kernel/power/swsusp.c

   1 /*
   2  * linux/kernel/power/swsusp.c
   3  *
   4  * This file provides code to write suspend image to swap and read it back.
   5  *
   6  * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu>
   7  * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@suse.cz>
   8  *
   9  * This file is released under the GPLv2.
  10  *
  11  * I'd like to thank the following people for their work:
  12  *
  13  * Pavel Machek <pavel@ucw.cz>:
  14  * Modifications, defectiveness pointing, being with me at the very beginning,
  15  * suspend to swap space, stop all tasks. Port to 2.4.18-ac and 2.5.17.
  16  *
  17  * Steve Doddi <dirk@loth.demon.co.uk>:
  18  * Support the possibility of hardware state restoring.
  19  *
  20  * Raph <grey.havens@earthling.net>:
  21  * Support for preserving states of network devices and virtual console
  22  * (including X and svgatextmode)
  23  *
  24  * Kurt Garloff <garloff@suse.de>:
  25  * Straightened the critical function in order to prevent compilers from
  26  * playing tricks with local variables.
  27  *
  28  * Andreas Mohr <a.mohr@mailto.de>
  29  *
  30  * Alex Badea <vampire@go.ro>:
  31  * Fixed runaway init
  32  *
  33  * Andreas Steinmetz <ast@domdv.de>:
  34  * Added encrypted suspend option
  35  *
  36  * More state savers are welcome. Especially for the scsi layer...
  37  *
  38  * For TODOs,FIXMEs also look in Documentation/power/swsusp.txt
  39  */
  40
  41 #include <linux/module.h>
  42 #include <linux/mm.h>
  43 #include <linux/suspend.h>
  44 #include <linux/smp_lock.h>
  45 #include <linux/file.h>
  46 #include <linux/utsname.h>
  47 #include <linux/version.h>
  48 #include <linux/delay.h>
  49 #include <linux/bitops.h>
  50 #include <linux/spinlock.h>
  51 #include <linux/genhd.h>
  52 #include <linux/kernel.h>
  53 #include <linux/major.h>
  54 #include <linux/swap.h>
  55 #include <linux/pm.h>
  56 #include <linux/device.h>
  57 #include <linux/buffer_head.h>
  58 #include <linux/swapops.h>
  59 #include <linux/bootmem.h>
  60 #include <linux/syscalls.h>
  61 #include <linux/highmem.h>
  62 #include <linux/bio.h>
  63
  64 #include <asm/uaccess.h>
  65 #include <asm/mmu_context.h>
  66 #include <asm/pgtable.h>
  67 #include <asm/tlbflush.h>
  68 #include <asm/io.h>
  69
  70 #include <linux/random.h>
  71 #include <linux/crypto.h>
  72 #include <asm/scatterlist.h>
  73
  74 #include "power.h"
  75
  76 #define CIPHER "aes"
  77 #define MAXKEY 32
  78 #define MAXIV  32
  79
  80 extern char resume_file[];
  81
  82 /* Local variables that should not be affected by save */
  83 unsigned int nr_copy_pages __nosavedata = 0;
  84
  85 /* Suspend pagedir is allocated before final copy, therefore it
  86    must be freed after resume
  87
  88    Warning: this is even more evil than it seems. Pagedirs this file
  89    talks about are completely different from page directories used by
  90    MMU hardware.
  91  */
  92 suspend_pagedir_t *pagedir_nosave __nosavedata = NULL;
  93
  94 #define SWSUSP_SIG      "S1SUSPEND"
  95
  96 static struct swsusp_header {
  97         char reserved[PAGE_SIZE - 20 - MAXKEY - MAXIV - sizeof(swp_entry_t)];
  98         u8 key_iv[MAXKEY+MAXIV];
  99         swp_entry_t swsusp_info;
 100         char    orig_sig[10];
 101         char    sig[10];
 102 } __attribute__((packed, aligned(PAGE_SIZE))) swsusp_header;
 103
 104 static struct swsusp_info swsusp_info;
 105
 106 /*
 107  * Saving part...
