Loading...
1/*
2 * linux/kernel/power/swap.c
3 *
4 * This file provides functions for reading the suspend image from
5 * and writing it to a swap partition.
6 *
7 * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@ucw.cz>
8 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
9 * Copyright (C) 2010-2012 Bojan Smojver <bojan@rexursive.com>
10 *
11 * This file is released under the GPLv2.
12 *
13 */
14
15#include <linux/module.h>
16#include <linux/file.h>
17#include <linux/delay.h>
18#include <linux/bitops.h>
19#include <linux/genhd.h>
20#include <linux/device.h>
21#include <linux/bio.h>
22#include <linux/blkdev.h>
23#include <linux/swap.h>
24#include <linux/swapops.h>
25#include <linux/pm.h>
26#include <linux/slab.h>
27#include <linux/lzo.h>
28#include <linux/vmalloc.h>
29#include <linux/cpumask.h>
30#include <linux/atomic.h>
31#include <linux/kthread.h>
32#include <linux/crc32.h>
33
34#include "power.h"
35
36#define HIBERNATE_SIG "S1SUSPEND"
37
38/*
39 * The swap map is a data structure used for keeping track of each page
40 * written to a swap partition. It consists of many swap_map_page
41 * structures that contain each an array of MAP_PAGE_ENTRIES swap entries.
42 * These structures are stored on the swap and linked together with the
43 * help of the .next_swap member.
44 *
45 * The swap map is created during suspend. The swap map pages are
46 * allocated and populated one at a time, so we only need one memory
47 * page to set up the entire structure.
48 *
49 * During resume we pick up all swap_map_page structures into a list.
50 */
51
52#define MAP_PAGE_ENTRIES (PAGE_SIZE / sizeof(sector_t) - 1)
53
54/*
55 * Number of free pages that are not high.
56 */
57static inline unsigned long low_free_pages(void)
58{
59 return nr_free_pages() - nr_free_highpages();
60}
61
62/*
63 * Number of pages required to be kept free while writing the image. Always
64 * half of all available low pages before the writing starts.
65 */
66static inline unsigned long reqd_free_pages(void)
67{
68 return low_free_pages() / 2;
69}
70
71struct swap_map_page {
72 sector_t entries[MAP_PAGE_ENTRIES];
73 sector_t next_swap;
74};
75
76struct swap_map_page_list {
77 struct swap_map_page *map;
78 struct swap_map_page_list *next;
79};
80
81/**
82 * The swap_map_handle structure is used for handling swap in
83 * a file-alike way
84 */
85
86struct swap_map_handle {
87 struct swap_map_page *cur;
88 struct swap_map_page_list *maps;
89 sector_t cur_swap;
90 sector_t first_sector;
91 unsigned int k;
92 unsigned long reqd_free_pages;
93 u32 crc32;
94};
95
96struct swsusp_header {
97 char reserved[PAGE_SIZE - 20 - sizeof(sector_t) - sizeof(int) -
98 sizeof(u32)];
99 u32 crc32;
100 sector_t image;
101 unsigned int flags; /* Flags to pass to the "boot" kernel */
102 char orig_sig[10];
103 char sig[10];
104} __attribute__((packed));
105
106static struct swsusp_header *swsusp_header;
107
108/**
109 * The following functions are used for tracing the allocated
110 * swap pages, so that they can be freed in case of an error.
111 */
112
113struct swsusp_extent {
114 struct rb_node node;
115 unsigned long start;
116 unsigned long end;
117};
118
119static struct rb_root swsusp_extents = RB_ROOT;
120
121static int swsusp_extents_insert(unsigned long swap_offset)
122{
123 struct rb_node **new = &(swsusp_extents.rb_node);
124 struct rb_node *parent = NULL;
125 struct swsusp_extent *ext;
126
127 /* Figure out where to put the new node */
128 while (*new) {
129 ext = container_of(*new, struct swsusp_extent, node);
130 parent = *new;
131 if (swap_offset < ext->start) {
132 /* Try to merge */
133 if (swap_offset == ext->start - 1) {
134 ext->start--;
135 return 0;
136 }
137 new = &((*new)->rb_left);
138 } else if (swap_offset > ext->end) {
139 /* Try to merge */
140 if (swap_offset == ext->end + 1) {
141 ext->end++;
142 return 0;
143 }
144 new = &((*new)->rb_right);
145 } else {
146 /* It already is in the tree */
147 return -EINVAL;
148 }
149 }
150 /* Add the new node and rebalance the tree. */
151 ext = kzalloc(sizeof(struct swsusp_extent), GFP_KERNEL);
152 if (!ext)
153 return -ENOMEM;
154
155 ext->start = swap_offset;
156 ext->end = swap_offset;
157 rb_link_node(&ext->node, parent, new);
158 rb_insert_color(&ext->node, &swsusp_extents);
159 return 0;
160}
161
162/**
163 * alloc_swapdev_block - allocate a swap page and register that it has
164 * been allocated, so that it can be freed in case of an error.
165 */
166
167sector_t alloc_swapdev_block(int swap)
168{
169 unsigned long offset;
170
171 offset = swp_offset(get_swap_page_of_type(swap));
172 if (offset) {
173 if (swsusp_extents_insert(offset))
174 swap_free(swp_entry(swap, offset));
175 else
176 return swapdev_block(swap, offset);
177 }
178 return 0;
179}
180
181/**
182 * free_all_swap_pages - free swap pages allocated for saving image data.
183 * It also frees the extents used to register which swap entries had been
184 * allocated.
185 */
186
187void free_all_swap_pages(int swap)
188{
189 struct rb_node *node;
190
191 while ((node = swsusp_extents.rb_node)) {
192 struct swsusp_extent *ext;
193 unsigned long offset;
194
195 ext = container_of(node, struct swsusp_extent, node);
196 rb_erase(node, &swsusp_extents);
197 for (offset = ext->start; offset <= ext->end; offset++)
198 swap_free(swp_entry(swap, offset));
199
200 kfree(ext);
201 }
202}
203
204int swsusp_swap_in_use(void)
205{
206 return (swsusp_extents.rb_node != NULL);
207}
208
209/*
210 * General things
211 */
212
213static unsigned short root_swap = 0xffff;
214struct block_device *hib_resume_bdev;
215
216/*
217 * Saving part
218 */
219
220static int mark_swapfiles(struct swap_map_handle *handle, unsigned int flags)
221{
222 int error;
223
224 hib_bio_read_page(swsusp_resume_block, swsusp_header, NULL);
225 if (!memcmp("SWAP-SPACE",swsusp_header->sig, 10) ||
226 !memcmp("SWAPSPACE2",swsusp_header->sig, 10)) {
227 memcpy(swsusp_header->orig_sig,swsusp_header->sig, 10);
228 memcpy(swsusp_header->sig, HIBERNATE_SIG, 10);
229 swsusp_header->image = handle->first_sector;
230 swsusp_header->flags = flags;
231 if (flags & SF_CRC32_MODE)
232 swsusp_header->crc32 = handle->crc32;
233 error = hib_bio_write_page(swsusp_resume_block,
234 swsusp_header, NULL);
235 } else {
236 printk(KERN_ERR "PM: Swap header not found!\n");
237 error = -ENODEV;
238 }
239 return error;
240}
241
242/**
243 * swsusp_swap_check - check if the resume device is a swap device
244 * and get its index (if so)
245 *
246 * This is called before saving image
247 */
248static int swsusp_swap_check(void)
249{
250 int res;
251
252 res = swap_type_of(swsusp_resume_device, swsusp_resume_block,
253 &hib_resume_bdev);
254 if (res < 0)
255 return res;
256
257 root_swap = res;
258 res = blkdev_get(hib_resume_bdev, FMODE_WRITE, NULL);
259 if (res)
260 return res;
261
262 res = set_blocksize(hib_resume_bdev, PAGE_SIZE);
263 if (res < 0)
264 blkdev_put(hib_resume_bdev, FMODE_WRITE);
265
266 return res;
267}
268
269/**
270 * write_page - Write one page to given swap location.
271 * @buf: Address we're writing.
272 * @offset: Offset of the swap page we're writing to.
273 * @bio_chain: Link the next write BIO here
274 */
275
276static int write_page(void *buf, sector_t offset, struct bio **bio_chain)
277{
278 void *src;
279 int ret;
280
281 if (!offset)
282 return -ENOSPC;
283
284 if (bio_chain) {
285 src = (void *)__get_free_page(__GFP_WAIT | __GFP_NOWARN |
286 __GFP_NORETRY);
287 if (src) {
288 copy_page(src, buf);
289 } else {
290 ret = hib_wait_on_bio_chain(bio_chain); /* Free pages */
291 if (ret)
292 return ret;
293 src = (void *)__get_free_page(__GFP_WAIT |
294 __GFP_NOWARN |
295 __GFP_NORETRY);
296 if (src) {
297 copy_page(src, buf);
298 } else {
299 WARN_ON_ONCE(1);
300 bio_chain = NULL; /* Go synchronous */
301 src = buf;
302 }
303 }
304 } else {
305 src = buf;
306 }
307 return hib_bio_write_page(offset, src, bio_chain);
308}
309
310static void release_swap_writer(struct swap_map_handle *handle)
311{
312 if (handle->cur)
313 free_page((unsigned long)handle->cur);
314 handle->cur = NULL;
315}
316
317static int get_swap_writer(struct swap_map_handle *handle)
318{
319 int ret;
320
321 ret = swsusp_swap_check();
322 if (ret) {
323 if (ret != -ENOSPC)
324 printk(KERN_ERR "PM: Cannot find swap device, try "
325 "swapon -a.\n");
326 return ret;
327 }
328 handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_KERNEL);
329 if (!handle->cur) {
330 ret = -ENOMEM;
331 goto err_close;
332 }
333 handle->cur_swap = alloc_swapdev_block(root_swap);
334 if (!handle->cur_swap) {
335 ret = -ENOSPC;
336 goto err_rel;
337 }
338 handle->k = 0;
339 handle->reqd_free_pages = reqd_free_pages();
340 handle->first_sector = handle->cur_swap;
341 return 0;
342err_rel:
343 release_swap_writer(handle);
344err_close:
345 swsusp_close(FMODE_WRITE);
346 return ret;
347}
348
349static int swap_write_page(struct swap_map_handle *handle, void *buf,
350 struct bio **bio_chain)
351{
352 int error = 0;
353 sector_t offset;
354
355 if (!handle->cur)
356 return -EINVAL;
357 offset = alloc_swapdev_block(root_swap);
358 error = write_page(buf, offset, bio_chain);
359 if (error)
360 return error;
361 handle->cur->entries[handle->k++] = offset;
362 if (handle->k >= MAP_PAGE_ENTRIES) {
363 offset = alloc_swapdev_block(root_swap);
364 if (!offset)
365 return -ENOSPC;
366 handle->cur->next_swap = offset;
367 error = write_page(handle->cur, handle->cur_swap, bio_chain);
368 if (error)
369 goto out;
370 clear_page(handle->cur);
371 handle->cur_swap = offset;
372 handle->k = 0;
373
374 if (bio_chain && low_free_pages() <= handle->reqd_free_pages) {
375 error = hib_wait_on_bio_chain(bio_chain);
376 if (error)
377 goto out;
378 /*
379 * Recalculate the number of required free pages, to
380 * make sure we never take more than half.
