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) 1998,2001-2005 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 extern char resume_file[];
  88
  89 /* Local variables that should not be affected by save */
  90 unsigned int nr_copy_pages __nosavedata = 0;
  91
  92 /* Suspend pagedir is allocated before final copy, therefore it
  93    must be freed after resume
  94
  95    Warning: this is evil. There are actually two pagedirs at time of
  96    resume. One is "pagedir_save", which is empty frame allocated at
  97    time of suspend, that must be freed. Second is "pagedir_nosave",
  98    allocated at time of resume, that travels through memory not to
  99    collide with anything.
 100
 101    Warning: this is even more evil than it seems. Pagedirs this file
 102    talks about are completely different from page directories used by
 103    MMU hardware.
 104  */
 105 suspend_pagedir_t *pagedir_nosave __nosavedata = NULL;
 106 suspend_pagedir_t *pagedir_save;
 107
 108 #define SWSUSP_SIG      "S1SUSPEND"
 109
 110 static struct swsusp_header {
 111         char reserved[PAGE_SIZE - 20 - MAXKEY - MAXIV - sizeof(swp_entry_t)];
 112         u8 key_iv[MAXKEY+MAXIV];
 113         swp_entry_t swsusp_info;
 114         char    orig_sig[10];
 115         char    sig[10];
 116 } __attribute__((packed, aligned(PAGE_SIZE))) swsusp_header;
 117
 118 static struct swsusp_info swsusp_info;
 119
 120 /*
 121  * Saving part...
 122  */
 123
 124 /* We memorize in swapfile_used what swap devices are used for suspension */
 125 #define SWAPFILE_UNUSED    0
 126 #define SWAPFILE_SUSPEND   1    /* This is the suspending device */
 127 #define SWAPFILE_IGNORED   2    /* Those are other swap devices ignored for suspension */
 128
 129 static unsigned short swapfile_used[MAX_SWAPFILES];
 130 static unsigned short root_swap;
 131
 132 static int write_page(unsigned long addr, swp_entry_t * loc);
 133 static int bio_read_page(pgoff_t page_off, void * page);
 134
 135 static u8 key_iv[MAXKEY+MAXIV];
 136
 137 #ifdef CONFIG_SWSUSP_ENCRYPT
 138
 139 static int crypto_init(int mode, void **mem)
 140 {
 141         int error = 0;
 142         int len;
 143         char *modemsg;
 144         struct crypto_tfm *tfm;
 145
 146         modemsg = mode ? "suspend not possible" : "resume not possible";
 147
 148         tfm = crypto_alloc_tfm(CIPHER, CRYPTO_TFM_MODE_CBC);
 149         if(!tfm) {
 150                 printk(KERN_ERR "swsusp: no tfm, %s\n", modemsg);
 151                 error = -EINVAL;
 152                 goto out;
 153         }
 154
 155         if(MAXKEY < crypto_tfm_alg_min_keysize(tfm)) {
 156                 printk(KERN_ERR "swsusp: key buffer too small, %s\n", modemsg);
 157                 error = -ENOKEY;
 158                 goto fail;
 159         }
 160
 161         if (mode)
 162                 get_random_bytes(key_iv, MAXKEY+MAXIV);
 163
 164         len = crypto_tfm_alg_max_keysize(tfm);
 165         if (len > MAXKEY)
 166                 len = MAXKEY;
 167
 168         if (crypto_cipher_setkey(tfm, key_iv, len)) {
 169                 printk(KERN_ERR "swsusp: key setup failure, %s\n", modemsg);
 170                 error = -EKEYREJECTED;
 171                 goto fail;
 172         }
 173
 174         len = crypto_tfm_alg_ivsize(tfm);
 175
 176         if (MAXIV < len) {
 177                 printk(KERN_ERR "swsusp: iv buffer too small, %s\n", modemsg);
 178                 error = -EOVERFLOW;
 179                 goto fail;
 180         }
 181
 182         crypto_cipher_set_iv(tfm, key_iv+MAXKEY, len);
 183
 184         *mem=(void *)tfm;
 185
 186         goto out;
 187
 188 fail:   crypto_free_tfm(tfm);
 189 out:    return error;
 190 }
