static unsigned long get_usable_page(unsigned gfp_mask)
1096 {
1097         unsigned long m;
1098
1099         m = get_zeroed_page(gfp_mask);
1100         while (!PageNosaveFree(virt_to_page(m))) {
1101                 eat_page((void *)m);
1102                 m = get_zeroed_page(gfp_mask);
1103                 if (!m)
1104                         break;
1105         }
1106         return m;
1107 }
1108
1109 static void free_eaten_memory(void)
1110 {
1111         unsigned long m;
1112         void **c;
1113         int i = 0;
1114
1115         c = eaten_memory;
1116         while (c) {
1117                 m = (unsigned long)c;
1118                 c = *c;
1119                 free_page(m);
1120                 i++;
1121         }
1122         eaten_memory = NULL;
1123         pr_debug("swsusp: %d unused pages freed\n", i);
1124 }
1125
1126 /**
1127  *      check_pagedir - We ensure here that pages that the PBEs point to
1128  *      won't collide with pages where we're going to restore from the loaded
1129  *      pages later
1130  */
1131
1132 static int check_pagedir(struct pbe *pblist)
1133 {
1134         struct pbe *p;
1135
1136         /* This is necessary, so that we can free allocated pages
1137          * in case of failure
1138          */
1139         for_each_pbe (p, pblist)
1140                 p->address = 0UL;
1141
1142         for_each_pbe (p, pblist) {
1143                 p->address = get_usable_page(GFP_ATOMIC);
1144                 if (!p->address)
1145                         return -ENOMEM;
1146         }
1147         return 0;
1148 }
1149
1150 /**
1151  *      swsusp_pagedir_relocate - It is possible, that some memory pages
1152  *      occupied by the list of PBEs collide with pages where we're going to
1153  *      restore from the loaded pages later.  We relocate them here.
1154  */
1155
1156 static struct pbe * swsusp_pagedir_relocate(struct pbe *pblist)
1157 {
1158         struct zone *zone;
1159         unsigned long zone_pfn;
1160         struct pbe *pbpage, *tail, *p;
1161         void *m;
1162         int rel = 0, error = 0;
1163
1164         if (!pblist) /* a sanity check */
1165                 return NULL;
1166
1167         pr_debug("swsusp: Relocating pagedir (%lu pages to check)\n",
1168                         swsusp_info.pagedir_pages);
1169
1170         /* Set page flags */
1171
1172         for_each_zone (zone) {
1173                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
1174                         SetPageNosaveFree(pfn_to_page(zone_pfn +
1175                                         zone->zone_start_pfn));
1176         }
1177
1178         /* Clear orig addresses */
1179
1180         for_each_pbe (p, pblist)
1181                 ClearPageNosaveFree(virt_to_page(p->orig_address));
1182
1183         tail = pblist + PB_PAGE_SKIP;
1184
1185         /* Relocate colliding pages */
1186
1187         for_each_pb_page (pbpage, pblist) {
1188                 if (!PageNosaveFree(virt_to_page((unsigned long)pbpage))) {
1189                         m = (void *)get_usable_page(GFP_ATOMIC | __GFP_COLD);
1190                         if (!m) {
1191                                 error = -ENOMEM;
1192                                 break;
1193                         }
1194                         memcpy(m, (void *)pbpage, PAGE_SIZE);
1195                         if (pbpage == pblist)
1196                                 pblist = (struct pbe *)m;
1197                         else
1198                                 tail->next = (struct pbe *)m;
1199
1200                         eat_page((void *)pbpage);
1201                         pbpage = (struct pbe *)m;
1202
1203                         /* We have to link the PBEs again */
1204
1205                         for (p = pbpage; p < pbpage + PB_PAGE_SKIP; p++)
1206                                 if (p->next) /* needed to save the end */
1207                                         p->next = p + 1;
1208
1209                         rel++;
1210                 }
1211                 tail = pbpage + PB_PAGE_SKIP;
1212         }
1213
1214         if (error) {
1215                 printk("\nswsusp: Out of memory\n\n");
1216                 free_pagedir(pblist);
1217                 free_eaten_memory();
1218                 pblist = NULL;
1219                 /* Is this even worth handling? It should never ever happen, and we
1220                    have just lost user's state, anyway... */
1221         } else
1222                 printk("swsusp: Relocated %d pages\n", rel);
1223
1224         return pblist;
1225 }
1226
1227 /*
1228  *      Using bio to read from swap.
1229  *      This code requires a bit more work than just using buffer heads
1230  *      but, it is the recommended way for 2.5/2.6.
1231  *      The following are to signal the beginning and end of I/O. Bios
1232  *      finish asynchronously, while we want them to happen synchronously.
1233  *      A simple atomic_t, and a wait loop take care of this problem.