381 */
382 handle->reqd_free_pages = reqd_free_pages();
383 }
384 }
385 out:
386 return error;
387}
388
389static int flush_swap_writer(struct swap_map_handle *handle)
390{
391 if (handle->cur && handle->cur_swap)
392 return write_page(handle->cur, handle->cur_swap, NULL);
393 else
394 return -EINVAL;
395}
396
397static int swap_writer_finish(struct swap_map_handle *handle,
398 unsigned int flags, int error)
399{
400 if (!error) {
401 flush_swap_writer(handle);
402 printk(KERN_INFO "PM: S");
403 error = mark_swapfiles(handle, flags);
404 printk("|\n");
405 }
406
407 if (error)
408 free_all_swap_pages(root_swap);
409 release_swap_writer(handle);
410 swsusp_close(FMODE_WRITE);
411
412 return error;
413}
414
415/* We need to remember how much compressed data we need to read. */
416#define LZO_HEADER sizeof(size_t)
417
418/* Number of pages/bytes we'll compress at one time. */
419#define LZO_UNC_PAGES 32
420#define LZO_UNC_SIZE (LZO_UNC_PAGES * PAGE_SIZE)
421
422/* Number of pages/bytes we need for compressed data (worst case). */
423#define LZO_CMP_PAGES DIV_ROUND_UP(lzo1x_worst_compress(LZO_UNC_SIZE) + \
424 LZO_HEADER, PAGE_SIZE)
425#define LZO_CMP_SIZE (LZO_CMP_PAGES * PAGE_SIZE)
426
427/* Maximum number of threads for compression/decompression. */
428#define LZO_THREADS 3
429
430/* Minimum/maximum number of pages for read buffering. */
431#define LZO_MIN_RD_PAGES 1024
432#define LZO_MAX_RD_PAGES 8192
433
434
435/**
436 * save_image - save the suspend image data
437 */
438
439static int save_image(struct swap_map_handle *handle,
440 struct snapshot_handle *snapshot,
441 unsigned int nr_to_write)
442{
443 unsigned int m;
444 int ret;
445 int nr_pages;
446 int err2;
447 struct bio *bio;
448 struct timeval start;
449 struct timeval stop;
450
451 printk(KERN_INFO "PM: Saving image data pages (%u pages) ... ",
452 nr_to_write);
453 m = nr_to_write / 100;
454 if (!m)
455 m = 1;
456 nr_pages = 0;
457 bio = NULL;
458 do_gettimeofday(&start);
459 while (1) {
460 ret = snapshot_read_next(snapshot);
461 if (ret <= 0)
462 break;
463 ret = swap_write_page(handle, data_of(*snapshot), &bio);
464 if (ret)
465 break;
466 if (!(nr_pages % m))
467 printk(KERN_CONT "\b\b\b\b%3d%%", nr_pages / m);
468 nr_pages++;
469 }
470 err2 = hib_wait_on_bio_chain(&bio);
471 do_gettimeofday(&stop);
472 if (!ret)
473 ret = err2;
474 if (!ret)
475 printk(KERN_CONT "\b\b\b\bdone\n");
476 else
477 printk(KERN_CONT "\n");
478 swsusp_show_speed(&start, &stop, nr_to_write, "Wrote");
479 return ret;
480}
481
482/**
483 * Structure used for CRC32.
484 */
485struct crc_data {
486 struct task_struct *thr; /* thread */
487 atomic_t ready; /* ready to start flag */
488 atomic_t stop; /* ready to stop flag */
489 unsigned run_threads; /* nr current threads */
490 wait_queue_head_t go; /* start crc update */
491 wait_queue_head_t done; /* crc update done */
492 u32 *crc32; /* points to handle's crc32 */
493 size_t *unc_len[LZO_THREADS]; /* uncompressed lengths */
494 unsigned char *unc[LZO_THREADS]; /* uncompressed data */
495};
496
497/**
498 * CRC32 update function that runs in its own thread.
499 */
500static int crc32_threadfn(void *data)
501{
502 struct crc_data *d = data;
503 unsigned i;
504
505 while (1) {
506 wait_event(d->go, atomic_read(&d->ready) ||
507 kthread_should_stop());
508 if (kthread_should_stop()) {
509 d->thr = NULL;
510 atomic_set(&d->stop, 1);
511 wake_up(&d->done);
512 break;
513 }
514 atomic_set(&d->ready, 0);
515
516 for (i = 0; i < d->run_threads; i++)
517 *d->crc32 = crc32_le(*d->crc32,
518 d->unc[i], *d->unc_len[i]);
519 atomic_set(&d->stop, 1);
520 wake_up(&d->done);
521 }
522 return 0;
523}
524/**
525 * Structure used for LZO data compression.
526 */
527struct cmp_data {
528 struct task_struct *thr; /* thread */
529 atomic_t ready; /* ready to start flag */
530 atomic_t stop; /* ready to stop flag */
531 int ret; /* return code */
532 wait_queue_head_t go; /* start compression */
533 wait_queue_head_t done; /* compression done */
534 size_t unc_len; /* uncompressed length */
535 size_t cmp_len; /* compressed length */
536 unsigned char unc[LZO_UNC_SIZE]; /* uncompressed buffer */
537 unsigned char cmp[LZO_CMP_SIZE]; /* compressed buffer */
538 unsigned char wrk[LZO1X_1_MEM_COMPRESS]; /* compression workspace */
539};
540
541/**
542 * Compression function that runs in its own thread.
543 */
544static int lzo_compress_threadfn(void *data)
545{
546 struct cmp_data *d = data;
547
548 while (1) {
549 wait_event(d->go, atomic_read(&d->ready) ||
550 kthread_should_stop());
551 if (kthread_should_stop()) {
552 d->thr = NULL;
553 d->ret = -1;
554 atomic_set(&d->stop, 1);
555 wake_up(&d->done);
556 break;
557 }
558 atomic_set(&d->ready, 0);
559
560 d->ret = lzo1x_1_compress(d->unc, d->unc_len,
561 d->cmp + LZO_HEADER, &d->cmp_len,
562 d->wrk);
563 atomic_set(&d->stop, 1);
564 wake_up(&d->done);
565 }
566 return 0;
567}
568
569/**
570 * save_image_lzo - Save the suspend image data compressed with LZO.
571 * @handle: Swap mam handle to use for saving the image.
572 * @snapshot: Image to read data from.
573 * @nr_to_write: Number of pages to save.
574 */
575static int save_image_lzo(struct swap_map_handle *handle,
576 struct snapshot_handle *snapshot,
577 unsigned int nr_to_write)
578{
579 unsigned int m;
580 int ret = 0;
581 int nr_pages;
582 int err2;
583 struct bio *bio;
584 struct timeval start;
585 struct timeval stop;
586 size_t off;
587 unsigned thr, run_threads, nr_threads;
588 unsigned char *page = NULL;
589 struct cmp_data *data = NULL;
590 struct crc_data *crc = NULL;
591
592 /*
593 * We'll limit the number of threads for compression to limit memory
594 * footprint.
595 */
596 nr_threads = num_online_cpus() - 1;
597 nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
598
599 page = (void *)__get_free_page(__GFP_WAIT | __GFP_HIGH);
600 if (!page) {
601 printk(KERN_ERR "PM: Failed to allocate LZO page\n");
602 ret = -ENOMEM;
603 goto out_clean;
604 }
605
606 data = vmalloc(sizeof(*data) * nr_threads);
607 if (!data) {
608 printk(KERN_ERR "PM: Failed to allocate LZO data\n");
609 ret = -ENOMEM;
610 goto out_clean;
611 }
612 for (thr = 0; thr < nr_threads; thr++)
613 memset(&data[thr], 0, offsetof(struct cmp_data, go));
614
615 crc = kmalloc(sizeof(*crc), GFP_KERNEL);
616 if (!crc) {
617 printk(KERN_ERR "PM: Failed to allocate crc\n");
618 ret = -ENOMEM;
619 goto out_clean;
620 }
621 memset(crc, 0, offsetof(struct crc_data, go));
622
623 /*
624 * Start the compression threads.
625 */
626 for (thr = 0; thr < nr_threads; thr++) {
627 init_waitqueue_head(&data[thr].go);
628 init_waitqueue_head(&data[thr].done);
629
630 data[thr].thr = kthread_run(lzo_compress_threadfn,
631 &data[thr],
632 "image_compress/%u", thr);
633 if (IS_ERR(data[thr].thr)) {
634 data[thr].thr = NULL;
635 printk(KERN_ERR
636 "PM: Cannot start compression threads\n");
637 ret = -ENOMEM;
638 goto out_clean;
639 }
640 }
641
642 /*
643 * Start the CRC32 thread.
644 */
645 init_waitqueue_head(&crc->go);
646 init_waitqueue_head(&crc->done);
647
648 handle->crc32 = 0;
649 crc->crc32 = &handle->crc32;
650 for (thr = 0; thr < nr_threads; thr++) {
651 crc->unc[thr] = data[thr].unc;
652 crc->unc_len[thr] = &data[thr].unc_len;
653 }
654
655 crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
656 if (IS_ERR(crc->thr)) {
657 crc->thr = NULL;
658 printk(KERN_ERR "PM: Cannot start CRC32 thread\n");
659 ret = -ENOMEM;
660 goto out_clean;
661 }
662
663 /*
664 * Adjust the number of required free pages after all allocations have
665 * been done. We don't want to run out of pages when writing.
666 */
667 handle->reqd_free_pages = reqd_free_pages();
668
669 printk(KERN_INFO
670 "PM: Using %u thread(s) for compression.\n"
671 "PM: Compressing and saving image data (%u pages) ... ",
672 nr_threads, nr_to_write);
673 m = nr_to_write / 100;
674 if (!m)
675 m = 1;
676 nr_pages = 0;
677 bio = NULL;
678 do_gettimeofday(&start);
679 for (;;) {
680 for (thr = 0; thr < nr_threads; thr++) {
681 for (off = 0; off < LZO_UNC_SIZE; off += PAGE_SIZE) {
682 ret = snapshot_read_next(snapshot);
683 if (ret < 0)
684 goto out_finish;
685
686 if (!ret)
687 break;
688
689 memcpy(data[thr].unc + off,
690 data_of(*snapshot), PAGE_SIZE);
691
692 if (!(nr_pages % m))
693 printk(KERN_CONT "\b\b\b\b%3d%%",
694 nr_pages / m);
695 nr_pages++;
696 }
697 if (!off)
698 break;
699
700 data[thr].unc_len = off;
701
702 atomic_set(&data[thr].ready, 1);
703 wake_up(&data[thr].go);
704 }
705
706 if (!thr)
707 break;
708
709 crc->run_threads = thr;
710 atomic_set(&crc->ready, 1);
711 wake_up(&crc->go);
712
713 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
714 wait_event(data[thr].done,
715 atomic_read(&data[thr].stop));
716 atomic_set(&data[thr].stop, 0);
717
718 ret = data[thr].ret;
719
720 if (ret < 0) {
721 printk(KERN_ERR "PM: LZO compression failed\n");
722 goto out_finish;
723 }
724
725 if (unlikely(!data[thr].cmp_len ||
726 data[thr].cmp_len >
727 lzo1x_worst_compress(data[thr].unc_len))) {
728 printk(KERN_ERR
729 "PM: Invalid LZO compressed length\n");
730 ret = -1;
731 goto out_finish;
732 }
733
734 *(size_t *)data[thr].cmp = data[thr].cmp_len;
735
736 /*
737 * Given we are writing one page at a time to disk, we
738 * copy that much from the buffer, although the last
739 * bit will likely be smaller than full page. This is
740 * OK - we saved the length of the compressed data, so
741 * any garbage at the end will be discarded when we
742 * read it.
743 */
744 for (off = 0;
745 off < LZO_HEADER + data[thr].cmp_len;
746 off += PAGE_SIZE) {
747 memcpy(page, data[thr].cmp + off, PAGE_SIZE);
748
749 ret = swap_write_page(handle, page, &bio);
750 if (ret)
751 goto out_finish;
752 }
753 }
754
755 wait_event(crc->done, atomic_read(&crc->stop));
756 atomic_set(&crc->stop, 0);
757 }
758
759out_finish:
760 err2 = hib_wait_on_bio_chain(&bio);
761 do_gettimeofday(&stop);
762 if (!ret)
763 ret = err2;
764 if (!ret) {
765 printk(KERN_CONT "\b\b\b\bdone\n");
766 } else {
767 printk(KERN_CONT "\n");
768 }
769 swsusp_show_speed(&start, &stop, nr_to_write, "Wrote");
770out_clean:
771 if (crc) {
772 if (crc->thr)
773 kthread_stop(crc->thr);
774 kfree(crc);
775 }
776 if (data) {
777 for (thr = 0; thr < nr_threads; thr++)
778 if (data[thr].thr)
779 kthread_stop(data[thr].thr);
780 vfree(data);
781 }
782 if (page) free_page((unsigned long)page);
783
784 return ret;
785}
786
787/**
788 * enough_swap - Make sure we have enough swap to save the image.
789 *
790 * Returns TRUE or FALSE after checking the total amount of swap
791 * space avaiable from the resume partition.
792 */
793
794static int enough_swap(unsigned int nr_pages, unsigned int flags)
795{
796 unsigned int free_swap = count_swap_pages(root_swap, 1);
797 unsigned int required;
798
799 pr_debug("PM: Free swap pages: %u\n", free_swap);
800
801 required = PAGES_FOR_IO + nr_pages;
802 return free_swap > required;
803}
804
805/**
806 * swsusp_write - Write entire image and metadata.