 191
 192 static __inline__ void crypto_exit(void *mem)
 193 {
 194         crypto_free_tfm((struct crypto_tfm *)mem);
 195 }
 196
 197 static __inline__ int crypto_write(struct pbe *p, void *mem)
 198 {
 199         int error = 0;
 200         struct scatterlist src, dst;
 201
 202         src.page   = virt_to_page(p->address);
 203         src.offset = 0;
 204         src.length = PAGE_SIZE;
 205         dst.page   = virt_to_page((void *)&swsusp_header);
 206         dst.offset = 0;
 207         dst.length = PAGE_SIZE;
 208
 209         error = crypto_cipher_encrypt((struct crypto_tfm *)mem, &dst, &src,
 210                                         PAGE_SIZE);
 211
 212         if (!error)
 213                 error = write_page((unsigned long)&swsusp_header,
 214                                 &(p->swap_address));
 215         return error;
 216 }
 217
 218 static __inline__ int crypto_read(struct pbe *p, void *mem)
 219 {
 220         int error = 0;
 221         struct scatterlist src, dst;
 222
 223         error = bio_read_page(swp_offset(p->swap_address), (void *)p->address);
 224         if (!error) {
 225                 src.offset = 0;
 226                 src.length = PAGE_SIZE;
 227                 dst.offset = 0;
 228                 dst.length = PAGE_SIZE;
 229                 src.page = dst.page = virt_to_page((void *)p->address);
 230
 231                 error = crypto_cipher_decrypt((struct crypto_tfm *)mem, &dst,
 232                                                 &src, PAGE_SIZE);
 233         }
 234         return error;
 235 }
 236 #else
 237 static __inline__ int crypto_init(int mode, void *mem)
 238 {
 239         return 0;
 240 }
 241
 242 static __inline__ void crypto_exit(void *mem)
 243 {
 244 }
 245
 246 static __inline__ int crypto_write(struct pbe *p, void *mem)
 247 {
 248         return write_page(p->address, &(p->swap_address));
 249 }
 250
 251 static __inline__ int crypto_read(struct pbe *p, void *mem)
 252 {
 253         return bio_read_page(swp_offset(p->swap_address), (void *)p->address);
 254 }
 255 #endif
 256
 257 static int mark_swapfiles(swp_entry_t prev)
 258 {
 259         int error;
 260
 261         rw_swap_page_sync(READ,
 262                           swp_entry(root_swap, 0),
 263                           virt_to_page((unsigned long)&swsusp_header));
 264         if (!memcmp("SWAP-SPACE",swsusp_header.sig, 10) ||
 265             !memcmp("SWAPSPACE2",swsusp_header.sig, 10)) {
 266                 memcpy(swsusp_header.orig_sig,swsusp_header.sig, 10);
 267                 memcpy(swsusp_header.sig,SWSUSP_SIG, 10);
 268                 memcpy(swsusp_header.key_iv, key_iv, MAXKEY+MAXIV);
 269                 swsusp_header.swsusp_info = prev;
 270                 error = rw_swap_page_sync(WRITE,
 271                                           swp_entry(root_swap, 0),
 272                                           virt_to_page((unsigned long)
 273                                                        &swsusp_header));
 274         } else {
 275                 pr_debug("swsusp: Partition is not swap space.\n");
 276                 error = -ENODEV;
 277         }
 278         return error;
 279 }
 280
 281 /*
 282  * Check whether the swap device is the specified resume
 283  * device, irrespective of whether they are specified by
 284  * identical names.
 285  *
 286  * (Thus, device inode aliasing is allowed.  You can say /dev/hda4
 287  * instead of /dev/ide/host0/bus0/target0/lun0/part4 [if using devfs]
 288  * and they'll be considered the same device.  This is *necessary* for
 289  * devfs, since the resume code can only recognize the form /dev/hda4,
 290  * but the suspend code would see the long name.)