 108  */
 109
 110 /* We memorize in swapfile_used what swap devices are used for suspension */
 111 #define SWAPFILE_UNUSED    0
 112 #define SWAPFILE_SUSPEND   1    /* This is the suspending device */
 113 #define SWAPFILE_IGNORED   2    /* Those are other swap devices ignored for suspension */
 114
 115 static unsigned short swapfile_used[MAX_SWAPFILES];
 116 static unsigned short root_swap;
 117
 118 static int write_page(unsigned long addr, swp_entry_t *loc);
 119 static int bio_read_page(pgoff_t page_off, void *page);
 120
 121 static u8 key_iv[MAXKEY+MAXIV];
 122
 123 #ifdef CONFIG_SWSUSP_ENCRYPT
 124
 125 static int crypto_init(int mode, void **mem)
 126 {
 127         int error = 0;
 128         int len;
 129         char *modemsg;
 130         struct crypto_tfm *tfm;
 131
 132         modemsg = mode ? "suspend not possible" : "resume not possible";
 133
 134         tfm = crypto_alloc_tfm(CIPHER, CRYPTO_TFM_MODE_CBC);
 135         if(!tfm) {
 136                 printk(KERN_ERR "swsusp: no tfm, %s\n", modemsg);
 137                 error = -EINVAL;
 138                 goto out;
 139         }
 140
 141         if(MAXKEY < crypto_tfm_alg_min_keysize(tfm)) {
 142                 printk(KERN_ERR "swsusp: key buffer too small, %s\n", modemsg);
 143                 error = -ENOKEY;
 144                 goto fail;
 145         }
 146
 147         if (mode)
 148                 get_random_bytes(key_iv, MAXKEY+MAXIV);
 149
 150         len = crypto_tfm_alg_max_keysize(tfm);
 151         if (len > MAXKEY)
 152                 len = MAXKEY;
 153
 154         if (crypto_cipher_setkey(tfm, key_iv, len)) {
 155                 printk(KERN_ERR "swsusp: key setup failure, %s\n", modemsg);
 156                 error = -EKEYREJECTED;
 157                 goto fail;
 158         }
 159
 160         len = crypto_tfm_alg_ivsize(tfm);
 161
 162         if (MAXIV < len) {
 163                 printk(KERN_ERR "swsusp: iv buffer too small, %s\n", modemsg);
 164                 error = -EOVERFLOW;
 165                 goto fail;
 166         }
 167
 168         crypto_cipher_set_iv(tfm, key_iv+MAXKEY, len);
 169
 170         *mem=(void *)tfm;
 171
 172         goto out;
 173
 174 fail:   crypto_free_tfm(tfm);
 175 out:    return error;
 176 }
 177
 178 static __inline__ void crypto_exit(void *mem)
 179 {
 180         crypto_free_tfm((struct crypto_tfm *)mem);
 181 }
 182
 183 static __inline__ int crypto_write(struct pbe *p, void *mem)
 184 {
 185         int error = 0;
 186         struct scatterlist src, dst;
 187
 188         src.page   = virt_to_page(p->address);
 189         src.offset = 0;
 190         src.length = PAGE_SIZE;
 191         dst.page   = virt_to_page((void *)&swsusp_header);
 192         dst.offset = 0;
 193         dst.length = PAGE_SIZE;
 194
 195         error = crypto_cipher_encrypt((struct crypto_tfm *)mem, &dst, &src,
 196                                         PAGE_SIZE);
 197
 198         if (!error)
 199                 error = write_page((unsigned long)&swsusp_header,
 200                                 &(p->swap_address));
 201         return error;
 202 }
 203
 204 static __inline__ int crypto_read(struct pbe *p, void *mem)
 205 {
 206         int error = 0;
 207         struct scatterlist src, dst;
 208
 209         error = bio_read_page(swp_offset(p->swap_address), (void *)p->address);
 210         if (!error) {
 211                 src.offset = 0;
 212                 src.length = PAGE_SIZE;
 213                 dst.offset = 0;
 214                 dst.length = PAGE_SIZE;
 215                 src.page = dst.page = virt_to_page((void *)p->address);
 216
 217                 error = crypto_cipher_decrypt((struct crypto_tfm *)mem, &dst,
 218                                                 &src, PAGE_SIZE);
 219         }
 220         return error;
 221 }
 222 #else
 223 static __inline__ int crypto_init(int mode, void *mem)
 224 {
 225         return 0;
 226 }
 227
 228 static __inline__ void crypto_exit(void *mem)
 229 {
 230 }
 231
 232 static __inline__ int crypto_write(struct pbe *p, void *mem)
 233 {
 234         return write_page(p->address, &(p->swap_address));
 235 }
 236
 237 static __inline__ int crypto_read(struct pbe *p, void *mem)
 238 {
 239         return bio_read_page(swp_offset(p->swap_address), (void *)p->address);
 240 }
 241 #endif
 242
 243 static int mark_swapfiles(swp_entry_t prev)
 244 {
 245         int error;
 246
 247         rw_swap_page_sync(READ,
 248                           swp_entry(root_swap, 0),
 249                           virt_to_page((unsigned long)&swsusp_header));
 250         if (!memcmp("SWAP-SPACE",swsusp_header.sig, 10) ||
 251             !memcmp("SWAPSPACE2",swsusp_header.sig, 10)) {
 252                 memcpy(swsusp_header.orig_sig,swsusp_header.sig, 10);
 253                 memcpy(swsusp_header.sig,SWSUSP_SIG, 10);
 254                 memcpy(swsusp_header.key_iv, key_iv, MAXKEY+MAXIV);
 255                 swsusp_header.swsusp_info = prev;
 256                 error = rw_swap_page_sync(WRITE,
 257                                           swp_entry(root_swap, 0),
 258                                           virt_to_page((unsigned long)
 259                                                        &swsusp_header));
 260         } else {
 261                 pr_debug("swsusp: Partition is not swap space.\n");
 262                 error = -ENODEV;
 263         }
 264         return error;
 265 }
 266
 267 /*
 268  * Check whether the swap device is the specified resume
 269  * device, irrespective of whether they are specified by
 270  * identical names.