807 * @flags: flags to pass to the "boot" kernel in the image header
808 *
809 * It is important _NOT_ to umount filesystems at this point. We want
810 * them synced (in case something goes wrong) but we DO not want to mark
811 * filesystem clean: it is not. (And it does not matter, if we resume
812 * correctly, we'll mark system clean, anyway.)
813 */
814
815int swsusp_write(unsigned int flags)
816{
817 struct swap_map_handle handle;
818 struct snapshot_handle snapshot;
819 struct swsusp_info *header;
820 unsigned long pages;
821 int error;
822
823 pages = snapshot_get_image_size();
824 error = get_swap_writer(&handle);
825 if (error) {
826 printk(KERN_ERR "PM: Cannot get swap writer\n");
827 return error;
828 }
829 if (flags & SF_NOCOMPRESS_MODE) {
830 if (!enough_swap(pages, flags)) {
831 printk(KERN_ERR "PM: Not enough free swap\n");
832 error = -ENOSPC;
833 goto out_finish;
834 }
835 }
836 memset(&snapshot, 0, sizeof(struct snapshot_handle));
837 error = snapshot_read_next(&snapshot);
838 if (error < PAGE_SIZE) {
839 if (error >= 0)
840 error = -EFAULT;
841
842 goto out_finish;
843 }
844 header = (struct swsusp_info *)data_of(snapshot);
845 error = swap_write_page(&handle, header, NULL);
846 if (!error) {
847 error = (flags & SF_NOCOMPRESS_MODE) ?
848 save_image(&handle, &snapshot, pages - 1) :
849 save_image_lzo(&handle, &snapshot, pages - 1);
850 }
851out_finish:
852 error = swap_writer_finish(&handle, flags, error);
853 return error;
854}
855
856/**
857 * The following functions allow us to read data using a swap map
858 * in a file-alike way
859 */
860
861static void release_swap_reader(struct swap_map_handle *handle)
862{
863 struct swap_map_page_list *tmp;
864
865 while (handle->maps) {
866 if (handle->maps->map)
867 free_page((unsigned long)handle->maps->map);
868 tmp = handle->maps;
869 handle->maps = handle->maps->next;
870 kfree(tmp);
871 }
872 handle->cur = NULL;
873}
874
875static int get_swap_reader(struct swap_map_handle *handle,
876 unsigned int *flags_p)
877{
878 int error;
879 struct swap_map_page_list *tmp, *last;
880 sector_t offset;
881
882 *flags_p = swsusp_header->flags;
883
884 if (!swsusp_header->image) /* how can this happen? */
885 return -EINVAL;
886
887 handle->cur = NULL;
888 last = handle->maps = NULL;
889 offset = swsusp_header->image;
890 while (offset) {
891 tmp = kmalloc(sizeof(*handle->maps), GFP_KERNEL);
892 if (!tmp) {
893 release_swap_reader(handle);
894 return -ENOMEM;
895 }
896 memset(tmp, 0, sizeof(*tmp));
897 if (!handle->maps)
898 handle->maps = tmp;
899 if (last)
900 last->next = tmp;
901 last = tmp;
902
903 tmp->map = (struct swap_map_page *)
904 __get_free_page(__GFP_WAIT | __GFP_HIGH);
905 if (!tmp->map) {
906 release_swap_reader(handle);
907 return -ENOMEM;
908 }
909
910 error = hib_bio_read_page(offset, tmp->map, NULL);
911 if (error) {
912 release_swap_reader(handle);
913 return error;
914 }
915 offset = tmp->map->next_swap;
916 }
917 handle->k = 0;
918 handle->cur = handle->maps->map;
919 return 0;
920}
921
922static int swap_read_page(struct swap_map_handle *handle, void *buf,
923 struct bio **bio_chain)
924{
925 sector_t offset;
926 int error;
927 struct swap_map_page_list *tmp;
928
929 if (!handle->cur)
930 return -EINVAL;
931 offset = handle->cur->entries[handle->k];
932 if (!offset)
933 return -EFAULT;
934 error = hib_bio_read_page(offset, buf, bio_chain);
935 if (error)
936 return error;
937 if (++handle->k >= MAP_PAGE_ENTRIES) {
938 handle->k = 0;
939 free_page((unsigned long)handle->maps->map);
940 tmp = handle->maps;
941 handle->maps = handle->maps->next;
942 kfree(tmp);
943 if (!handle->maps)
944 release_swap_reader(handle);
945 else
946 handle->cur = handle->maps->map;
947 }
948 return error;
949}
950
951static int swap_reader_finish(struct swap_map_handle *handle)
952{
953 release_swap_reader(handle);
954
955 return 0;
956}
957
958/**
959 * load_image - load the image using the swap map handle
960 * @handle and the snapshot handle @snapshot
961 * (assume there are @nr_pages pages to load)
962 */
963
964static int load_image(struct swap_map_handle *handle,
965 struct snapshot_handle *snapshot,
966 unsigned int nr_to_read)
967{
968 unsigned int m;
969 int ret = 0;
970 struct timeval start;
971 struct timeval stop;
972 struct bio *bio;
973 int err2;
974 unsigned nr_pages;
975
976 printk(KERN_INFO "PM: Loading image data pages (%u pages) ... ",
977 nr_to_read);
978 m = nr_to_read / 100;
979 if (!m)
980 m = 1;
981 nr_pages = 0;
982 bio = NULL;
983 do_gettimeofday(&start);
984 for ( ; ; ) {
985 ret = snapshot_write_next(snapshot);
986 if (ret <= 0)
987 break;
988 ret = swap_read_page(handle, data_of(*snapshot), &bio);
989 if (ret)
990 break;
991 if (snapshot->sync_read)
992 ret = hib_wait_on_bio_chain(&bio);
993 if (ret)
994 break;
995 if (!(nr_pages % m))
996 printk("\b\b\b\b%3d%%", nr_pages / m);
997 nr_pages++;
998 }
999 err2 = hib_wait_on_bio_chain(&bio);
1000 do_gettimeofday(&stop);
1001 if (!ret)
1002 ret = err2;
1003 if (!ret) {
1004 printk("\b\b\b\bdone\n");
1005 snapshot_write_finalize(snapshot);
1006 if (!snapshot_image_loaded(snapshot))
1007 ret = -ENODATA;
1008 } else
1009 printk("\n");
1010 swsusp_show_speed(&start, &stop, nr_to_read, "Read");
1011 return ret;
1012}
1013
1014/**
1015 * Structure used for LZO data decompression.
1016 */
1017struct dec_data {
1018 struct task_struct *thr; /* thread */
1019 atomic_t ready; /* ready to start flag */
1020 atomic_t stop; /* ready to stop flag */
1021 int ret; /* return code */
1022 wait_queue_head_t go; /* start decompression */
1023 wait_queue_head_t done; /* decompression done */
1024 size_t unc_len; /* uncompressed length */
1025 size_t cmp_len; /* compressed length */
1026 unsigned char unc[LZO_UNC_SIZE]; /* uncompressed buffer */
1027 unsigned char cmp[LZO_CMP_SIZE]; /* compressed buffer */
1028};
1029
1030/**
1031 * Deompression function that runs in its own thread.
1032 */
1033static int lzo_decompress_threadfn(void *data)
1034{
1035 struct dec_data *d = data;
1036
1037 while (1) {
1038 wait_event(d->go, atomic_read(&d->ready) ||
1039 kthread_should_stop());
1040 if (kthread_should_stop()) {
1041 d->thr = NULL;
1042 d->ret = -1;
1043 atomic_set(&d->stop, 1);
1044 wake_up(&d->done);
1045 break;
1046 }
1047 atomic_set(&d->ready, 0);
1048
1049 d->unc_len = LZO_UNC_SIZE;
1050 d->ret = lzo1x_decompress_safe(d->cmp + LZO_HEADER, d->cmp_len,
1051 d->unc, &d->unc_len);
1052 atomic_set(&d->stop, 1);
1053 wake_up(&d->done);
1054 }
1055 return 0;
1056}
1057
1058/**
1059 * load_image_lzo - Load compressed image data and decompress them with LZO.
1060 * @handle: Swap map handle to use for loading data.
1061 * @snapshot: Image to copy uncompressed data into.
1062 * @nr_to_read: Number of pages to load.
1063 */
1064static int load_image_lzo(struct swap_map_handle *handle,
1065 struct snapshot_handle *snapshot,
1066 unsigned int nr_to_read)
1067{
1068 unsigned int m;
1069 int ret = 0;
1070 int eof = 0;
1071 struct bio *bio;
1072 struct timeval start;
1073 struct timeval stop;
1074 unsigned nr_pages;
1075 size_t off;
1076 unsigned i, thr, run_threads, nr_threads;
1077 unsigned ring = 0, pg = 0, ring_size = 0,
1078 have = 0, want, need, asked = 0;
1079 unsigned long read_pages = 0;
1080 unsigned char **page = NULL;
1081 struct dec_data *data = NULL;
1082 struct crc_data *crc = NULL;
1083
1084 /*
1085 * We'll limit the number of threads for decompression to limit memory
1086 * footprint.
1087 */
1088 nr_threads = num_online_cpus() - 1;
1089 nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
1090
1091 page = vmalloc(sizeof(*page) * LZO_MAX_RD_PAGES);
1092 if (!page) {
1093 printk(KERN_ERR "PM: Failed to allocate LZO page\n");
1094 ret = -ENOMEM;
1095 goto out_clean;
1096 }
1097
1098 data = vmalloc(sizeof(*data) * nr_threads);
1099 if (!data) {
1100 printk(KERN_ERR "PM: Failed to allocate LZO data\n");
1101 ret = -ENOMEM;
1102 goto out_clean;
1103 }
1104 for (thr = 0; thr < nr_threads; thr++)
1105 memset(&data[thr], 0, offsetof(struct dec_data, go));
1106
1107 crc = kmalloc(sizeof(*crc), GFP_KERNEL);
1108 if (!crc) {
1109 printk(KERN_ERR "PM: Failed to allocate crc\n");
1110 ret = -ENOMEM;
1111 goto out_clean;
1112 }
1113 memset(crc, 0, offsetof(struct crc_data, go));
1114
1115 /*
1116 * Start the decompression threads.
1117 */
1118 for (thr = 0; thr < nr_threads; thr++) {
1119 init_waitqueue_head(&data[thr].go);
1120 init_waitqueue_head(&data[thr].done);
1121
1122 data[thr].thr = kthread_run(lzo_decompress_threadfn,
1123 &data[thr],
1124 "image_decompress/%u", thr);
1125 if (IS_ERR(data[thr].thr)) {
1126 data[thr].thr = NULL;
1127 printk(KERN_ERR
1128 "PM: Cannot start decompression threads\n");
1129 ret = -ENOMEM;
1130 goto out_clean;
1131 }
1132 }
1133
1134 /*
1135 * Start the CRC32 thread.
1136 */
1137 init_waitqueue_head(&crc->go);
1138 init_waitqueue_head(&crc->done);
1139
1140 handle->crc32 = 0;
1141 crc->crc32 = &handle->crc32;
1142 for (thr = 0; thr < nr_threads; thr++) {
1143 crc->unc[thr] = data[thr].unc;
1144 crc->unc_len[thr] = &data[thr].unc_len;
1145 }
1146
1147 crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
1148 if (IS_ERR(crc->thr)) {
1149 crc->thr = NULL;
1150 printk(KERN_ERR "PM: Cannot start CRC32 thread\n");
1151 ret = -ENOMEM;
1152 goto out_clean;
1153 }
1154
1155 /*
1156 * Set the number of pages for read buffering.
1157 * This is complete guesswork, because we'll only know the real
1158 * picture once prepare_image() is called, which is much later on
1159 * during the image load phase. We'll assume the worst case and
1160 * say that none of the image pages are from high memory.
1161 */
1162 if (low_free_pages() > snapshot_get_image_size())
1163 read_pages = (low_free_pages() - snapshot_get_image_size()) / 2;
1164 read_pages = clamp_val(read_pages, LZO_MIN_RD_PAGES, LZO_MAX_RD_PAGES);
1165
1166 for (i = 0; i < read_pages; i++) {
1167 page[i] = (void *)__get_free_page(i < LZO_CMP_PAGES ?