 291  */
 292 static int is_resume_device(const struct swap_info_struct *swap_info)
 293 {
 294         struct file *file = swap_info->swap_file;
 295         struct inode *inode = file->f_dentry->d_inode;
 296
 297         return S_ISBLK(inode->i_mode) &&
 298                 swsusp_resume_device == MKDEV(imajor(inode), iminor(inode));
 299 }
 300
 301 static int swsusp_swap_check(void) /* This is called before saving image */
 302 {
 303         int i, len;
 304
 305         len=strlen(resume_file);
 306         root_swap = 0xFFFF;
 307
 308         spin_lock(&swap_lock);
 309         for (i=0; i<MAX_SWAPFILES; i++) {
 310                 if (!(swap_info[i].flags & SWP_WRITEOK)) {
 311                         swapfile_used[i]=SWAPFILE_UNUSED;
 312                 } else {
 313                         if (!len) {
 314                                 printk(KERN_WARNING "resume= option should be used to set suspend device" );
 315                                 if (root_swap == 0xFFFF) {
 316                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 317                                         root_swap = i;
 318                                 } else
 319                                         swapfile_used[i] = SWAPFILE_IGNORED;
 320                         } else {
 321                                 /* we ignore all swap devices that are not the resume_file */
 322                                 if (is_resume_device(&swap_info[i])) {
 323                                         swapfile_used[i] = SWAPFILE_SUSPEND;
 324                                         root_swap = i;
 325                                 } else {
 326                                         swapfile_used[i] = SWAPFILE_IGNORED;
 327                                 }
 328                         }
 329                 }
 330         }
 331         spin_unlock(&swap_lock);
 332         return (root_swap != 0xffff) ? 0 : -ENODEV;
 333 }
 334
 335 /**
 336  * This is called after saving image so modification
 337  * will be lost after resume... and that's what we want.
 338  * we make the device unusable. A new call to
 339  * lock_swapdevices can unlock the devices.
 340  */
 341 static void lock_swapdevices(void)
 342 {
 343         int i;
 344
 345         spin_lock(&swap_lock);
 346         for (i = 0; i< MAX_SWAPFILES; i++)
 347                 if (swapfile_used[i] == SWAPFILE_IGNORED) {
 348                         swap_info[i].flags ^= SWP_WRITEOK;
 349                 }
 350         spin_unlock(&swap_lock);
 351 }
 352
 353 /**
 354  *      write_page - Write one page to a fresh swap location.
 355  *      @addr:  Address we're writing.
 356  *      @loc:   Place to store the entry we used.
 357  *
 358  *      Allocate a new swap entry and 'sync' it. Note we discard -EIO
 359  *      errors. That is an artifact left over from swsusp. It did not
 360  *      check the return of rw_swap_page_sync() at all, since most pages
 361  *      written back to swap would return -EIO.
 362  *      This is a partial improvement, since we will at least return other
 363  *      errors, though we need to eventually fix the damn code.
 364  */
 365 static int write_page(unsigned long addr, swp_entry_t * loc)
 366 {
 367         swp_entry_t entry;
 368         int error = 0;
 369
 370         entry = get_swap_page();
 371         if (swp_offset(entry) &&
 372             swapfile_used[swp_type(entry)] == SWAPFILE_SUSPEND) {
 373                 error = rw_swap_page_sync(WRITE, entry,
 374                                           virt_to_page(addr));
 375                 if (error == -EIO)
 376                         error = 0;
 377                 if (!error)
 378                         *loc = entry;
 379         } else
 380                 error = -ENOSPC;
 381         return error;
 382 }
 383
 384 /**
 385  *      data_free - Free the swap entries used by the saved image.
 386  *
 387  *      Walk the list of used swap entries and free each one.
 388  *      This is only used for cleanup when suspend fails.
 389  */
 390 static void data_free(void)
 391 {
 392         swp_entry_t entry;
 393         struct pbe * p;
 394
 395         for_each_pbe(p, pagedir_nosave) {
 396                 entry = p->swap_address;
 397                 if (entry.val)
 398                         swap_free(entry);
 399                 else
 400                         break;
 401         }
 402 }
 403
 404 /**
 405  *      data_write - Write saved image to swap.
 406  *
 407  *      Walk the list of pages in the image and sync each one to swap.