 271  *
 272  * (Thus, device inode aliasing is allowed.  You can say /dev/hda4
 273  * instead of /dev/ide/host0/bus0/target0/lun0/part4 [if using devfs]
 274  * and they'll be considered the same device.  This is *necessary* for
 275  * devfs, since the resume code can only recognize the form /dev/hda4,
 276  * but the suspend code would see the long name.)
 277  */
 278 static int is_resume_device(const struct swap_info_struct *swap_info)
 279 {
 280         struct file *file = swap_info->swap_file;
 281         struct inode *inode = file->f_dentry->d_inode;
 282
 283         return S_ISBLK(inode->i_mode) &&
 284                 swsusp_resume_device == MKDEV(imajor(inode), iminor(inode));
 285 }
 286
 287 static int swsusp_swap_check(void) /* This is called before saving image */
 288 {
 289         int i, len;
 290
 291         len=strlen(resume_file);
 292         root_swap = 0xFFFF;
 293
 294         spin_lock(&swap_lock);
 295         for (i=0; i<MAX_SWAPFILES; i++) {
 296                 if (!(swap_info[i].flags & SWP_WRITEOK)) {
 297                         swapfile_used[i]=SWAPFILE_UNUSED;
 298                 } else {
 299                         if (!len) {
 300                                 printk(KERN_WARNING "resume= option should be used to set suspend device" );
 301                                 if (root_swap == 0xFFFF) {
 302                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 303                                         root_swap = i;
 304                                 } else
 305                                         swapfile_used[i] = SWAPFILE_IGNORED;
 306                         } else {
 307                                 /* we ignore all swap devices that are not the resume_file */
 308                                 if (is_resume_device(&swap_info[i])) {
 309                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 310                                         root_swap = i;
 311                                 } else {
 312                                         swapfile_used[i] = SWAPFILE_IGNORED;
 313                                 }
 314                         }
 315                 }
 316         }
 317         spin_unlock(&swap_lock);
 318         return (root_swap != 0xffff) ? 0 : -ENODEV;
 319 }
 320
 321 /**
 322  * This is called after saving image so modification
 323  * will be lost after resume... and that's what we want.
 324  * we make the device unusable. A new call to
 325  * lock_swapdevices can unlock the devices.
 326  */
 327 static void lock_swapdevices(void)
 328 {
 329         int i;
 330
 331         spin_lock(&swap_lock);
 332         for (i = 0; i< MAX_SWAPFILES; i++)
 333                 if (swapfile_used[i] == SWAPFILE_IGNORED) {
 334                         swap_info[i].flags ^= SWP_WRITEOK;
 335                 }
 336         spin_unlock(&swap_lock);
 337 }
 338
 339 /**
 340  *      write_page - Write one page to a fresh swap location.
 341  *      @addr:  Address we're writing.
 342  *      @loc:   Place to store the entry we used.
 343  *
 344  *      Allocate a new swap entry and 'sync' it. Note we discard -EIO
 345  *      errors. That is an artifact left over from swsusp. It did not
 346  *      check the return of rw_swap_page_sync() at all, since most pages
 347  *      written back to swap would return -EIO.
 348  *      This is a partial improvement, since we will at least return other
 349  *      errors, though we need to eventually fix the damn code.