1168 __GFP_WAIT | __GFP_HIGH :
1169 __GFP_WAIT | __GFP_NOWARN |
1170 __GFP_NORETRY);
1171
1172 if (!page[i]) {
1173 if (i < LZO_CMP_PAGES) {
1174 ring_size = i;
1175 printk(KERN_ERR
1176 "PM: Failed to allocate LZO pages\n");
1177 ret = -ENOMEM;
1178 goto out_clean;
1179 } else {
1180 break;
1181 }
1182 }
1183 }
1184 want = ring_size = i;
1185
1186 printk(KERN_INFO
1187 "PM: Using %u thread(s) for decompression.\n"
1188 "PM: Loading and decompressing image data (%u pages) ... ",
1189 nr_threads, nr_to_read);
1190 m = nr_to_read / 100;
1191 if (!m)
1192 m = 1;
1193 nr_pages = 0;
1194 bio = NULL;
1195 do_gettimeofday(&start);
1196
1197 ret = snapshot_write_next(snapshot);
1198 if (ret <= 0)
1199 goto out_finish;
1200
1201 for(;;) {
1202 for (i = 0; !eof && i < want; i++) {
1203 ret = swap_read_page(handle, page[ring], &bio);
1204 if (ret) {
1205 /*
1206 * On real read error, finish. On end of data,
1207 * set EOF flag and just exit the read loop.
1208 */
1209 if (handle->cur &&
1210 handle->cur->entries[handle->k]) {
1211 goto out_finish;
1212 } else {
1213 eof = 1;
1214 break;
1215 }
1216 }
1217 if (++ring >= ring_size)
1218 ring = 0;
1219 }
1220 asked += i;
1221 want -= i;
1222
1223 /*
1224 * We are out of data, wait for some more.
1225 */
1226 if (!have) {
1227 if (!asked)
1228 break;
1229
1230 ret = hib_wait_on_bio_chain(&bio);
1231 if (ret)
1232 goto out_finish;
1233 have += asked;
1234 asked = 0;
1235 if (eof)
1236 eof = 2;
1237 }
1238
1239 if (crc->run_threads) {
1240 wait_event(crc->done, atomic_read(&crc->stop));
1241 atomic_set(&crc->stop, 0);
1242 crc->run_threads = 0;
1243 }
1244
1245 for (thr = 0; have && thr < nr_threads; thr++) {
1246 data[thr].cmp_len = *(size_t *)page[pg];
1247 if (unlikely(!data[thr].cmp_len ||
1248 data[thr].cmp_len >
1249 lzo1x_worst_compress(LZO_UNC_SIZE))) {
1250 printk(KERN_ERR
1251 "PM: Invalid LZO compressed length\n");
1252 ret = -1;
1253 goto out_finish;
1254 }
1255
1256 need = DIV_ROUND_UP(data[thr].cmp_len + LZO_HEADER,
1257 PAGE_SIZE);
1258 if (need > have) {
1259 if (eof > 1) {
1260 ret = -1;
1261 goto out_finish;
1262 }
1263 break;
1264 }
1265
1266 for (off = 0;
1267 off < LZO_HEADER + data[thr].cmp_len;
1268 off += PAGE_SIZE) {
1269 memcpy(data[thr].cmp + off,
1270 page[pg], PAGE_SIZE);
1271 have--;
1272 want++;
1273 if (++pg >= ring_size)
1274 pg = 0;
1275 }
1276
1277 atomic_set(&data[thr].ready, 1);
1278 wake_up(&data[thr].go);
1279 }
1280
1281 /*
1282 * Wait for more data while we are decompressing.
1283 */
1284 if (have < LZO_CMP_PAGES && asked) {
1285 ret = hib_wait_on_bio_chain(&bio);
1286 if (ret)
1287 goto out_finish;
1288 have += asked;
1289 asked = 0;
1290 if (eof)
1291 eof = 2;
1292 }
1293
1294 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
1295 wait_event(data[thr].done,
1296 atomic_read(&data[thr].stop));
1297 atomic_set(&data[thr].stop, 0);
1298
1299 ret = data[thr].ret;
1300
1301 if (ret < 0) {
1302 printk(KERN_ERR
1303 "PM: LZO decompression failed\n");
1304 goto out_finish;
1305 }
1306
1307 if (unlikely(!data[thr].unc_len ||
1308 data[thr].unc_len > LZO_UNC_SIZE ||
1309 data[thr].unc_len & (PAGE_SIZE - 1))) {
1310 printk(KERN_ERR
1311 "PM: Invalid LZO uncompressed length\n");
1312 ret = -1;
1313 goto out_finish;
1314 }
1315
1316 for (off = 0;
1317 off < data[thr].unc_len; off += PAGE_SIZE) {
1318 memcpy(data_of(*snapshot),
1319 data[thr].unc + off, PAGE_SIZE);
1320
1321 if (!(nr_pages % m))
1322 printk("\b\b\b\b%3d%%", nr_pages / m);
1323 nr_pages++;
1324
1325 ret = snapshot_write_next(snapshot);
1326 if (ret <= 0) {
1327 crc->run_threads = thr + 1;
1328 atomic_set(&crc->ready, 1);
1329 wake_up(&crc->go);
1330 goto out_finish;
1331 }
1332 }
1333 }
1334
1335 crc->run_threads = thr;
1336 atomic_set(&crc->ready, 1);
1337 wake_up(&crc->go);
1338 }
1339
1340out_finish:
1341 if (crc->run_threads) {
1342 wait_event(crc->done, atomic_read(&crc->stop));
1343 atomic_set(&crc->stop, 0);
1344 }
1345 do_gettimeofday(&stop);
1346 if (!ret) {
1347 printk("\b\b\b\bdone\n");
1348 snapshot_write_finalize(snapshot);
1349 if (!snapshot_image_loaded(snapshot))
1350 ret = -ENODATA;
1351 if (!ret) {
1352 if (swsusp_header->flags & SF_CRC32_MODE) {
1353 if(handle->crc32 != swsusp_header->crc32) {
1354 printk(KERN_ERR
1355 "PM: Invalid image CRC32!\n");
1356 ret = -ENODATA;
1357 }
1358 }
1359 }
1360 } else
1361 printk("\n");
1362 swsusp_show_speed(&start, &stop, nr_to_read, "Read");
1363out_clean:
1364 for (i = 0; i < ring_size; i++)
1365 free_page((unsigned long)page[i]);
1366 if (crc) {
1367 if (crc->thr)
1368 kthread_stop(crc->thr);
1369 kfree(crc);
1370 }
1371 if (data) {
1372 for (thr = 0; thr < nr_threads; thr++)
1373 if (data[thr].thr)
1374 kthread_stop(data[thr].thr);
1375 vfree(data);
1376 }
1377 if (page) vfree(page);
1378
1379 return ret;
1380}
1381
1382/**
1383 * swsusp_read - read the hibernation image.
1384 * @flags_p: flags passed by the "frozen" kernel in the image header should
1385 * be written into this memory location
1386 */
1387
1388int swsusp_read(unsigned int *flags_p)
1389{
1390 int error;
1391 struct swap_map_handle handle;
1392 struct snapshot_handle snapshot;
1393 struct swsusp_info *header;
1394
1395 memset(&snapshot, 0, sizeof(struct snapshot_handle));
1396 error = snapshot_write_next(&snapshot);
1397 if (error < PAGE_SIZE)
1398 return error < 0 ? error : -EFAULT;
1399 header = (struct swsusp_info *)data_of(snapshot);
1400 error = get_swap_reader(&handle, flags_p);
1401 if (error)
1402 goto end;
1403 if (!error)
1404 error = swap_read_page(&handle, header, NULL);
1405 if (!error) {
1406 error = (*flags_p & SF_NOCOMPRESS_MODE) ?
1407 load_image(&handle, &snapshot, header->pages - 1) :
1408 load_image_lzo(&handle, &snapshot, header->pages - 1);
1409 }
1410 swap_reader_finish(&handle);
1411end:
1412 if (!error)
1413 pr_debug("PM: Image successfully loaded\n");
1414 else
1415 pr_debug("PM: Error %d resuming\n", error);
1416 return error;
1417}
1418
1419/**
1420 * swsusp_check - Check for swsusp signature in the resume device
1421 */
1422
1423int swsusp_check(void)
1424{
1425 int error;
1426
1427 hib_resume_bdev = blkdev_get_by_dev(swsusp_resume_device,
1428 FMODE_READ, NULL);
1429 if (!IS_ERR(hib_resume_bdev)) {
1430 set_blocksize(hib_resume_bdev, PAGE_SIZE);
1431 clear_page(swsusp_header);
1432 error = hib_bio_read_page(swsusp_resume_block,
1433 swsusp_header, NULL);
1434 if (error)
1435 goto put;
1436
1437 if (!memcmp(HIBERNATE_SIG, swsusp_header->sig, 10)) {
1438 memcpy(swsusp_header->sig, swsusp_header->orig_sig, 10);
1439 /* Reset swap signature now */
1440 error = hib_bio_write_page(swsusp_resume_block,
1441 swsusp_header, NULL);
1442 } else {
1443 error = -EINVAL;
1444 }
1445
1446put:
1447 if (error)
1448 blkdev_put(hib_resume_bdev, FMODE_READ);
1449 else
1450 pr_debug("PM: Image signature found, resuming\n");
1451 } else {
1452 error = PTR_ERR(hib_resume_bdev);
1453 }
1454
1455 if (error)
1456 pr_debug("PM: Image not found (code %d)\n", error);
1457
1458 return error;
1459}
1460
1461/**
1462 * swsusp_close - close swap device.
1463 */
1464
1465void swsusp_close(fmode_t mode)
1466{
1467 if (IS_ERR(hib_resume_bdev)) {
1468 pr_debug("PM: Image device not initialised\n");
1469 return;
1470 }
1471
1472 blkdev_put(hib_resume_bdev, mode);
1473}
1474
1475static int swsusp_header_init(void)
1476{
1477 swsusp_header = (struct swsusp_header*) __get_free_page(GFP_KERNEL);
1478 if (!swsusp_header)
1479 panic("Could not allocate memory for swsusp_header\n");
1480 return 0;
1481}
1482
1483core_initcall(swsusp_header_init);
1/*
2 * linux/kernel/power/swap.c
3 *
4 * This file provides functions for reading the suspend image from
5 * and writing it to a swap partition.
6 *
7 * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@ucw.cz>
8 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
9 * Copyright (C) 2010-2012 Bojan Smojver <bojan@rexursive.com>
10 *
11 * This file is released under the GPLv2.
12 *
13 */
14
15#include <linux/module.h>
16#include <linux/file.h>
17#include <linux/delay.h>
18#include <linux/bitops.h>
19#include <linux/genhd.h>
20#include <linux/device.h>
21#include <linux/bio.h>
22#include <linux/blkdev.h>
23#include <linux/swap.h>
24#include <linux/swapops.h>
25#include <linux/pm.h>
26#include <linux/slab.h>
27#include <linux/lzo.h>
28#include <linux/vmalloc.h>
29#include <linux/cpumask.h>
30#include <linux/atomic.h>
31#include <linux/kthread.h>
32#include <linux/crc32.h>
33#include <linux/ktime.h>
34
35#include "power.h"
36
37#define HIBERNATE_SIG "S1SUSPEND"
38
39/*
40 * When reading an {un,}compressed image, we may restore pages in place,
41 * in which case some architectures need these pages cleaning before they
42 * can be executed. We don't know which pages these may be, so clean the lot.
43 */
44static bool clean_pages_on_read;
45static bool clean_pages_on_decompress;
46
47/*
48 * The swap map is a data structure used for keeping track of each page
49 * written to a swap partition. It consists of many swap_map_page
50 * structures that contain each an array of MAP_PAGE_ENTRIES swap entries.
51 * These structures are stored on the swap and linked together with the
52 * help of the .next_swap member.
53 *
54 * The swap map is created during suspend. The swap map pages are
55 * allocated and populated one at a time, so we only need one memory
56 * page to set up the entire structure.
57 *
58 * During resume we pick up all swap_map_page structures into a list.
59 */
60
61#define MAP_PAGE_ENTRIES (PAGE_SIZE / sizeof(sector_t) - 1)
62
63/*
64 * Number of free pages that are not high.
65 */
66static inline unsigned long low_free_pages(void)
67{
68 return nr_free_pages() - nr_free_highpages();
69}
70
71/*
72 * Number of pages required to be kept free while writing the image. Always
73 * half of all available low pages before the writing starts.