 408  */
 409 static int data_write(void)
 410 {
 411         int error = 0, i = 0;
 412         unsigned int mod = nr_copy_pages / 100;
 413         struct pbe *p;
 414         void *tfm;
 415
 416         if ((error = crypto_init(1, &tfm)))
 417                 return error;
 418
 419         if (!mod)
 420                 mod = 1;
 421
 422         printk( "Writing data to swap (%d pages)...     ", nr_copy_pages );
 423         for_each_pbe (p, pagedir_nosave) {
 424                 if (!(i%mod))
 425                         printk( "\b\b\b\b%3d%%", i / mod );
 426                 if ((error = crypto_write(p, tfm))) {
 427                         crypto_exit(tfm);
 428                         return error;
 429                 }
 430                 i++;
 431         }
 432         printk("\b\b\b\bdone\n");
 433         crypto_exit(tfm);
 434         return error;
 435 }
 436
 437 static void dump_info(void)
 438 {
 439         pr_debug(" swsusp: Version: %u\n",swsusp_info.version_code);
 440         pr_debug(" swsusp: Num Pages: %ld\n",swsusp_info.num_physpages);
 441         pr_debug(" swsusp: UTS Sys: %s\n",swsusp_info.uts.sysname);
 442         pr_debug(" swsusp: UTS Node: %s\n",swsusp_info.uts.nodename);
 443         pr_debug(" swsusp: UTS Release: %s\n",swsusp_info.uts.release);
 444         pr_debug(" swsusp: UTS Version: %s\n",swsusp_info.uts.version);
 445         pr_debug(" swsusp: UTS Machine: %s\n",swsusp_info.uts.machine);
 446         pr_debug(" swsusp: UTS Domain: %s\n",swsusp_info.uts.domainname);
 447         pr_debug(" swsusp: CPUs: %d\n",swsusp_info.cpus);
 448         pr_debug(" swsusp: Image: %ld Pages\n",swsusp_info.image_pages);
 449         pr_debug(" swsusp: Pagedir: %ld Pages\n",swsusp_info.pagedir_pages);
 450 }
 451
 452 static void init_header(void)
 453 {
 454         memset(&swsusp_info, 0, sizeof(swsusp_info));
 455         swsusp_info.version_code = LINUX_VERSION_CODE;
 456         swsusp_info.num_physpages = num_physpages;
 457         memcpy(&swsusp_info.uts, &system_utsname, sizeof(system_utsname));
 458
 459         swsusp_info.suspend_pagedir = pagedir_nosave;
 460         swsusp_info.cpus = num_online_cpus();
 461         swsusp_info.image_pages = nr_copy_pages;
 462 }
 463
 464 static int close_swap(void)
 465 {
 466         swp_entry_t entry;
 467         int error;
 468
 469         dump_info();
 470         error = write_page((unsigned long)&swsusp_info, &entry);
 471         if (!error) {
 472                 printk( "S" );
 473                 error = mark_swapfiles(entry);
 474                 printk( "|\n" );
 475         }
 476         return error;
 477 }
 478
 479 /**
 480  *      free_pagedir_entries - Free pages used by the page directory.
 481  *
 482  *      This is used during suspend for error recovery.
 483  */
 484
 485 static void free_pagedir_entries(void)
 486 {
 487         int i;
 488
 489         for (i = 0; i < swsusp_info.pagedir_pages; i++)
 490                 swap_free(swsusp_info.pagedir[i]);
 491 }
 492
 493
 494 /**
 495  *      write_pagedir - Write the array of pages holding the page directory.
 496  *      @last:  Last swap entry we write (needed for header).
 497  */
 498
 499 static int write_pagedir(void)
 500 {
 501         int error = 0;
 502         unsigned n = 0;
 503         struct pbe * pbe;
 504
 505         printk( "Writing pagedir...");
 506         for_each_pb_page (pbe, pagedir_nosave) {
 507                 if ((error = write_page((unsigned long)pbe, &swsusp_info.pagedir[n++])))
 508                         return error;
 509         }
 510
 511         swsusp_info.pagedir_pages = n;
 512         printk("done (%u pages)\n", n);
 513         return error;
 514 }
 515
 516 /**
 517  *      write_suspend_image - Write entire image and metadata.
 518  *
 519  */
 520 static int write_suspend_image(void)
 521 {
 522         int error;
 523
 524         init_header();
 525         if ((error = data_write()))
 526                 goto FreeData;
 527
 528         if ((error = write_pagedir()))
 529                 goto FreePagedir;
 530
 531         if ((error = close_swap()))
 532                 goto FreePagedir;
 533  Done:
 534         memset(key_iv, 0, MAXKEY+MAXIV);
 535         return error;
 536  FreePagedir:
 537         free_pagedir_entries();
 538  FreeData:
 539         data_free();
 540         goto Done;
 541 }
 542
 543 /**
 544  *      enough_swap - Make sure we have enough swap to save the image.
 545  *
 546  *      Returns TRUE or FALSE after checking the total amount of swap
 547  *      space avaiable.
 548  *
 549  *      FIXME: si_swapinfo(&i) returns all swap devices information.
 550  *      We should only consider resume_device.