 350  */
 351 static int write_page(unsigned long addr, swp_entry_t *loc)
 352 {
 353         swp_entry_t entry;
 354         int error = 0;
 355
 356         entry = get_swap_page();
 357         if (swp_offset(entry) &&
 358             swapfile_used[swp_type(entry)] == SWAPFILE_SUSPEND) {
 359                 error = rw_swap_page_sync(WRITE, entry,
 360                                           virt_to_page(addr));
 361                 if (error == -EIO)
 362                         error = 0;
 363                 if (!error)
 364                         *loc = entry;
 365         } else
 366                 error = -ENOSPC;
 367         return error;
 368 }
 369
 370 /**
 371  *      data_free - Free the swap entries used by the saved image.
 372  *
 373  *      Walk the list of used swap entries and free each one.
 374  *      This is only used for cleanup when suspend fails.
 375  */
 376 static void data_free(void)
 377 {
 378         swp_entry_t entry;
 379         struct pbe *p;
 380
 381         for_each_pbe (p, pagedir_nosave) {
 382                 entry = p->swap_address;
 383                 if (entry.val)
 384                         swap_free(entry);
 385                 else
 386                         break;
 387         }
 388 }
 389
 390 /**
 391  *      data_write - Write saved image to swap.
 392  *
 393  *      Walk the list of pages in the image and sync each one to swap.
 394  */
 395 static int data_write(void)
 396 {
 397         int error = 0, i = 0;
 398         unsigned int mod = nr_copy_pages / 100;
 399         struct pbe *p;
 400         void *tfm;
 401
 402         if ((error = crypto_init(1, &tfm)))
 403                 return error;
 404
 405         if (!mod)
 406                 mod = 1;
 407
 408         printk( "Writing data to swap (%d pages)...     ", nr_copy_pages );
 409         for_each_pbe (p, pagedir_nosave) {
 410                 if (!(i%mod))
 411                         printk( "\b\b\b\b%3d%%", i / mod );
 412                 if ((error = crypto_write(p, tfm))) {
 413                         crypto_exit(tfm);
 414                         return error;
 415                 }
 416                 i++;
 417         }
 418         printk("\b\b\b\bdone\n");
 419         crypto_exit(tfm);
 420         return error;
 421 }
 422
 423 static void dump_info(void)
 424 {
 425         pr_debug(" swsusp: Version: %u\n",swsusp_info.version_code);
 426         pr_debug(" swsusp: Num Pages: %ld\n",swsusp_info.num_physpages);
 427         pr_debug(" swsusp: UTS Sys: %s\n",swsusp_info.uts.sysname);
 428         pr_debug(" swsusp: UTS Node: %s\n",swsusp_info.uts.nodename);
 429         pr_debug(" swsusp: UTS Release: %s\n",swsusp_info.uts.release);
 430         pr_debug(" swsusp: UTS Version: %s\n",swsusp_info.uts.version);
 431         pr_debug(" swsusp: UTS Machine: %s\n",swsusp_info.uts.machine);
 432         pr_debug(" swsusp: UTS Domain: %s\n",swsusp_info.uts.domainname);
 433         pr_debug(" swsusp: CPUs: %d\n",swsusp_info.cpus);
 434         pr_debug(" swsusp: Image: %ld Pages\n",swsusp_info.image_pages);
 435         pr_debug(" swsusp: Pagedir: %ld Pages\n",swsusp_info.pagedir_pages);
 436 }
 437
 438 static void init_header(void)
 439 {
 440         memset(&swsusp_info, 0, sizeof(swsusp_info));
 441         swsusp_info.version_code = LINUX_VERSION_CODE;
 442         swsusp_info.num_physpages = num_physpages;
 443         memcpy(&swsusp_info.uts, &system_utsname, sizeof(system_utsname));
 444
 445         swsusp_info.suspend_pagedir = pagedir_nosave;
 446         swsusp_info.cpus = num_online_cpus();
 447         swsusp_info.image_pages = nr_copy_pages;
 448 }
 449
 450 static int close_swap(void)
 451 {
 452         swp_entry_t entry;
 453         int error;
 454
 455         dump_info();
 456         error = write_page((unsigned long)&swsusp_info, &entry);
 457         if (!error) {
 458                 printk( "S" );
 459                 error = mark_swapfiles(entry);
 460                 printk( "|\n" );
 461         }
 462         return error;
 463 }
 464
 465 /**
 466  *      free_pagedir_entries - Free pages used by the page directory.
 467  *
 468  *      This is used during suspend for error recovery.