74 */
75static inline unsigned long reqd_free_pages(void)
76{
77 return low_free_pages() / 2;
78}
79
80struct swap_map_page {
81 sector_t entries[MAP_PAGE_ENTRIES];
82 sector_t next_swap;
83};
84
85struct swap_map_page_list {
86 struct swap_map_page *map;
87 struct swap_map_page_list *next;
88};
89
90/**
91 * The swap_map_handle structure is used for handling swap in
92 * a file-alike way
93 */
94
95struct swap_map_handle {
96 struct swap_map_page *cur;
97 struct swap_map_page_list *maps;
98 sector_t cur_swap;
99 sector_t first_sector;
100 unsigned int k;
101 unsigned long reqd_free_pages;
102 u32 crc32;
103};
104
105struct swsusp_header {
106 char reserved[PAGE_SIZE - 20 - sizeof(sector_t) - sizeof(int) -
107 sizeof(u32)];
108 u32 crc32;
109 sector_t image;
110 unsigned int flags; /* Flags to pass to the "boot" kernel */
111 char orig_sig[10];
112 char sig[10];
113} __packed;
114
115static struct swsusp_header *swsusp_header;
116
117/**
118 * The following functions are used for tracing the allocated
119 * swap pages, so that they can be freed in case of an error.
120 */
121
122struct swsusp_extent {
123 struct rb_node node;
124 unsigned long start;
125 unsigned long end;
126};
127
128static struct rb_root swsusp_extents = RB_ROOT;
129
130static int swsusp_extents_insert(unsigned long swap_offset)
131{
132 struct rb_node **new = &(swsusp_extents.rb_node);
133 struct rb_node *parent = NULL;
134 struct swsusp_extent *ext;
135
136 /* Figure out where to put the new node */
137 while (*new) {
138 ext = rb_entry(*new, struct swsusp_extent, node);
139 parent = *new;
140 if (swap_offset < ext->start) {
141 /* Try to merge */
142 if (swap_offset == ext->start - 1) {
143 ext->start--;
144 return 0;
145 }
146 new = &((*new)->rb_left);
147 } else if (swap_offset > ext->end) {
148 /* Try to merge */
149 if (swap_offset == ext->end + 1) {
150 ext->end++;
151 return 0;
152 }
153 new = &((*new)->rb_right);
154 } else {
155 /* It already is in the tree */
156 return -EINVAL;
157 }
158 }
159 /* Add the new node and rebalance the tree. */
160 ext = kzalloc(sizeof(struct swsusp_extent), GFP_KERNEL);
161 if (!ext)
162 return -ENOMEM;
163
164 ext->start = swap_offset;
165 ext->end = swap_offset;
166 rb_link_node(&ext->node, parent, new);
167 rb_insert_color(&ext->node, &swsusp_extents);
168 return 0;
169}
170
171/**
172 * alloc_swapdev_block - allocate a swap page and register that it has
173 * been allocated, so that it can be freed in case of an error.
174 */
175
176sector_t alloc_swapdev_block(int swap)
177{
178 unsigned long offset;
179
180 offset = swp_offset(get_swap_page_of_type(swap));
181 if (offset) {
182 if (swsusp_extents_insert(offset))
183 swap_free(swp_entry(swap, offset));
184 else
185 return swapdev_block(swap, offset);
186 }
187 return 0;
188}
189
190/**
191 * free_all_swap_pages - free swap pages allocated for saving image data.
192 * It also frees the extents used to register which swap entries had been
193 * allocated.
194 */
195
196void free_all_swap_pages(int swap)
197{
198 struct rb_node *node;
199
200 while ((node = swsusp_extents.rb_node)) {
201 struct swsusp_extent *ext;
202 unsigned long offset;
203
204 ext = container_of(node, struct swsusp_extent, node);
205 rb_erase(node, &swsusp_extents);
206 for (offset = ext->start; offset <= ext->end; offset++)
207 swap_free(swp_entry(swap, offset));
208
209 kfree(ext);
210 }
211}
212
213int swsusp_swap_in_use(void)
214{
215 return (swsusp_extents.rb_node != NULL);
216}
217
218/*
219 * General things
220 */
221
222static unsigned short root_swap = 0xffff;
223static struct block_device *hib_resume_bdev;
224
225struct hib_bio_batch {
226 atomic_t count;
227 wait_queue_head_t wait;
228 int error;
229};
230
231static void hib_init_batch(struct hib_bio_batch *hb)
232{
233 atomic_set(&hb->count, 0);
234 init_waitqueue_head(&hb->wait);
235 hb->error = 0;
236}
237
238static void hib_end_io(struct bio *bio)
239{
240 struct hib_bio_batch *hb = bio->bi_private;
241 struct page *page = bio->bi_io_vec[0].bv_page;
242
243 if (bio->bi_error) {
244 printk(KERN_ALERT "Read-error on swap-device (%u:%u:%Lu)\n",
245 imajor(bio->bi_bdev->bd_inode),
246 iminor(bio->bi_bdev->bd_inode),
247 (unsigned long long)bio->bi_iter.bi_sector);
248 }
249
250 if (bio_data_dir(bio) == WRITE)
251 put_page(page);
252 else if (clean_pages_on_read)
253 flush_icache_range((unsigned long)page_address(page),
254 (unsigned long)page_address(page) + PAGE_SIZE);
255
256 if (bio->bi_error && !hb->error)
257 hb->error = bio->bi_error;
258 if (atomic_dec_and_test(&hb->count))
259 wake_up(&hb->wait);
260
261 bio_put(bio);
262}
263
264static int hib_submit_io(int op, int op_flags, pgoff_t page_off, void *addr,
265 struct hib_bio_batch *hb)
266{
267 struct page *page = virt_to_page(addr);
268 struct bio *bio;
269 int error = 0;
270
271 bio = bio_alloc(__GFP_RECLAIM | __GFP_HIGH, 1);
272 bio->bi_iter.bi_sector = page_off * (PAGE_SIZE >> 9);
273 bio->bi_bdev = hib_resume_bdev;
274 bio_set_op_attrs(bio, op, op_flags);
275
276 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
277 printk(KERN_ERR "PM: Adding page to bio failed at %llu\n",
278 (unsigned long long)bio->bi_iter.bi_sector);
279 bio_put(bio);
280 return -EFAULT;
281 }
282
283 if (hb) {
284 bio->bi_end_io = hib_end_io;
285 bio->bi_private = hb;
286 atomic_inc(&hb->count);
287 submit_bio(bio);
288 } else {
289 error = submit_bio_wait(bio);
290 bio_put(bio);
291 }
292
293 return error;
294}
295
296static int hib_wait_io(struct hib_bio_batch *hb)
297{
298 wait_event(hb->wait, atomic_read(&hb->count) == 0);
299 return hb->error;
300}
301
302/*
303 * Saving part
304 */
305
306static int mark_swapfiles(struct swap_map_handle *handle, unsigned int flags)
307{
308 int error;
309
310 hib_submit_io(REQ_OP_READ, 0, swsusp_resume_block,
311 swsusp_header, NULL);
312 if (!memcmp("SWAP-SPACE",swsusp_header->sig, 10) ||
313 !memcmp("SWAPSPACE2",swsusp_header->sig, 10)) {
314 memcpy(swsusp_header->orig_sig,swsusp_header->sig, 10);
315 memcpy(swsusp_header->sig, HIBERNATE_SIG, 10);
316 swsusp_header->image = handle->first_sector;
317 swsusp_header->flags = flags;
318 if (flags & SF_CRC32_MODE)
319 swsusp_header->crc32 = handle->crc32;
320 error = hib_submit_io(REQ_OP_WRITE, REQ_SYNC,
321 swsusp_resume_block, swsusp_header, NULL);
322 } else {
323 printk(KERN_ERR "PM: Swap header not found!\n");
324 error = -ENODEV;
325 }
326 return error;
327}
328
329/**
330 * swsusp_swap_check - check if the resume device is a swap device
331 * and get its index (if so)
332 *
333 * This is called before saving image
334 */
335static int swsusp_swap_check(void)
336{
337 int res;
338
339 res = swap_type_of(swsusp_resume_device, swsusp_resume_block,
340 &hib_resume_bdev);
341 if (res < 0)
342 return res;
343
344 root_swap = res;
345 res = blkdev_get(hib_resume_bdev, FMODE_WRITE, NULL);
346 if (res)
347 return res;
348
349 res = set_blocksize(hib_resume_bdev, PAGE_SIZE);
350 if (res < 0)
351 blkdev_put(hib_resume_bdev, FMODE_WRITE);
352
353 /*
354 * Update the resume device to the one actually used,
355 * so the test_resume mode can use it in case it is
356 * invoked from hibernate() to test the snapshot.
357 */
358 swsusp_resume_device = hib_resume_bdev->bd_dev;
359 return res;
360}
361
362/**
363 * write_page - Write one page to given swap location.
364 * @buf: Address we're writing.
365 * @offset: Offset of the swap page we're writing to.
366 * @hb: bio completion batch
367 */
368
369static int write_page(void *buf, sector_t offset, struct hib_bio_batch *hb)
370{
371 void *src;
372 int ret;
373
374 if (!offset)
375 return -ENOSPC;
376
377 if (hb) {
378 src = (void *)__get_free_page(__GFP_RECLAIM | __GFP_NOWARN |
379 __GFP_NORETRY);
380 if (src) {
381 copy_page(src, buf);
382 } else {
383 ret = hib_wait_io(hb); /* Free pages */
384 if (ret)
385 return ret;
386 src = (void *)__get_free_page(__GFP_RECLAIM |
387 __GFP_NOWARN |
388 __GFP_NORETRY);
389 if (src) {
390 copy_page(src, buf);
391 } else {
392 WARN_ON_ONCE(1);
393 hb = NULL; /* Go synchronous */
394 src = buf;
395 }
396 }
397 } else {
398 src = buf;
399 }
400 return hib_submit_io(REQ_OP_WRITE, REQ_SYNC, offset, src, hb);
401}
402
403static void release_swap_writer(struct swap_map_handle *handle)
404{
405 if (handle->cur)
406 free_page((unsigned long)handle->cur);
407 handle->cur = NULL;
408}
409
410static int get_swap_writer(struct swap_map_handle *handle)
411{
412 int ret;
413
414 ret = swsusp_swap_check();
415 if (ret) {
416 if (ret != -ENOSPC)
417 printk(KERN_ERR "PM: Cannot find swap device, try "
418 "swapon -a.\n");
419 return ret;
420 }
421 handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_KERNEL);
422 if (!handle->cur) {
423 ret = -ENOMEM;
424 goto err_close;
425 }
426 handle->cur_swap = alloc_swapdev_block(root_swap);
427 if (!handle->cur_swap) {
428 ret = -ENOSPC;
429 goto err_rel;
430 }
431 handle->k = 0;
432 handle->reqd_free_pages = reqd_free_pages();
433 handle->first_sector = handle->cur_swap;
434 return 0;
435err_rel:
436 release_swap_writer(handle);
437err_close:
438 swsusp_close(FMODE_WRITE);
439 return ret;
440}
441
442static int swap_write_page(struct swap_map_handle *handle, void *buf,
443 struct hib_bio_batch *hb)
444{
445 int error = 0;
446 sector_t offset;
447
448 if (!handle->cur)
449 return -EINVAL;
450 offset = alloc_swapdev_block(root_swap);
451 error = write_page(buf, offset, hb);
452 if (error)
453 return error;
454 handle->cur->entries[handle->k++] = offset;
455 if (handle->k >= MAP_PAGE_ENTRIES) {
456 offset = alloc_swapdev_block(root_swap);
457 if (!offset)
458 return -ENOSPC;
459 handle->cur->next_swap = offset;
460 error = write_page(handle->cur, handle->cur_swap, hb);
461 if (error)
462 goto out;
463 clear_page(handle->cur);
464 handle->cur_swap = offset;
465 handle->k = 0;
466
467 if (hb && low_free_pages() <= handle->reqd_free_pages) {
468 error = hib_wait_io(hb);
469 if (error)
470 goto out;
471 /*
472 * Recalculate the number of required free pages, to
473 * make sure we never take more than half.