 551  */
 552
 553 int enough_swap(unsigned nr_pages)
 554 {
 555         struct sysinfo i;
 556
 557         si_swapinfo(&i);
 558         pr_debug("swsusp: available swap: %lu pages\n", i.freeswap);
 559         return i.freeswap > (nr_pages + PAGES_FOR_IO +
 560                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 561 }
 562
 563
 564 /* It is important _NOT_ to umount filesystems at this point. We want
 565  * them synced (in case something goes wrong) but we DO not want to mark
 566  * filesystem clean: it is not. (And it does not matter, if we resume
 567  * correctly, we'll mark system clean, anyway.)
 568  */
 569 int swsusp_write(void)
 570 {
 571         int error;
 572         device_resume();
 573         lock_swapdevices();
 574         error = write_suspend_image();
 575         /* This will unlock ignored swap devices since writing is finished */
 576         lock_swapdevices();
 577         return error;
 578
 579 }
 580
 581
 582
 583 int swsusp_suspend(void)
 584 {
 585         int error;
 586         if ((error = arch_prepare_suspend()))
 587                 return error;
 588         local_irq_disable();
 589         /* At this point, device_suspend() has been called, but *not*
 590          * device_power_down(). We *must* device_power_down() now.
 591          * Otherwise, drivers for some devices (e.g. interrupt controllers)
 592          * become desynchronized with the actual state of the hardware
 593          * at resume time, and evil weirdness ensues.
 594          */
 595         if ((error = device_power_down(PMSG_FREEZE))) {
 596                 printk(KERN_ERR "Some devices failed to power down, aborting suspend\n");
 597                 local_irq_enable();
 598                 return error;
 599         }
 600
 601         if ((error = swsusp_swap_check())) {
 602                 printk(KERN_ERR "swsusp: cannot find swap device, try swapon -a.\n");
 603                 device_power_up();
 604                 local_irq_enable();
 605                 return error;
 606         }
 607
 608         save_processor_state();
 609         if ((error = swsusp_arch_suspend()))
 610                 printk(KERN_ERR "Error %d suspending\n", error);
 611         /* Restore control flow magically appears here */
 612         restore_processor_state();
 613         restore_highmem();
 614         device_power_up();
 615         local_irq_enable();
 616         return error;
 617 }
 618
 619 int swsusp_resume(void)
 620 {
 621         int error;
 622         local_irq_disable();
 623         if (device_power_down(PMSG_FREEZE))
 624                 printk(KERN_ERR "Some devices failed to power down, very bad\n");
 625         /* We'll ignore saved state, but this gets preempt count (etc) right */
 626         save_processor_state();
 627         error = swsusp_arch_resume();
 628         /* Code below is only ever reached in case of failure. Otherwise
 629          * execution continues at place where swsusp_arch_suspend was called
 630          */
 631         BUG_ON(!error);
 632         /* The only reason why swsusp_arch_resume() can fail is memory being
 633          * very tight, so we have to free it as soon as we can to avoid
 634          * subsequent failures
 635          */
 636         swsusp_free();
 637         restore_processor_state();
 638         restore_highmem();
 639         touch_softlockup_watchdog();
 640         device_power_up();
 641         local_irq_enable();
 642         return error;
 643 }
 644
 645 /**
 646  *      On resume, for storing the PBE list and the image,
 647  *      we can only use memory pages that do not conflict with the pages
 648  *      which had been used before suspend.
 649  *
 650  *      We don't know which pages are usable until we allocate them.
 651  *
 652  *      Allocated but unusable (ie eaten) memory pages are marked so that
 653  *      swsusp_free() can release them
 654  */
 655
 656 unsigned long get_safe_page(gfp_t gfp_mask)
 657 {
 658         unsigned long m;
 659
 660         do {
 661                 m = get_zeroed_page(gfp_mask);
 662                 if (m && PageNosaveFree(virt_to_page(m)))
 663                         /* This is for swsusp_free() */
 664                         SetPageNosave(virt_to_page(m));
 665         } while (m && PageNosaveFree(virt_to_page(m)));
 666         if (m) {
 667                 /* This is for swsusp_free() */
 668                 SetPageNosave(virt_to_page(m));
 669                 SetPageNosaveFree(virt_to_page(m));
 670         }
 671         return m;
 672 }
 673
 674 /**
 675  *      check_pagedir - We ensure here that pages that the PBEs point to
 676  *      won't collide with pages where we're going to restore from the loaded
 677  *      pages later
 678  */
 679
 680 static int check_pagedir(struct pbe *pblist)
 681 {
 682         struct pbe *p;
 683
 684         /* This is necessary, so that we can free allocated pages
 685          * in case of failure
 686          */
 687         for_each_pbe (p, pblist)
 688                 p->address = 0UL;
 689
 690         for_each_pbe (p, pblist) {
 691                 p->address = get_safe_page(GFP_ATOMIC);
 692                 if (!p->address)
 693                         return -ENOMEM;
 694         }
 695         return 0;
 696 }
 697
 698 /**
 699  *      swsusp_pagedir_relocate - It is possible, that some memory pages
 700  *      occupied by the list of PBEs collide with pages where we're going to
 701  *      restore from the loaded pages later.  We relocate them here.