 469  */
 470
 471 static void free_pagedir_entries(void)
 472 {
 473         int i;
 474
 475         for (i = 0; i < swsusp_info.pagedir_pages; i++)
 476                 swap_free(swsusp_info.pagedir[i]);
 477 }
 478
 479
 480 /**
 481  *      write_pagedir - Write the array of pages holding the page directory.
 482  *      @last:  Last swap entry we write (needed for header).
 483  */
 484
 485 static int write_pagedir(void)
 486 {
 487         int error = 0;
 488         unsigned int n = 0;
 489         struct pbe *pbe;
 490
 491         printk( "Writing pagedir...");
 492         for_each_pb_page (pbe, pagedir_nosave) {
 493                 if ((error = write_page((unsigned long)pbe, &swsusp_info.pagedir[n++])))
 494                         return error;
 495         }
 496
 497         swsusp_info.pagedir_pages = n;
 498         printk("done (%u pages)\n", n);
 499         return error;
 500 }
 501
 502 /**
 503  *      write_suspend_image - Write entire image and metadata.
 504  *
 505  */
 506 static int write_suspend_image(void)
 507 {
 508         int error;
 509
 510         init_header();
 511         if ((error = data_write()))
 512                 goto FreeData;
 513
 514         if ((error = write_pagedir()))
 515                 goto FreePagedir;
 516
 517         if ((error = close_swap()))
 518                 goto FreePagedir;
 519  Done:
 520         memset(key_iv, 0, MAXKEY+MAXIV);
 521         return error;
 522  FreePagedir:
 523         free_pagedir_entries();
 524  FreeData:
 525         data_free();
 526         goto Done;
 527 }
 528
 529 /**
 530  *      enough_swap - Make sure we have enough swap to save the image.
 531  *
 532  *      Returns TRUE or FALSE after checking the total amount of swap
 533  *      space avaiable.
 534  *
 535  *      FIXME: si_swapinfo(&i) returns all swap devices information.
 536  *      We should only consider resume_device.
 537  */
 538
 539 int enough_swap(unsigned int nr_pages)
 540 {
 541         struct sysinfo i;
 542
 543         si_swapinfo(&i);
 544         pr_debug("swsusp: available swap: %lu pages\n", i.freeswap);
 545         return i.freeswap > (nr_pages + PAGES_FOR_IO +
 546                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 547 }
 548
 549
 550 /* It is important _NOT_ to umount filesystems at this point. We want
 551  * them synced (in case something goes wrong) but we DO not want to mark
 552  * filesystem clean: it is not. (And it does not matter, if we resume
 553  * correctly, we'll mark system clean, anyway.)
 554  */
 555 int swsusp_write(void)
 556 {
 557         int error;
 558
 559         lock_swapdevices();
 560         error = write_suspend_image();
 561         /* This will unlock ignored swap devices since writing is finished */
 562         lock_swapdevices();
 563         return error;
 564
 565 }
 566
 567
 568
 569 int swsusp_suspend(void)
 570 {
 571         int error;
 572         if ((error = arch_prepare_suspend()))
 573                 return error;
 574         local_irq_disable();
 575         /* At this point, device_suspend() has been called, but *not*
 576          * device_power_down(). We *must* device_power_down() now.
 577          * Otherwise, drivers for some devices (e.g. interrupt controllers)
 578          * become desynchronized with the actual state of the hardware
 579          * at resume time, and evil weirdness ensues.
 580          */
 581         if ((error = device_power_down(PMSG_FREEZE))) {
 582                 printk(KERN_ERR "Some devices failed to power down, aborting suspend\n");
 583                 local_irq_enable();
 584                 return error;
 585         }
 586
 587         if ((error = swsusp_swap_check())) {
 588                 printk(KERN_ERR "swsusp: cannot find swap device, try swapon -a.\n");
 589                 device_power_up();
 590                 local_irq_enable();
 591                 return error;
 592         }
 593
 594         save_processor_state();
 595         if ((error = swsusp_arch_suspend()))
 596                 printk(KERN_ERR "Error %d suspending\n", error);
 597         /* Restore control flow magically appears here */
 598         restore_processor_state();
 599         restore_highmem();
 600         device_power_up();
 601         local_irq_enable();
 602         return error;
 603 }
 604
 605 int swsusp_resume(void)
 606 {
 607         int error;
 608         local_irq_disable();
 609         if (device_power_down(PMSG_FREEZE))
 610                 printk(KERN_ERR "Some devices failed to power down, very bad\n");
 611         /* We'll ignore saved state, but this gets preempt count (etc) right */
 612         save_processor_state();
 613         error = swsusp_arch_resume();
 614         /* Code below is only ever reached in case of failure. Otherwise
 615          * execution continues at place where swsusp_arch_suspend was called
 616          */
 617         BUG_ON(!error);
 618         /* The only reason why swsusp_arch_resume() can fail is memory being
 619          * very tight, so we have to free it as soon as we can to avoid
 620          * subsequent failures
 621          */
 622         swsusp_free();
 623         restore_processor_state();
 624         restore_highmem();
 625         touch_softlockup_watchdog();
 626         device_power_up();
 627         local_irq_enable();
 628         return error;
 629 }
 630
 631 /**
 632  *      swsusp_pagedir_relocate - It is possible, that some memory pages
 633  *      occupied by the list of PBEs collide with pages where we're going to
 634  *      restore from the loaded pages later.  We relocate them here.