474 */
475 handle->reqd_free_pages = reqd_free_pages();
476 }
477 }
478 out:
479 return error;
480}
481
482static int flush_swap_writer(struct swap_map_handle *handle)
483{
484 if (handle->cur && handle->cur_swap)
485 return write_page(handle->cur, handle->cur_swap, NULL);
486 else
487 return -EINVAL;
488}
489
490static int swap_writer_finish(struct swap_map_handle *handle,
491 unsigned int flags, int error)
492{
493 if (!error) {
494 flush_swap_writer(handle);
495 printk(KERN_INFO "PM: S");
496 error = mark_swapfiles(handle, flags);
497 printk("|\n");
498 }
499
500 if (error)
501 free_all_swap_pages(root_swap);
502 release_swap_writer(handle);
503 swsusp_close(FMODE_WRITE);
504
505 return error;
506}
507
508/* We need to remember how much compressed data we need to read. */
509#define LZO_HEADER sizeof(size_t)
510
511/* Number of pages/bytes we'll compress at one time. */
512#define LZO_UNC_PAGES 32
513#define LZO_UNC_SIZE (LZO_UNC_PAGES * PAGE_SIZE)
514
515/* Number of pages/bytes we need for compressed data (worst case). */
516#define LZO_CMP_PAGES DIV_ROUND_UP(lzo1x_worst_compress(LZO_UNC_SIZE) + \
517 LZO_HEADER, PAGE_SIZE)
518#define LZO_CMP_SIZE (LZO_CMP_PAGES * PAGE_SIZE)
519
520/* Maximum number of threads for compression/decompression. */
521#define LZO_THREADS 3
522
523/* Minimum/maximum number of pages for read buffering. */
524#define LZO_MIN_RD_PAGES 1024
525#define LZO_MAX_RD_PAGES 8192
526
527
528/**
529 * save_image - save the suspend image data
530 */
531
532static int save_image(struct swap_map_handle *handle,
533 struct snapshot_handle *snapshot,
534 unsigned int nr_to_write)
535{
536 unsigned int m;
537 int ret;
538 int nr_pages;
539 int err2;
540 struct hib_bio_batch hb;
541 ktime_t start;
542 ktime_t stop;
543
544 hib_init_batch(&hb);
545
546 printk(KERN_INFO "PM: Saving image data pages (%u pages)...\n",
547 nr_to_write);
548 m = nr_to_write / 10;
549 if (!m)
550 m = 1;
551 nr_pages = 0;
552 start = ktime_get();
553 while (1) {
554 ret = snapshot_read_next(snapshot);
555 if (ret <= 0)
556 break;
557 ret = swap_write_page(handle, data_of(*snapshot), &hb);
558 if (ret)
559 break;
560 if (!(nr_pages % m))
561 printk(KERN_INFO "PM: Image saving progress: %3d%%\n",
562 nr_pages / m * 10);
563 nr_pages++;
564 }
565 err2 = hib_wait_io(&hb);
566 stop = ktime_get();
567 if (!ret)
568 ret = err2;
569 if (!ret)
570 printk(KERN_INFO "PM: Image saving done.\n");
571 swsusp_show_speed(start, stop, nr_to_write, "Wrote");
572 return ret;
573}
574
575/**
576 * Structure used for CRC32.
577 */
578struct crc_data {
579 struct task_struct *thr; /* thread */
580 atomic_t ready; /* ready to start flag */
581 atomic_t stop; /* ready to stop flag */
582 unsigned run_threads; /* nr current threads */
583 wait_queue_head_t go; /* start crc update */
584 wait_queue_head_t done; /* crc update done */
585 u32 *crc32; /* points to handle's crc32 */
586 size_t *unc_len[LZO_THREADS]; /* uncompressed lengths */
587 unsigned char *unc[LZO_THREADS]; /* uncompressed data */
588};
589
590/**
591 * CRC32 update function that runs in its own thread.
592 */
593static int crc32_threadfn(void *data)
594{
595 struct crc_data *d = data;
596 unsigned i;
597
598 while (1) {
599 wait_event(d->go, atomic_read(&d->ready) ||
600 kthread_should_stop());
601 if (kthread_should_stop()) {
602 d->thr = NULL;
603 atomic_set(&d->stop, 1);
604 wake_up(&d->done);
605 break;
606 }
607 atomic_set(&d->ready, 0);
608
609 for (i = 0; i < d->run_threads; i++)
610 *d->crc32 = crc32_le(*d->crc32,
611 d->unc[i], *d->unc_len[i]);
612 atomic_set(&d->stop, 1);
613 wake_up(&d->done);
614 }
615 return 0;
616}
617/**
618 * Structure used for LZO data compression.
619 */
620struct cmp_data {
621 struct task_struct *thr; /* thread */
622 atomic_t ready; /* ready to start flag */
623 atomic_t stop; /* ready to stop flag */
624 int ret; /* return code */
625 wait_queue_head_t go; /* start compression */
626 wait_queue_head_t done; /* compression done */
627 size_t unc_len; /* uncompressed length */
628 size_t cmp_len; /* compressed length */
629 unsigned char unc[LZO_UNC_SIZE]; /* uncompressed buffer */
630 unsigned char cmp[LZO_CMP_SIZE]; /* compressed buffer */
631 unsigned char wrk[LZO1X_1_MEM_COMPRESS]; /* compression workspace */
632};
633
634/**
635 * Compression function that runs in its own thread.
636 */
637static int lzo_compress_threadfn(void *data)
638{
639 struct cmp_data *d = data;
640
641 while (1) {
642 wait_event(d->go, atomic_read(&d->ready) ||
643 kthread_should_stop());
644 if (kthread_should_stop()) {
645 d->thr = NULL;
646 d->ret = -1;
647 atomic_set(&d->stop, 1);
648 wake_up(&d->done);
649 break;
650 }
651 atomic_set(&d->ready, 0);
652
653 d->ret = lzo1x_1_compress(d->unc, d->unc_len,
654 d->cmp + LZO_HEADER, &d->cmp_len,
655 d->wrk);
656 atomic_set(&d->stop, 1);
657 wake_up(&d->done);
658 }
659 return 0;
660}
661
662/**
663 * save_image_lzo - Save the suspend image data compressed with LZO.
664 * @handle: Swap map handle to use for saving the image.
665 * @snapshot: Image to read data from.
666 * @nr_to_write: Number of pages to save.
667 */
668static int save_image_lzo(struct swap_map_handle *handle,
669 struct snapshot_handle *snapshot,
670 unsigned int nr_to_write)
671{
672 unsigned int m;
673 int ret = 0;
674 int nr_pages;
675 int err2;
676 struct hib_bio_batch hb;
677 ktime_t start;
678 ktime_t stop;
679 size_t off;
680 unsigned thr, run_threads, nr_threads;
681 unsigned char *page = NULL;
682 struct cmp_data *data = NULL;
683 struct crc_data *crc = NULL;
684
685 hib_init_batch(&hb);
686
687 /*
688 * We'll limit the number of threads for compression to limit memory
689 * footprint.
690 */
691 nr_threads = num_online_cpus() - 1;
692 nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
693
694 page = (void *)__get_free_page(__GFP_RECLAIM | __GFP_HIGH);
695 if (!page) {
696 printk(KERN_ERR "PM: Failed to allocate LZO page\n");
697 ret = -ENOMEM;
698 goto out_clean;
699 }
700
701 data = vmalloc(sizeof(*data) * nr_threads);
702 if (!data) {
703 printk(KERN_ERR "PM: Failed to allocate LZO data\n");
704 ret = -ENOMEM;
705 goto out_clean;
706 }
707 for (thr = 0; thr < nr_threads; thr++)
708 memset(&data[thr], 0, offsetof(struct cmp_data, go));
709
710 crc = kmalloc(sizeof(*crc), GFP_KERNEL);
711 if (!crc) {
712 printk(KERN_ERR "PM: Failed to allocate crc\n");
713 ret = -ENOMEM;
714 goto out_clean;
715 }
716 memset(crc, 0, offsetof(struct crc_data, go));
717
718 /*
719 * Start the compression threads.
720 */
721 for (thr = 0; thr < nr_threads; thr++) {
722 init_waitqueue_head(&data[thr].go);
723 init_waitqueue_head(&data[thr].done);
724
725 data[thr].thr = kthread_run(lzo_compress_threadfn,
726 &data[thr],
727 "image_compress/%u", thr);
728 if (IS_ERR(data[thr].thr)) {
729 data[thr].thr = NULL;
730 printk(KERN_ERR
731 "PM: Cannot start compression threads\n");
732 ret = -ENOMEM;
733 goto out_clean;
734 }
735 }
736
737 /*
738 * Start the CRC32 thread.
739 */
740 init_waitqueue_head(&crc->go);
741 init_waitqueue_head(&crc->done);
742
743 handle->crc32 = 0;
744 crc->crc32 = &handle->crc32;
745 for (thr = 0; thr < nr_threads; thr++) {
746 crc->unc[thr] = data[thr].unc;
747 crc->unc_len[thr] = &data[thr].unc_len;
748 }
749
750 crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
751 if (IS_ERR(crc->thr)) {
752 crc->thr = NULL;
753 printk(KERN_ERR "PM: Cannot start CRC32 thread\n");
754 ret = -ENOMEM;
755 goto out_clean;
756 }
757
758 /*
759 * Adjust the number of required free pages after all allocations have
760 * been done. We don't want to run out of pages when writing.
761 */
762 handle->reqd_free_pages = reqd_free_pages();
763
764 printk(KERN_INFO
765 "PM: Using %u thread(s) for compression.\n"
766 "PM: Compressing and saving image data (%u pages)...\n",
767 nr_threads, nr_to_write);
768 m = nr_to_write / 10;
769 if (!m)
770 m = 1;
771 nr_pages = 0;
772 start = ktime_get();
773 for (;;) {
774 for (thr = 0; thr < nr_threads; thr++) {
775 for (off = 0; off < LZO_UNC_SIZE; off += PAGE_SIZE) {
776 ret = snapshot_read_next(snapshot);
777 if (ret < 0)
778 goto out_finish;
779
780 if (!ret)
781 break;
782
783 memcpy(data[thr].unc + off,
784 data_of(*snapshot), PAGE_SIZE);
785
786 if (!(nr_pages % m))
787 printk(KERN_INFO
788 "PM: Image saving progress: "
789 "%3d%%\n",
790 nr_pages / m * 10);
791 nr_pages++;
792 }
793 if (!off)
794 break;
795
796 data[thr].unc_len = off;
797
798 atomic_set(&data[thr].ready, 1);
799 wake_up(&data[thr].go);
800 }
801
802 if (!thr)
803 break;
804
805 crc->run_threads = thr;
806 atomic_set(&crc->ready, 1);
807 wake_up(&crc->go);
808
809 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
810 wait_event(data[thr].done,
811 atomic_read(&data[thr].stop));
812 atomic_set(&data[thr].stop, 0);
813
814 ret = data[thr].ret;
815
816 if (ret < 0) {
817 printk(KERN_ERR "PM: LZO compression failed\n");
818 goto out_finish;
819 }
820
821 if (unlikely(!data[thr].cmp_len ||
822 data[thr].cmp_len >
823 lzo1x_worst_compress(data[thr].unc_len))) {
824 printk(KERN_ERR
825 "PM: Invalid LZO compressed length\n");
826 ret = -1;
827 goto out_finish;
828 }
829
830 *(size_t *)data[thr].cmp = data[thr].cmp_len;
831
832 /*
833 * Given we are writing one page at a time to disk, we
834 * copy that much from the buffer, although the last
835 * bit will likely be smaller than full page. This is
836 * OK - we saved the length of the compressed data, so
837 * any garbage at the end will be discarded when we
838 * read it.
839 */
840 for (off = 0;
841 off < LZO_HEADER + data[thr].cmp_len;
842 off += PAGE_SIZE) {
843 memcpy(page, data[thr].cmp + off, PAGE_SIZE);
844
845 ret = swap_write_page(handle, page, &hb);
846 if (ret)
847 goto out_finish;
848 }
849 }
850
851 wait_event(crc->done, atomic_read(&crc->stop));
852 atomic_set(&crc->stop, 0);
853 }
854
855out_finish:
856 err2 = hib_wait_io(&hb);
857 stop = ktime_get();
858 if (!ret)
859 ret = err2;
860 if (!ret)
861 printk(KERN_INFO "PM: Image saving done.\n");
862 swsusp_show_speed(start, stop, nr_to_write, "Wrote");
863out_clean:
864 if (crc) {
865 if (crc->thr)
866 kthread_stop(crc->thr);
867 kfree(crc);
868 }
869 if (data) {
870 for (thr = 0; thr < nr_threads; thr++)
871 if (data[thr].thr)
872 kthread_stop(data[thr].thr);
873 vfree(data);
874 }
875 if (page) free_page((unsigned long)page);
876
877 return ret;
878}
879
880/**
881 * enough_swap - Make sure we have enough swap to save the image.