 702  */
 703
 704 static struct pbe * swsusp_pagedir_relocate(struct pbe *pblist)
 705 {
 706         struct zone *zone;
 707         unsigned long zone_pfn;
 708         struct pbe *pbpage, *tail, *p;
 709         void *m;
 710         int rel = 0;
 711
 712         if (!pblist) /* a sanity check */
 713                 return NULL;
 714
 715         pr_debug("swsusp: Relocating pagedir (%lu pages to check)\n",
 716                         swsusp_info.pagedir_pages);
 717
 718         /* Clear page flags */
 719
 720         for_each_zone (zone) {
 721                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 722                         if (pfn_valid(zone_pfn + zone->zone_start_pfn))
 723                                 ClearPageNosaveFree(pfn_to_page(zone_pfn +
 724                                         zone->zone_start_pfn));
 725         }
 726
 727         /* Mark orig addresses */
 728
 729         for_each_pbe (p, pblist)
 730                 SetPageNosaveFree(virt_to_page(p->orig_address));
 731
 732         tail = pblist + PB_PAGE_SKIP;
 733
 734         /* Relocate colliding pages */
 735
 736         for_each_pb_page (pbpage, pblist) {
 737                 if (PageNosaveFree(virt_to_page((unsigned long)pbpage))) {
 738                         m = (void *)get_safe_page(GFP_ATOMIC | __GFP_COLD);
 739                         if (!m)
 740                                 return NULL;
 741                         memcpy(m, (void *)pbpage, PAGE_SIZE);
 742                         if (pbpage == pblist)
 743                                 pblist = (struct pbe *)m;
 744                         else
 745                                 tail->next = (struct pbe *)m;
 746                         pbpage = (struct pbe *)m;
 747
 748                         /* We have to link the PBEs again */
 749                         for (p = pbpage; p < pbpage + PB_PAGE_SKIP; p++)
 750                                 if (p->next) /* needed to save the end */
 751                                         p->next = p + 1;
 752
 753                         rel++;
 754                 }
 755                 tail = pbpage + PB_PAGE_SKIP;
 756         }
 757
 758         /* This is for swsusp_free() */
 759         for_each_pb_page (pbpage, pblist) {
 760                 SetPageNosave(virt_to_page(pbpage));
 761                 SetPageNosaveFree(virt_to_page(pbpage));
 762         }
 763
 764         printk("swsusp: Relocated %d pages\n", rel);
 765
 766         return pblist;
 767 }
 768
 769 /*
 770  *      Using bio to read from swap.
 771  *      This code requires a bit more work than just using buffer heads
 772  *      but, it is the recommended way for 2.5/2.6.
 773  *      The following are to signal the beginning and end of I/O. Bios
 774  *      finish asynchronously, while we want them to happen synchronously.
 775  *      A simple atomic_t, and a wait loop take care of this problem.
 776  */
 777
 778 static atomic_t io_done = ATOMIC_INIT(0);
 779
 780 static int end_io(struct bio * bio, unsigned int num, int err)
 781 {
 782         if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
 783                 panic("I/O error reading memory image");
 784         atomic_set(&io_done, 0);
 785         return 0;
 786 }
 787
 788 static struct block_device * resume_bdev;
 789
 790 /**
 791  *      submit - submit BIO request.
 792  *      @rw:    READ or WRITE.
 793  *      @off    physical offset of page.
 794  *      @page:  page we're reading or writing.
 795  *
 796  *      Straight from the textbook - allocate and initialize the bio.
 797  *      If we're writing, make sure the page is marked as dirty.
 798  *      Then submit it and wait.