1234  */
1235
1236 static atomic_t io_done = ATOMIC_INIT(0);
1237
1238 static int end_io(struct bio * bio, unsigned int num, int err)
1239 {
1240         if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1241                 panic("I/O error reading memory image");
1242         atomic_set(&io_done, 0);
1243         return 0;
1244 }
1245
1246 static struct block_device * resume_bdev;
1247
1248 /**
1249  *      submit - submit BIO request.
1250  *      @rw:    READ or WRITE.
1251  *      @off    physical offset of page.
1252  *      @page:  page we're reading or writing.
1253  *
1254  *      Straight from the textbook - allocate and initialize the bio.
1255  *      If we're writing, make sure the page is marked as dirty.
1256  *      Then submit it and wait.
1257  */
1258
1259 static int submit(int rw, pgoff_t page_off, void * page)
1260 {
1261         int error = 0;
1262         struct bio * bio;
1263
1264         bio = bio_alloc(GFP_ATOMIC, 1);
1265         if (!bio)
1266                 return -ENOMEM;
1267         bio->bi_sector = page_off * (PAGE_SIZE >> 9);
1268         bio_get(bio);
1269         bio->bi_bdev = resume_bdev;
1270         bio->bi_end_io = end_io;
1271
1272         if (bio_add_page(bio, virt_to_page(page), PAGE_SIZE, 0) < PAGE_SIZE) {
1273                 printk("swsusp: ERROR: adding page to bio at %ld\n",page_off);
1274                 error = -EFAULT;
1275                 goto Done;
1276         }
1277
1278         if (rw == WRITE)
1279                 bio_set_pages_dirty(bio);
1280
1281         atomic_set(&io_done, 1);
1282         submit_bio(rw | (1 << BIO_RW_SYNC), bio);
1283         while (atomic_read(&io_done))
1284                 yield();
1285
1286  Done:
1287         bio_put(bio);
1288         return error;
1289 }
1290
1291 static int bio_read_page(pgoff_t page_off, void * page)
1292 {
1293         return submit(READ, page_off, page);
1294 }
1295
1296 static int bio_write_page(pgoff_t page_off, void * page)
1297 {
1298         return submit(WRITE, page_off, page);
1299 }
1300
1301 /*
1302  * Sanity check if this image makes sense with this kernel/swap context
1303  * I really don't think that it's foolproof but more than nothing..
1304  */
1305
1306 static const char * sanity_check(void)
1307 {
1308         dump_info();
1309         if (swsusp_info.version_code != LINUX_VERSION_CODE)
1310                 return "kernel version";
1311         if (swsusp_info.num_physpages != num_physpages)
1312                 return "memory size";
1313         if (strcmp(swsusp_info.uts.sysname,system_utsname.sysname))
1314                 return "system type";
1315         if (strcmp(swsusp_info.uts.release,system_utsname.release))
1316                 return "kernel release";
1317         if (strcmp(swsusp_info.uts.version,system_utsname.version))
1318                 return "version";
1319         if (strcmp(swsusp_info.uts.machine,system_utsname.machine))
1320                 return "machine";
1321 #if 0
1322         /* We can't use number of online CPUs when we use hotplug to remove them ;-))) */
1323         if (swsusp_info.cpus != num_possible_cpus())
1324                 return "number of cpus";
1325 #endif
1326         return NULL;
1327 }
1328
1329
1330 static int check_header(void)
1331 {
1332         const char * reason = NULL;
1333         int error;
1334
1335         if ((error = bio_read_page(swp_offset(swsusp_header.swsusp_info), &swsusp_info)))
1336                 return error;
1337
1338         /* Is this same machine? */
1339         if ((reason = sanity_check())) {
1340                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n",reason);
1341                 return -EPERM;
1342         }
1343         nr_copy_pages = swsusp_info.image_pages;
1344         return error;
1345 }
1346
1347 static int check_sig(void)
1348 {
1349         int error;
1350
1351         memset(&swsusp_header, 0, sizeof(swsusp_header));
1352         if ((error = bio_read_page(0, &swsusp_header)))
1353                 return error;
1354         if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
1355                 memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
1356                 memcpy(key_iv, swsusp_header.key_iv, MAXKEY+MAXIV);
1357                 memset(swsusp_header.key_iv, 0, MAXKEY+MAXIV);
1358
1359                 /*
1360                  * Reset swap signature now.
1361                  */
1362                 error = bio_write_page(0, &swsusp_header);
1363         } else {
1364                 return -EINVAL;
1365         }
1366         if (!error)
1367                 pr_debug("swsusp: Signature found, resuming\n");
1368         return error;
1369 }
1370
1371 /**
1372  *      data_read - Read image pages from swap.
1373  *
1374  *      You do not need to check for overlaps, check_pagedir()
1375  *      already did that.