882 *
883 * Returns TRUE or FALSE after checking the total amount of swap
884 * space avaiable from the resume partition.
885 */
886
887static int enough_swap(unsigned int nr_pages, unsigned int flags)
888{
889 unsigned int free_swap = count_swap_pages(root_swap, 1);
890 unsigned int required;
891
892 pr_debug("PM: Free swap pages: %u\n", free_swap);
893
894 required = PAGES_FOR_IO + nr_pages;
895 return free_swap > required;
896}
897
898/**
899 * swsusp_write - Write entire image and metadata.
900 * @flags: flags to pass to the "boot" kernel in the image header
901 *
902 * It is important _NOT_ to umount filesystems at this point. We want
903 * them synced (in case something goes wrong) but we DO not want to mark
904 * filesystem clean: it is not. (And it does not matter, if we resume
905 * correctly, we'll mark system clean, anyway.)
906 */
907
908int swsusp_write(unsigned int flags)
909{
910 struct swap_map_handle handle;
911 struct snapshot_handle snapshot;
912 struct swsusp_info *header;
913 unsigned long pages;
914 int error;
915
916 pages = snapshot_get_image_size();
917 error = get_swap_writer(&handle);
918 if (error) {
919 printk(KERN_ERR "PM: Cannot get swap writer\n");
920 return error;
921 }
922 if (flags & SF_NOCOMPRESS_MODE) {
923 if (!enough_swap(pages, flags)) {
924 printk(KERN_ERR "PM: Not enough free swap\n");
925 error = -ENOSPC;
926 goto out_finish;
927 }
928 }
929 memset(&snapshot, 0, sizeof(struct snapshot_handle));
930 error = snapshot_read_next(&snapshot);
931 if (error < PAGE_SIZE) {
932 if (error >= 0)
933 error = -EFAULT;
934
935 goto out_finish;
936 }
937 header = (struct swsusp_info *)data_of(snapshot);
938 error = swap_write_page(&handle, header, NULL);
939 if (!error) {
940 error = (flags & SF_NOCOMPRESS_MODE) ?
941 save_image(&handle, &snapshot, pages - 1) :
942 save_image_lzo(&handle, &snapshot, pages - 1);
943 }
944out_finish:
945 error = swap_writer_finish(&handle, flags, error);
946 return error;
947}
948
949/**
950 * The following functions allow us to read data using a swap map
951 * in a file-alike way
952 */
953
954static void release_swap_reader(struct swap_map_handle *handle)
955{
956 struct swap_map_page_list *tmp;
957
958 while (handle->maps) {
959 if (handle->maps->map)
960 free_page((unsigned long)handle->maps->map);
961 tmp = handle->maps;
962 handle->maps = handle->maps->next;
963 kfree(tmp);
964 }
965 handle->cur = NULL;
966}
967
968static int get_swap_reader(struct swap_map_handle *handle,
969 unsigned int *flags_p)
970{
971 int error;
972 struct swap_map_page_list *tmp, *last;
973 sector_t offset;
974
975 *flags_p = swsusp_header->flags;
976
977 if (!swsusp_header->image) /* how can this happen? */
978 return -EINVAL;
979
980 handle->cur = NULL;
981 last = handle->maps = NULL;
982 offset = swsusp_header->image;
983 while (offset) {
984 tmp = kmalloc(sizeof(*handle->maps), GFP_KERNEL);
985 if (!tmp) {
986 release_swap_reader(handle);
987 return -ENOMEM;
988 }
989 memset(tmp, 0, sizeof(*tmp));
990 if (!handle->maps)
991 handle->maps = tmp;
992 if (last)
993 last->next = tmp;
994 last = tmp;
995
996 tmp->map = (struct swap_map_page *)
997 __get_free_page(__GFP_RECLAIM | __GFP_HIGH);
998 if (!tmp->map) {
999 release_swap_reader(handle);
1000 return -ENOMEM;
1001 }
1002
1003 error = hib_submit_io(REQ_OP_READ, 0, offset, tmp->map, NULL);
1004 if (error) {
1005 release_swap_reader(handle);
1006 return error;
1007 }
1008 offset = tmp->map->next_swap;
1009 }
1010 handle->k = 0;
1011 handle->cur = handle->maps->map;
1012 return 0;
1013}
1014
1015static int swap_read_page(struct swap_map_handle *handle, void *buf,
1016 struct hib_bio_batch *hb)
1017{
1018 sector_t offset;
1019 int error;
1020 struct swap_map_page_list *tmp;
1021
1022 if (!handle->cur)
1023 return -EINVAL;
1024 offset = handle->cur->entries[handle->k];
1025 if (!offset)
1026 return -EFAULT;
1027 error = hib_submit_io(REQ_OP_READ, 0, offset, buf, hb);
1028 if (error)
1029 return error;
1030 if (++handle->k >= MAP_PAGE_ENTRIES) {
1031 handle->k = 0;
1032 free_page((unsigned long)handle->maps->map);
1033 tmp = handle->maps;
1034 handle->maps = handle->maps->next;
1035 kfree(tmp);
1036 if (!handle->maps)
1037 release_swap_reader(handle);
1038 else
1039 handle->cur = handle->maps->map;
1040 }
1041 return error;
1042}
1043
1044static int swap_reader_finish(struct swap_map_handle *handle)
1045{
1046 release_swap_reader(handle);
1047
1048 return 0;
1049}
1050
1051/**
1052 * load_image - load the image using the swap map handle
1053 * @handle and the snapshot handle @snapshot
1054 * (assume there are @nr_pages pages to load)
1055 */
1056
1057static int load_image(struct swap_map_handle *handle,
1058 struct snapshot_handle *snapshot,
1059 unsigned int nr_to_read)
1060{
1061 unsigned int m;
1062 int ret = 0;
1063 ktime_t start;
1064 ktime_t stop;
1065 struct hib_bio_batch hb;
1066 int err2;
1067 unsigned nr_pages;
1068
1069 hib_init_batch(&hb);
1070
1071 clean_pages_on_read = true;
1072 printk(KERN_INFO "PM: Loading image data pages (%u pages)...\n",
1073 nr_to_read);
1074 m = nr_to_read / 10;
1075 if (!m)
1076 m = 1;
1077 nr_pages = 0;
1078 start = ktime_get();
1079 for ( ; ; ) {
1080 ret = snapshot_write_next(snapshot);
1081 if (ret <= 0)
1082 break;
1083 ret = swap_read_page(handle, data_of(*snapshot), &hb);
1084 if (ret)
1085 break;
1086 if (snapshot->sync_read)
1087 ret = hib_wait_io(&hb);
1088 if (ret)
1089 break;
1090 if (!(nr_pages % m))
1091 printk(KERN_INFO "PM: Image loading progress: %3d%%\n",
1092 nr_pages / m * 10);
1093 nr_pages++;
1094 }
1095 err2 = hib_wait_io(&hb);
1096 stop = ktime_get();
1097 if (!ret)
1098 ret = err2;
1099 if (!ret) {
1100 printk(KERN_INFO "PM: Image loading done.\n");
1101 snapshot_write_finalize(snapshot);
1102 if (!snapshot_image_loaded(snapshot))
1103 ret = -ENODATA;
1104 }
1105 swsusp_show_speed(start, stop, nr_to_read, "Read");
1106 return ret;
1107}
1108
1109/**
1110 * Structure used for LZO data decompression.
1111 */
1112struct dec_data {
1113 struct task_struct *thr; /* thread */
1114 atomic_t ready; /* ready to start flag */
1115 atomic_t stop; /* ready to stop flag */
1116 int ret; /* return code */
1117 wait_queue_head_t go; /* start decompression */
1118 wait_queue_head_t done; /* decompression done */
1119 size_t unc_len; /* uncompressed length */
1120 size_t cmp_len; /* compressed length */
1121 unsigned char unc[LZO_UNC_SIZE]; /* uncompressed buffer */
1122 unsigned char cmp[LZO_CMP_SIZE]; /* compressed buffer */
1123};
1124
1125/**
1126 * Deompression function that runs in its own thread.
1127 */
1128static int lzo_decompress_threadfn(void *data)
1129{
1130 struct dec_data *d = data;
1131
1132 while (1) {
1133 wait_event(d->go, atomic_read(&d->ready) ||
1134 kthread_should_stop());
1135 if (kthread_should_stop()) {
1136 d->thr = NULL;
1137 d->ret = -1;
1138 atomic_set(&d->stop, 1);
1139 wake_up(&d->done);
1140 break;
1141 }
1142 atomic_set(&d->ready, 0);
1143
1144 d->unc_len = LZO_UNC_SIZE;
1145 d->ret = lzo1x_decompress_safe(d->cmp + LZO_HEADER, d->cmp_len,
1146 d->unc, &d->unc_len);
1147 if (clean_pages_on_decompress)
1148 flush_icache_range((unsigned long)d->unc,
1149 (unsigned long)d->unc + d->unc_len);
1150
1151 atomic_set(&d->stop, 1);
1152 wake_up(&d->done);
1153 }
1154 return 0;
1155}
1156
1157/**
1158 * load_image_lzo - Load compressed image data and decompress them with LZO.
1159 * @handle: Swap map handle to use for loading data.
1160 * @snapshot: Image to copy uncompressed data into.
1161 * @nr_to_read: Number of pages to load.
1162 */
1163static int load_image_lzo(struct swap_map_handle *handle,
1164 struct snapshot_handle *snapshot,
1165 unsigned int nr_to_read)
1166{
1167 unsigned int m;
1168 int ret = 0;
1169 int eof = 0;
1170 struct hib_bio_batch hb;
1171 ktime_t start;
1172 ktime_t stop;
1173 unsigned nr_pages;
1174 size_t off;
1175 unsigned i, thr, run_threads, nr_threads;
1176 unsigned ring = 0, pg = 0, ring_size = 0,
1177 have = 0, want, need, asked = 0;
1178 unsigned long read_pages = 0;
1179 unsigned char **page = NULL;
1180 struct dec_data *data = NULL;
1181 struct crc_data *crc = NULL;
1182
1183 hib_init_batch(&hb);
1184
1185 /*
1186 * We'll limit the number of threads for decompression to limit memory
1187 * footprint.
1188 */
1189 nr_threads = num_online_cpus() - 1;
1190 nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
1191
1192 page = vmalloc(sizeof(*page) * LZO_MAX_RD_PAGES);
1193 if (!page) {
1194 printk(KERN_ERR "PM: Failed to allocate LZO page\n");
1195 ret = -ENOMEM;
1196 goto out_clean;
1197 }
1198
1199 data = vmalloc(sizeof(*data) * nr_threads);
1200 if (!data) {
1201 printk(KERN_ERR "PM: Failed to allocate LZO data\n");
1202 ret = -ENOMEM;
1203 goto out_clean;
1204 }
1205 for (thr = 0; thr < nr_threads; thr++)
1206 memset(&data[thr], 0, offsetof(struct dec_data, go));
1207
1208 crc = kmalloc(sizeof(*crc), GFP_KERNEL);
1209 if (!crc) {
1210 printk(KERN_ERR "PM: Failed to allocate crc\n");
1211 ret = -ENOMEM;
1212 goto out_clean;
1213 }
1214 memset(crc, 0, offsetof(struct crc_data, go));
1215
1216 clean_pages_on_decompress = true;
1217
1218 /*
1219 * Start the decompression threads.
1220 */
1221 for (thr = 0; thr < nr_threads; thr++) {
1222 init_waitqueue_head(&data[thr].go);
1223 init_waitqueue_head(&data[thr].done);
1224
1225 data[thr].thr = kthread_run(lzo_decompress_threadfn,
1226 &data[thr],
1227 "image_decompress/%u", thr);
1228 if (IS_ERR(data[thr].thr)) {
1229 data[thr].thr = NULL;
1230 printk(KERN_ERR
1231 "PM: Cannot start decompression threads\n");
1232 ret = -ENOMEM;
1233 goto out_clean;
1234 }
1235 }
1236
1237 /*
1238 * Start the CRC32 thread.