 799  */
 800
 801 static int submit(int rw, pgoff_t page_off, void * page)
 802 {
 803         int error = 0;
 804         struct bio * bio;
 805
 806         bio = bio_alloc(GFP_ATOMIC, 1);
 807         if (!bio)
 808                 return -ENOMEM;
 809         bio->bi_sector = page_off * (PAGE_SIZE >> 9);
 810         bio_get(bio);
 811         bio->bi_bdev = resume_bdev;
 812         bio->bi_end_io = end_io;
 813
 814         if (bio_add_page(bio, virt_to_page(page), PAGE_SIZE, 0) < PAGE_SIZE) {
 815                 printk("swsusp: ERROR: adding page to bio at %ld\n",page_off);
 816                 error = -EFAULT;
 817                 goto Done;
 818         }
 819
 820         if (rw == WRITE)
 821                 bio_set_pages_dirty(bio);
 822
 823         atomic_set(&io_done, 1);
 824         submit_bio(rw | (1 << BIO_RW_SYNC), bio);
 825         while (atomic_read(&io_done))
 826                 yield();
 827
 828  Done:
 829         bio_put(bio);
 830         return error;
 831 }
 832
 833 static int bio_read_page(pgoff_t page_off, void * page)
 834 {
 835         return submit(READ, page_off, page);
 836 }
 837
 838 static int bio_write_page(pgoff_t page_off, void * page)
 839 {
 840         return submit(WRITE, page_off, page);
 841 }
 842
 843 /*
 844  * Sanity check if this image makes sense with this kernel/swap context
 845  * I really don't think that it's foolproof but more than nothing..
 846  */
 847
 848 static const char * sanity_check(void)
 849 {
 850         dump_info();
 851         if (swsusp_info.version_code != LINUX_VERSION_CODE)
 852                 return "kernel version";
 853         if (swsusp_info.num_physpages != num_physpages)
 854                 return "memory size";
 855         if (strcmp(swsusp_info.uts.sysname,system_utsname.sysname))
 856                 return "system type";
 857         if (strcmp(swsusp_info.uts.release,system_utsname.release))
 858                 return "kernel release";
 859         if (strcmp(swsusp_info.uts.version,system_utsname.version))
 860                 return "version";
 861         if (strcmp(swsusp_info.uts.machine,system_utsname.machine))
 862                 return "machine";
 863 #if 0
 864         /* We can't use number of online CPUs when we use hotplug to remove them ;-))) */
 865         if (swsusp_info.cpus != num_possible_cpus())
 866                 return "number of cpus";
 867 #endif
 868         return NULL;
 869 }
 870
 871
 872 static int check_header(void)
 873 {
 874         const char * reason = NULL;
 875         int error;
 876
 877         if ((error = bio_read_page(swp_offset(swsusp_header.swsusp_info), &swsusp_info)))
 878                 return error;
 879
 880         /* Is this same machine? */
 881         if ((reason = sanity_check())) {
 882                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n",reason);
 883                 return -EPERM;
 884         }
 885         nr_copy_pages = swsusp_info.image_pages;
 886         return error;
 887 }
 888
 889 static int check_sig(void)
 890 {
 891         int error;
 892
 893         memset(&swsusp_header, 0, sizeof(swsusp_header));
 894         if ((error = bio_read_page(0, &swsusp_header)))
 895                 return error;
 896         if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
 897                 memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
 898                 memcpy(key_iv, swsusp_header.key_iv, MAXKEY+MAXIV);
 899                 memset(swsusp_header.key_iv, 0, MAXKEY+MAXIV);
 900
 901                 /*
 902                  * Reset swap signature now.
 903                  */
 904                 error = bio_write_page(0, &swsusp_header);
 905         } else {
 906                 return -EINVAL;
 907         }
 908         if (!error)
 909                 pr_debug("swsusp: Signature found, resuming\n");
 910         return error;
 911 }
 912
 913 /**
 914  *      data_read - Read image pages from swap.
 915  *
 916  *      You do not need to check for overlaps, check_pagedir()
 917  *      already did that.
