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