1239 */
1240 init_waitqueue_head(&crc->go);
1241 init_waitqueue_head(&crc->done);
1242
1243 handle->crc32 = 0;
1244 crc->crc32 = &handle->crc32;
1245 for (thr = 0; thr < nr_threads; thr++) {
1246 crc->unc[thr] = data[thr].unc;
1247 crc->unc_len[thr] = &data[thr].unc_len;
1248 }
1249
1250 crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
1251 if (IS_ERR(crc->thr)) {
1252 crc->thr = NULL;
1253 printk(KERN_ERR "PM: Cannot start CRC32 thread\n");
1254 ret = -ENOMEM;
1255 goto out_clean;
1256 }
1257
1258 /*
1259 * Set the number of pages for read buffering.
1260 * This is complete guesswork, because we'll only know the real
1261 * picture once prepare_image() is called, which is much later on
1262 * during the image load phase. We'll assume the worst case and
1263 * say that none of the image pages are from high memory.
1264 */
1265 if (low_free_pages() > snapshot_get_image_size())
1266 read_pages = (low_free_pages() - snapshot_get_image_size()) / 2;
1267 read_pages = clamp_val(read_pages, LZO_MIN_RD_PAGES, LZO_MAX_RD_PAGES);
1268
1269 for (i = 0; i < read_pages; i++) {
1270 page[i] = (void *)__get_free_page(i < LZO_CMP_PAGES ?
1271 __GFP_RECLAIM | __GFP_HIGH :
1272 __GFP_RECLAIM | __GFP_NOWARN |
1273 __GFP_NORETRY);
1274
1275 if (!page[i]) {
1276 if (i < LZO_CMP_PAGES) {
1277 ring_size = i;
1278 printk(KERN_ERR
1279 "PM: Failed to allocate LZO pages\n");
1280 ret = -ENOMEM;
1281 goto out_clean;
1282 } else {
1283 break;
1284 }
1285 }
1286 }
1287 want = ring_size = i;
1288
1289 printk(KERN_INFO
1290 "PM: Using %u thread(s) for decompression.\n"
1291 "PM: Loading and decompressing image data (%u pages)...\n",
1292 nr_threads, nr_to_read);
1293 m = nr_to_read / 10;
1294 if (!m)
1295 m = 1;
1296 nr_pages = 0;
1297 start = ktime_get();
1298
1299 ret = snapshot_write_next(snapshot);
1300 if (ret <= 0)
1301 goto out_finish;
1302
1303 for(;;) {
1304 for (i = 0; !eof && i < want; i++) {
1305 ret = swap_read_page(handle, page[ring], &hb);
1306 if (ret) {
1307 /*
1308 * On real read error, finish. On end of data,
1309 * set EOF flag and just exit the read loop.
1310 */
1311 if (handle->cur &&
1312 handle->cur->entries[handle->k]) {
1313 goto out_finish;
1314 } else {
1315 eof = 1;
1316 break;
1317 }
1318 }
1319 if (++ring >= ring_size)
1320 ring = 0;
1321 }
1322 asked += i;
1323 want -= i;
1324
1325 /*
1326 * We are out of data, wait for some more.
1327 */
1328 if (!have) {
1329 if (!asked)
1330 break;
1331
1332 ret = hib_wait_io(&hb);
1333 if (ret)
1334 goto out_finish;
1335 have += asked;
1336 asked = 0;
1337 if (eof)
1338 eof = 2;
1339 }
1340
1341 if (crc->run_threads) {
1342 wait_event(crc->done, atomic_read(&crc->stop));
1343 atomic_set(&crc->stop, 0);
1344 crc->run_threads = 0;
1345 }
1346
1347 for (thr = 0; have && thr < nr_threads; thr++) {
1348 data[thr].cmp_len = *(size_t *)page[pg];
1349 if (unlikely(!data[thr].cmp_len ||
1350 data[thr].cmp_len >
1351 lzo1x_worst_compress(LZO_UNC_SIZE))) {
1352 printk(KERN_ERR
1353 "PM: Invalid LZO compressed length\n");
1354 ret = -1;
1355 goto out_finish;
1356 }
1357
1358 need = DIV_ROUND_UP(data[thr].cmp_len + LZO_HEADER,
1359 PAGE_SIZE);
1360 if (need > have) {
1361 if (eof > 1) {
1362 ret = -1;
1363 goto out_finish;
1364 }
1365 break;
1366 }
1367
1368 for (off = 0;
1369 off < LZO_HEADER + data[thr].cmp_len;
1370 off += PAGE_SIZE) {
1371 memcpy(data[thr].cmp + off,
1372 page[pg], PAGE_SIZE);
1373 have--;
1374 want++;
1375 if (++pg >= ring_size)
1376 pg = 0;
1377 }
1378
1379 atomic_set(&data[thr].ready, 1);
1380 wake_up(&data[thr].go);
1381 }
1382
1383 /*
1384 * Wait for more data while we are decompressing.
1385 */
1386 if (have < LZO_CMP_PAGES && asked) {
1387 ret = hib_wait_io(&hb);
1388 if (ret)
1389 goto out_finish;
1390 have += asked;
1391 asked = 0;
1392 if (eof)
1393 eof = 2;
1394 }
1395
1396 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
1397 wait_event(data[thr].done,
1398 atomic_read(&data[thr].stop));
1399 atomic_set(&data[thr].stop, 0);
1400
1401 ret = data[thr].ret;
1402
1403 if (ret < 0) {
1404 printk(KERN_ERR
1405 "PM: LZO decompression failed\n");
1406 goto out_finish;
1407 }
1408
1409 if (unlikely(!data[thr].unc_len ||
1410 data[thr].unc_len > LZO_UNC_SIZE ||
1411 data[thr].unc_len & (PAGE_SIZE - 1))) {
1412 printk(KERN_ERR
1413 "PM: Invalid LZO uncompressed length\n");
1414 ret = -1;
1415 goto out_finish;
1416 }
1417
1418 for (off = 0;
1419 off < data[thr].unc_len; off += PAGE_SIZE) {
1420 memcpy(data_of(*snapshot),
1421 data[thr].unc + off, PAGE_SIZE);
1422
1423 if (!(nr_pages % m))
1424 printk(KERN_INFO
1425 "PM: Image loading progress: "
1426 "%3d%%\n",
1427 nr_pages / m * 10);
1428 nr_pages++;
1429
1430 ret = snapshot_write_next(snapshot);
1431 if (ret <= 0) {
1432 crc->run_threads = thr + 1;
1433 atomic_set(&crc->ready, 1);
1434 wake_up(&crc->go);
1435 goto out_finish;
1436 }
1437 }
1438 }
1439
1440 crc->run_threads = thr;
1441 atomic_set(&crc->ready, 1);
1442 wake_up(&crc->go);
1443 }
1444
1445out_finish:
1446 if (crc->run_threads) {
1447 wait_event(crc->done, atomic_read(&crc->stop));
1448 atomic_set(&crc->stop, 0);
1449 }
1450 stop = ktime_get();
1451 if (!ret) {
1452 printk(KERN_INFO "PM: Image loading done.\n");
1453 snapshot_write_finalize(snapshot);
1454 if (!snapshot_image_loaded(snapshot))
1455 ret = -ENODATA;
1456 if (!ret) {
1457 if (swsusp_header->flags & SF_CRC32_MODE) {
1458 if(handle->crc32 != swsusp_header->crc32) {
1459 printk(KERN_ERR
1460 "PM: Invalid image CRC32!\n");
1461 ret = -ENODATA;
1462 }
1463 }
1464 }
1465 }
1466 swsusp_show_speed(start, stop, nr_to_read, "Read");
1467out_clean:
1468 for (i = 0; i < ring_size; i++)
1469 free_page((unsigned long)page[i]);
1470 if (crc) {
1471 if (crc->thr)
1472 kthread_stop(crc->thr);
1473 kfree(crc);
1474 }
1475 if (data) {
1476 for (thr = 0; thr < nr_threads; thr++)
1477 if (data[thr].thr)
1478 kthread_stop(data[thr].thr);
1479 vfree(data);
1480 }
1481 vfree(page);
1482
1483 return ret;
1484}
1485
1486/**
1487 * swsusp_read - read the hibernation image.
1488 * @flags_p: flags passed by the "frozen" kernel in the image header should
1489 * be written into this memory location
1490 */
1491
1492int swsusp_read(unsigned int *flags_p)
1493{
1494 int error;
1495 struct swap_map_handle handle;
1496 struct snapshot_handle snapshot;
1497 struct swsusp_info *header;
1498
1499 memset(&snapshot, 0, sizeof(struct snapshot_handle));
1500 error = snapshot_write_next(&snapshot);
1501 if (error < PAGE_SIZE)
1502 return error < 0 ? error : -EFAULT;
1503 header = (struct swsusp_info *)data_of(snapshot);
1504 error = get_swap_reader(&handle, flags_p);
1505 if (error)
1506 goto end;
1507 if (!error)
1508 error = swap_read_page(&handle, header, NULL);
1509 if (!error) {
1510 error = (*flags_p & SF_NOCOMPRESS_MODE) ?
1511 load_image(&handle, &snapshot, header->pages - 1) :
1512 load_image_lzo(&handle, &snapshot, header->pages - 1);
1513 }
1514 swap_reader_finish(&handle);
1515end:
1516 if (!error)
1517 pr_debug("PM: Image successfully loaded\n");
1518 else
1519 pr_debug("PM: Error %d resuming\n", error);
1520 return error;
1521}
1522
1523/**
1524 * swsusp_check - Check for swsusp signature in the resume device
1525 */
1526
1527int swsusp_check(void)
1528{
1529 int error;
1530
1531 hib_resume_bdev = blkdev_get_by_dev(swsusp_resume_device,
1532 FMODE_READ, NULL);
1533 if (!IS_ERR(hib_resume_bdev)) {
1534 set_blocksize(hib_resume_bdev, PAGE_SIZE);
1535 clear_page(swsusp_header);
1536 error = hib_submit_io(REQ_OP_READ, 0,
1537 swsusp_resume_block,
1538 swsusp_header, NULL);
1539 if (error)
1540 goto put;
1541
1542 if (!memcmp(HIBERNATE_SIG, swsusp_header->sig, 10)) {
1543 memcpy(swsusp_header->sig, swsusp_header->orig_sig, 10);
1544 /* Reset swap signature now */
1545 error = hib_submit_io(REQ_OP_WRITE, REQ_SYNC,
1546 swsusp_resume_block,
1547 swsusp_header, NULL);
1548 } else {
1549 error = -EINVAL;
1550 }
1551
1552put:
1553 if (error)
1554 blkdev_put(hib_resume_bdev, FMODE_READ);
1555 else
1556 pr_debug("PM: Image signature found, resuming\n");
1557 } else {
1558 error = PTR_ERR(hib_resume_bdev);
1559 }
1560
1561 if (error)
1562 pr_debug("PM: Image not found (code %d)\n", error);
1563
1564 return error;
1565}
1566
1567/**
1568 * swsusp_close - close swap device.
1569 */
1570
1571void swsusp_close(fmode_t mode)
1572{
1573 if (IS_ERR(hib_resume_bdev)) {
1574 pr_debug("PM: Image device not initialised\n");
1575 return;
1576 }
1577
1578 blkdev_put(hib_resume_bdev, mode);
1579}
1580
1581/**
1582 * swsusp_unmark - Unmark swsusp signature in the resume device
1583 */
1584
1585#ifdef CONFIG_SUSPEND
1586int swsusp_unmark(void)
1587{
1588 int error;
1589
1590 hib_submit_io(REQ_OP_READ, 0, swsusp_resume_block,
1591 swsusp_header, NULL);
1592 if (!memcmp(HIBERNATE_SIG,swsusp_header->sig, 10)) {
1593 memcpy(swsusp_header->sig,swsusp_header->orig_sig, 10);
1594 error = hib_submit_io(REQ_OP_WRITE, REQ_SYNC,
1595 swsusp_resume_block,
1596 swsusp_header, NULL);
1597 } else {
1598 printk(KERN_ERR "PM: Cannot find swsusp signature!\n");
1599 error = -ENODEV;
1600 }
1601
1602 /*
1603 * We just returned from suspend, we don't need the image any more.
1604 */
1605 free_all_swap_pages(root_swap);
1606
1607 return error;
1608}
1609#endif
1610
1611static int swsusp_header_init(void)
1612{
1613 swsusp_header = (struct swsusp_header*) __get_free_page(GFP_KERNEL);
1614 if (!swsusp_header)
1615 panic("Could not allocate memory for swsusp_header\n");
1616 return 0;
1617}
1618
1619core_initcall(swsusp_header_init);