 918  */
 919
 920 static int data_read(struct pbe *pblist)
 921 {
 922         struct pbe * p;
 923         int error = 0;
 924         int i = 0;
 925         int mod = swsusp_info.image_pages / 100;
 926         void *tfm;
 927
 928         if ((error = crypto_init(0, &tfm)))
 929                 return error;
 930
 931         if (!mod)
 932                 mod = 1;
 933
 934         printk("swsusp: Reading image data (%lu pages):     ",
 935                         swsusp_info.image_pages);
 936
 937         for_each_pbe (p, pblist) {
 938                 if (!(i % mod))
 939                         printk("\b\b\b\b%3d%%", i / mod);
 940
 941                 if ((error = crypto_read(p, tfm))) {
 942                         crypto_exit(tfm);
 943                         return error;
 944                 }
 945
 946                 i++;
 947         }
 948         printk("\b\b\b\bdone\n");
 949         crypto_exit(tfm);
 950         return error;
 951 }
 952
 953 /**
 954  *      read_pagedir - Read page backup list pages from swap
 955  */
 956
 957 static int read_pagedir(struct pbe *pblist)
 958 {
 959         struct pbe *pbpage, *p;
 960         unsigned i = 0;
 961         int error;
 962
 963         if (!pblist)
 964                 return -EFAULT;
 965
 966         printk("swsusp: Reading pagedir (%lu pages)\n",
 967                         swsusp_info.pagedir_pages);
 968
 969         for_each_pb_page (pbpage, pblist) {
 970                 unsigned long offset = swp_offset(swsusp_info.pagedir[i++]);
 971
 972                 error = -EFAULT;
 973                 if (offset) {
 974                         p = (pbpage + PB_PAGE_SKIP)->next;
 975                         error = bio_read_page(offset, (void *)pbpage);
 976                         (pbpage + PB_PAGE_SKIP)->next = p;
 977                 }
 978                 if (error)
 979                         break;
 980         }
 981
 982         if (!error)
 983                 BUG_ON(i != swsusp_info.pagedir_pages);
 984
 985         return error;
 986 }
 987
 988
 989 static int check_suspend_image(void)
 990 {
 991         int error = 0;
 992
 993         if ((error = check_sig()))
 994                 return error;
 995
 996         if ((error = check_header()))
 997                 return error;
 998
 999         return 0;
1000 }
1001
1002 static int read_suspend_image(void)
1003 {
1004         int error = 0;
1005         struct pbe *p;
1006
1007         if (!(p = alloc_pagedir(nr_copy_pages)))
1008                 return -ENOMEM;
1009
1010         if ((error = read_pagedir(p)))
1011                 return error;
1012
1013         create_pbe_list(p, nr_copy_pages);
1014
1015         if (!(pagedir_nosave = swsusp_pagedir_relocate(p)))
1016                 return -ENOMEM;
1017
1018         /* Allocate memory for the image and read the data from swap */
1019
1020         error = check_pagedir(pagedir_nosave);
1021
1022         if (!error)
1023                 error = data_read(pagedir_nosave);
1024
1025         return error;
1026 }
1027
1028 /**
1029  *      swsusp_check - Check for saved image in swap
1030  */
1031
1032 int swsusp_check(void)
1033 {
1034         int error;
1035
1036         resume_bdev = open_by_devnum(swsusp_resume_device, FMODE_READ);
1037         if (!IS_ERR(resume_bdev)) {
1038                 set_blocksize(resume_bdev, PAGE_SIZE);
1039                 error = check_suspend_image();
1040                 if (error)
1041                     blkdev_put(resume_bdev);
1042         } else
1043                 error = PTR_ERR(resume_bdev);
1044
1045         if (!error)
1046                 pr_debug("swsusp: resume file found\n");
1047         else
1048                 pr_debug("swsusp: Error %d check for resume file\n", error);
1049         return error;
1050 }
1051
1052 /**
1053  *      swsusp_read - Read saved image from swap.
1054  */
1055
1056 int swsusp_read(void)
1057 {
1058         int error;
1059
1060         if (IS_ERR(resume_bdev)) {
1061                 pr_debug("swsusp: block device not initialised\n");
1062                 return PTR_ERR(resume_bdev);
1063         }
1064
1065         error = read_suspend_image();
1066         blkdev_put(resume_bdev);
1067         memset(key_iv, 0, MAXKEY+MAXIV);
1068
1069         if (!error)
1070                 pr_debug("swsusp: Reading resume file was successful\n");
1071         else
1072                 pr_debug("swsusp: Error %d resuming\n", error);
1073         return error;
1074 }
1075
1076 /**
1077  *      swsusp_close - close swap device.
1078  */
1079
1080 void swsusp_close(void)
1081 {
1082         if (IS_ERR(resume_bdev)) {
1083                 pr_debug("swsusp: block device not initialised\n");
1084                 return;
1085         }
1086
1087         blkdev_put(resume_bdev);
1088 }