 635  */
 636
 637 static struct pbe *swsusp_pagedir_relocate(struct pbe *pblist)
 638 {
 639         struct zone *zone;
 640         unsigned long zone_pfn;
 641         struct pbe *pbpage, *tail, *p;
 642         void *m;
 643         int rel = 0;
 644
 645         if (!pblist) /* a sanity check */
 646                 return NULL;
 647
 648         pr_debug("swsusp: Relocating pagedir (%lu pages to check)\n",
 649                         swsusp_info.pagedir_pages);
 650
 651         /* Clear page flags */
 652
 653         for_each_zone (zone) {
 654                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 655                         if (pfn_valid(zone_pfn + zone->zone_start_pfn))
 656                                 ClearPageNosaveFree(pfn_to_page(zone_pfn +
 657                                         zone->zone_start_pfn));
 658         }
 659
 660         /* Mark orig addresses */
 661
 662         for_each_pbe (p, pblist)
 663                 SetPageNosaveFree(virt_to_page(p->orig_address));
 664
 665         tail = pblist + PB_PAGE_SKIP;
 666
 667         /* Relocate colliding pages */
 668
 669         for_each_pb_page (pbpage, pblist) {
 670                 if (PageNosaveFree(virt_to_page((unsigned long)pbpage))) {
 671                         m = (void *)get_safe_page(GFP_ATOMIC | __GFP_COLD);
 672                         if (!m)
 673                                 return NULL;
 674                         memcpy(m, (void *)pbpage, PAGE_SIZE);
 675                         if (pbpage == pblist)
 676                                 pblist = (struct pbe *)m;
 677                         else
 678                                 tail->next = (struct pbe *)m;
 679                         pbpage = (struct pbe *)m;
 680
 681                         /* We have to link the PBEs again */
 682                         for (p = pbpage; p < pbpage + PB_PAGE_SKIP; p++)
 683                                 if (p->next) /* needed to save the end */
 684                                         p->next = p + 1;
 685
 686                         rel++;
 687                 }
 688                 tail = pbpage + PB_PAGE_SKIP;
 689         }
 690
 691         /* This is for swsusp_free() */
 692         for_each_pb_page (pbpage, pblist) {
 693                 SetPageNosave(virt_to_page(pbpage));
 694                 SetPageNosaveFree(virt_to_page(pbpage));
 695         }
 696
 697         printk("swsusp: Relocated %d pages\n", rel);
 698
 699         return pblist;
 700 }
 701
 702 /*
 703  *      Using bio to read from swap.
 704  *      This code requires a bit more work than just using buffer heads
 705  *      but, it is the recommended way for 2.5/2.6.
 706  *      The following are to signal the beginning and end of I/O. Bios
 707  *      finish asynchronously, while we want them to happen synchronously.
 708  *      A simple atomic_t, and a wait loop take care of this problem.
 709  */
 710
 711 static atomic_t io_done = ATOMIC_INIT(0);
 712
 713 static int end_io(struct bio *bio, unsigned int num, int err)
 714 {
 715         if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
 716                 panic("I/O error reading memory image");
 717         atomic_set(&io_done, 0);
 718         return 0;
 719 }
 720
 721 static struct block_device *resume_bdev;
 722
 723 /**
 724  *      submit - submit BIO request.
 725  *      @rw:    READ or WRITE.
 726  *      @off    physical offset of page.
 727  *      @page:  page we're reading or writing.
 728  *
 729  *      Straight from the textbook - allocate and initialize the bio.
 730  *      If we're writing, make sure the page is marked as dirty.
 731  *      Then submit it and wait.