1376  */
1377
1378 static int data_read(struct pbe *pblist)
1379 {
1380         struct pbe * p;
1381         int error = 0;
1382         int i = 0;
1383         int mod = swsusp_info.image_pages / 100;
1384         void *tfm;
1385
1386         if ((error = crypto_init(0, &tfm)))
1387                 return error;
1388
1389         if (!mod)
1390                 mod = 1;
1391
1392         printk("swsusp: Reading image data (%lu pages):     ",
1393                         swsusp_info.image_pages);
1394
1395         for_each_pbe (p, pblist) {
1396                 if (!(i % mod))
1397                         printk("\b\b\b\b%3d%%", i / mod);
1398
1399                 if ((error = crypto_read(p, tfm))) {
1400                         crypto_exit(tfm);
1401                         return error;
1402                 }
1403
1404                 i++;
1405         }
1406         printk("\b\b\b\bdone\n");
1407         crypto_exit(tfm);
1408         return error;
1409 }
1410
1411 /**
1412  *      read_pagedir - Read page backup list pages from swap
1413  */
1414
1415 static int read_pagedir(struct pbe *pblist)
1416 {
1417         struct pbe *pbpage, *p;
1418         unsigned i = 0;
1419         int error;
1420
1421         if (!pblist)
1422                 return -EFAULT;
1423
1424         printk("swsusp: Reading pagedir (%lu pages)\n",
1425                         swsusp_info.pagedir_pages);
1426
1427         for_each_pb_page (pbpage, pblist) {
1428                 unsigned long offset = swp_offset(swsusp_info.pagedir[i++]);
1429
1430                 error = -EFAULT;
1431                 if (offset) {
1432                         p = (pbpage + PB_PAGE_SKIP)->next;
1433                         error = bio_read_page(offset, (void *)pbpage);
1434                         (pbpage + PB_PAGE_SKIP)->next = p;
1435                 }
1436                 if (error)
1437                         break;
1438         }
1439
1440         if (error)
1441                 free_page((unsigned long)pblist);
1442
1443         BUG_ON(i != swsusp_info.pagedir_pages);
1444
1445         return error;
1446 }
1447
1448
1449 static int check_suspend_image(void)
1450 {
1451         int error = 0;
1452
1453         if ((error = check_sig()))
1454                 return error;
1455
1456         if ((error = check_header()))
1457                 return error;
1458
1459         return 0;
1460 }
1461
1462 static int read_suspend_image(void)
1463 {
1464         int error = 0;
1465         struct pbe *p;
1466
1467         if (!(p = alloc_pagedir(nr_copy_pages)))
1468                 return -ENOMEM;
1469
1470         if ((error = read_pagedir(p)))
1471                 return error;
1472
1473         create_pbe_list(p, nr_copy_pages);
1474
1475         if (!(pagedir_nosave = swsusp_pagedir_relocate(p)))
1476                 return -ENOMEM;
1477
1478         /* Allocate memory for the image and read the data from swap */
1479
1480         error = check_pagedir(pagedir_nosave);
1481         free_eaten_memory();
1482         if (!error)
1483                 error = data_read(pagedir_nosave);
1484
1485         if (error) { /* We fail cleanly */
1486                 for_each_pbe (p, pagedir_nosave)
1487                         if (p->address) {
1488                                 free_page(p->address);
1489                                 p->address = 0UL;
1490                         }
1491                 free_pagedir(pagedir_nosave);
1492         }
1493         return error;
1494 }
1495
1496 /**
1497  *      swsusp_check - Check for saved image in swap
1498  */
1499
1500 int swsusp_check(void)
1501 {
1502         int error;
1503
1504         resume_bdev = open_by_devnum(swsusp_resume_device, FMODE_READ);
1505         if (!IS_ERR(resume_bdev)) {
1506                 set_blocksize(resume_bdev, PAGE_SIZE);
1507                 error = check_suspend_image();
1508                 if (error)
1509                     blkdev_put(resume_bdev);
1510         } else
1511                 error = PTR_ERR(resume_bdev);
1512
1513         if (!error)
1514                 pr_debug("swsusp: resume file found\n");
1515         else
1516                 pr_debug("swsusp: Error %d check for resume file\n", error);
1517         return error;
1518 }
1519
1520 /**
1521  *      swsusp_read - Read saved image from swap.
1522  */
1523
1524 int swsusp_read(void)
1525 {
1526         int error;
1527
1528         if (IS_ERR(resume_bdev)) {
1529                 pr_debug("swsusp: block device not initialised\n");
1530                 return PTR_ERR(resume_bdev);
1531         }
1532
1533         error = read_suspend_image();
1534         blkdev_put(resume_bdev);
1535         memset(key_iv, 0, MAXKEY+MAXIV);
1536
1537         if (!error)
1538                 pr_debug("swsusp: Reading resume file was successful\n");
1539         else
1540                 pr_debug("swsusp: Error %d resuming\n", error);
1541         return error;
1542 }
1543
1544 /**
1545  *      swsusp_close - close swap device.
1546  */
1547
1548 void swsusp_close(void)
1549 {
1550         if (IS_ERR(resume_bdev)) {
1551                 pr_debug("swsusp: block device not initialised\n");
1552                 return;
1553         }
1554
1555         blkdev_put(resume_bdev);
1556 }