 732  */
 733
 734 static int submit(int rw, pgoff_t page_off, void *page)
 735 {
 736         int error = 0;
 737         struct bio *bio;
 738
 739         bio = bio_alloc(GFP_ATOMIC, 1);
 740         if (!bio)
 741                 return -ENOMEM;
 742         bio->bi_sector = page_off * (PAGE_SIZE >> 9);
 743         bio_get(bio);
 744         bio->bi_bdev = resume_bdev;
 745         bio->bi_end_io = end_io;
 746
 747         if (bio_add_page(bio, virt_to_page(page), PAGE_SIZE, 0) < PAGE_SIZE) {
 748                 printk("swsusp: ERROR: adding page to bio at %ld\n",page_off);
 749                 error = -EFAULT;
 750                 goto Done;
 751         }
 752
 753         if (rw == WRITE)
 754                 bio_set_pages_dirty(bio);
 755
 756         atomic_set(&io_done, 1);
 757         submit_bio(rw | (1 << BIO_RW_SYNC), bio);
 758         while (atomic_read(&io_done))
 759                 yield();
 760
 761  Done:
 762         bio_put(bio);
 763         return error;
 764 }
 765
 766 static int bio_read_page(pgoff_t page_off, void *page)
 767 {
 768         return submit(READ, page_off, page);
 769 }
 770
 771 static int bio_write_page(pgoff_t page_off, void *page)
 772 {
 773         return submit(WRITE, page_off, page);
 774 }
 775
 776 /*
 777  * Sanity check if this image makes sense with this kernel/swap context
 778  * I really don't think that it's foolproof but more than nothing..
 779  */
 780
 781 static const char *sanity_check(void)
 782 {
 783         dump_info();
 784         if (swsusp_info.version_code != LINUX_VERSION_CODE)
 785                 return "kernel version";
 786         if (swsusp_info.num_physpages != num_physpages)
 787                 return "memory size";
 788         if (strcmp(swsusp_info.uts.sysname,system_utsname.sysname))
 789                 return "system type";
 790         if (strcmp(swsusp_info.uts.release,system_utsname.release))
 791                 return "kernel release";
 792         if (strcmp(swsusp_info.uts.version,system_utsname.version))
 793                 return "version";
 794         if (strcmp(swsusp_info.uts.machine,system_utsname.machine))
 795                 return "machine";
 796 #if 0
 797         /* We can't use number of online CPUs when we use hotplug to remove them ;-))) */
 798         if (swsusp_info.cpus != num_possible_cpus())
 799                 return "number of cpus";
 800 #endif
 801         return NULL;
 802 }
 803
 804
 805 static int check_header(void)
 806 {
 807         const char *reason = NULL;
 808         int error;
 809
 810         if ((error = bio_read_page(swp_offset(swsusp_header.swsusp_info), &swsusp_info)))
 811                 return error;
 812
 813         /* Is this same machine? */
 814         if ((reason = sanity_check())) {
 815                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n",reason);
 816                 return -EPERM;
 817         }
 818         nr_copy_pages = swsusp_info.image_pages;
 819         return error;
 820 }
 821
 822 static int check_sig(void)
 823 {
 824         int error;
 825
 826         memset(&swsusp_header, 0, sizeof(swsusp_header));
 827         if ((error = bio_read_page(0, &swsusp_header)))
 828                 return error;
 829         if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
 830                 memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
 831                 memcpy(key_iv, swsusp_header.key_iv, MAXKEY+MAXIV);
 832                 memset(swsusp_header.key_iv, 0, MAXKEY+MAXIV);
 833
 834                 /*
 835                  * Reset swap signature now.
 836                  */
 837                 error = bio_write_page(0, &swsusp_header);
 838         } else {
 839                 return -EINVAL;
 840         }
 841         if (!error)
 842                 pr_debug("swsusp: Signature found, resuming\n");
 843         return error;
 844 }
 845
 846 /**
 847  *      data_read - Read image pages from swap.
 848  *
 849  *      You do not need to check for overlaps, check_pagedir()
 850  *      already did that.
 851  */
 852
 853 static int data_read(struct pbe *pblist)
 854 {
 855         struct pbe *p;
 856         int error = 0;
 857         int i = 0;
 858         int mod = swsusp_info.image_pages / 100;
 859         void *tfm;
 860
 861         if ((error = crypto_init(0, &tfm)))
 862                 return error;
 863
 864         if (!mod)
 865                 mod = 1;
 866
 867         printk("swsusp: Reading image data (%lu pages):     ",
 868                         swsusp_info.image_pages);
 869
 870         for_each_pbe (p, pblist) {
 871                 if (!(i % mod))
 872                         printk("\b\b\b\b%3d%%", i / mod);
 873
 874                 if ((error = crypto_read(p, tfm))) {
 875                         crypto_exit(tfm);
 876                         return error;
 877                 }
 878
 879                 i++;
 880         }
 881         printk("\b\b\b\bdone\n");
 882         crypto_exit(tfm);
 883         return error;
 884 }
 885
 886 /**
 887  *      read_pagedir - Read page backup list pages from swap
 888  */
 889
 890 static int read_pagedir(struct pbe *pblist)
 891 {
 892         struct pbe *pbpage, *p;
 893         unsigned int i = 0;
 894         int error;
 895
 896         if (!pblist)
 897                 return -EFAULT;
 898
 899         printk("swsusp: Reading pagedir (%lu pages)\n",
 900                         swsusp_info.pagedir_pages);
 901
 902         for_each_pb_page (pbpage, pblist) {
 903                 unsigned long offset = swp_offset(swsusp_info.pagedir[i++]);
 904
 905                 error = -EFAULT;
 906                 if (offset) {
 907                         p = (pbpage + PB_PAGE_SKIP)->next;
 908                         error = bio_read_page(offset, (void *)pbpage);
 909                         (pbpage + PB_PAGE_SKIP)->next = p;
 910                 }
 911                 if (error)
 912                         break;
 913         }
 914
 915         if (!error)
 916                 BUG_ON(i != swsusp_info.pagedir_pages);
 917
 918         return error;
 919 }
 920
 921
 922 static int check_suspend_image(void)
 923 {
 924         int error = 0;
 925
 926         if ((error = check_sig()))
 927                 return error;
 928
 929         if ((error = check_header()))
 930                 return error;
 931
 932         return 0;
 933 }
 934
 935 static int read_suspend_image(void)
 936 {
 937         int error = 0;
 938         struct pbe *p;
 939
 940         if (!(p = alloc_pagedir(nr_copy_pages, GFP_ATOMIC, 0)))
 941                 return -ENOMEM;
 942
 943         if ((error = read_pagedir(p)))
 944                 return error;
 945
 946         create_pbe_list(p, nr_copy_pages);
 947
 948         if (!(pagedir_nosave = swsusp_pagedir_relocate(p)))
 949                 return -ENOMEM;
 950
 951         /* Allocate memory for the image and read the data from swap */
 952
 953         error = alloc_data_pages(pagedir_nosave, GFP_ATOMIC, 1);
 954
 955         if (!error)
 956                 error = data_read(pagedir_nosave);
 957
 958         return error;
 959 }
 960
 961 /**
 962  *      swsusp_check - Check for saved image in swap
 963  */
 964
 965 int swsusp_check(void)
 966 {
 967         int error;
 968
 969         resume_bdev = open_by_devnum(swsusp_resume_device, FMODE_READ);
 970         if (!IS_ERR(resume_bdev)) {
 971                 set_blocksize(resume_bdev, PAGE_SIZE);
 972                 error = check_suspend_image();
 973                 if (error)
 974                     blkdev_put(resume_bdev);
 975         } else
 976                 error = PTR_ERR(resume_bdev);
 977
 978         if (!error)
 979                 pr_debug("swsusp: resume file found\n");
 980         else
 981                 pr_debug("swsusp: Error %d check for resume file\n", error);
 982         return error;
 983 }
 984
 985 /**
 986  *      swsusp_read - Read saved image from swap.
 987  */
 988
 989 int swsusp_read(void)
 990 {
 991         int error;
 992
 993         if (IS_ERR(resume_bdev)) {
 994                 pr_debug("swsusp: block device not initialised\n");
 995                 return PTR_ERR(resume_bdev);
 996         }
 997
 998         error = read_suspend_image();
 999         blkdev_put(resume_bdev);
1000         memset(key_iv, 0, MAXKEY+MAXIV);
1001
1002         if (!error)
1003                 pr_debug("swsusp: Reading resume file was successful\n");
1004         else
1005                 pr_debug("swsusp: Error %d resuming\n", error);
1006         return error;
1007 }
1008
1009 /**
1010  *      swsusp_close - close swap device.
1011  */
1012
1013 void swsusp_close(void)
1014 {
1015         if (IS_ERR(resume_bdev)) {
1016                 pr_debug("swsusp: block device not initialised\n");
1017                 return;
1018         }
1019
1020         blkdev_put(resume_bdev);
1021 }