Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/power/swap.c
4 *
5 * This file provides functions for reading the suspend image from
6 * and writing it to a swap partition.
7 *
8 * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@ucw.cz>
9 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
10 * Copyright (C) 2010-2012 Bojan Smojver <bojan@rexursive.com>
11 */
12
13#define pr_fmt(fmt) "PM: " fmt
14
15#include <linux/module.h>
16#include <linux/file.h>
17#include <linux/delay.h>
18#include <linux/bitops.h>
19#include <linux/device.h>
20#include <linux/bio.h>
21#include <linux/blkdev.h>
22#include <linux/swap.h>
23#include <linux/swapops.h>
24#include <linux/pm.h>
25#include <linux/slab.h>
26#include <linux/lzo.h>
27#include <linux/vmalloc.h>
28#include <linux/cpumask.h>
29#include <linux/atomic.h>
30#include <linux/kthread.h>
31#include <linux/crc32.h>
32#include <linux/ktime.h>
33
34#include "power.h"
35
36#define HIBERNATE_SIG "S1SUSPEND"
37
38u32 swsusp_hardware_signature;
39
40/*
41 * When reading an {un,}compressed image, we may restore pages in place,
42 * in which case some architectures need these pages cleaning before they
43 * can be executed. We don't know which pages these may be, so clean the lot.
44 */
45static bool clean_pages_on_read;
46static bool clean_pages_on_decompress;
47
48/*
49 * The swap map is a data structure used for keeping track of each page
50 * written to a swap partition. It consists of many swap_map_page
51 * structures that contain each an array of MAP_PAGE_ENTRIES swap entries.
52 * These structures are stored on the swap and linked together with the
53 * help of the .next_swap member.
54 *
55 * The swap map is created during suspend. The swap map pages are
56 * allocated and populated one at a time, so we only need one memory
57 * page to set up the entire structure.
58 *
59 * During resume we pick up all swap_map_page structures into a list.
60 */
61
62#define MAP_PAGE_ENTRIES (PAGE_SIZE / sizeof(sector_t) - 1)
63
64/*
65 * Number of free pages that are not high.
66 */
67static inline unsigned long low_free_pages(void)
68{
69 return nr_free_pages() - nr_free_highpages();
70}
71
72/*
73 * Number of pages required to be kept free while writing the image. Always
74 * half of all available low pages before the writing starts.
75 */
76static inline unsigned long reqd_free_pages(void)
77{
78 return low_free_pages() / 2;
79}
80
81struct swap_map_page {
82 sector_t entries[MAP_PAGE_ENTRIES];
83 sector_t next_swap;
84};
85
86struct swap_map_page_list {
87 struct swap_map_page *map;
88 struct swap_map_page_list *next;
89};
90
91/*
92 * The swap_map_handle structure is used for handling swap in
93 * a file-alike way
94 */
95
96struct swap_map_handle {
97 struct swap_map_page *cur;
98 struct swap_map_page_list *maps;
99 sector_t cur_swap;
100 sector_t first_sector;
101 unsigned int k;
102 unsigned long reqd_free_pages;
103 u32 crc32;
104};
105
106struct swsusp_header {
107 char reserved[PAGE_SIZE - 20 - sizeof(sector_t) - sizeof(int) -
108 sizeof(u32) - sizeof(u32)];
109 u32 hw_sig;
110 u32 crc32;
111 sector_t image;
112 unsigned int flags; /* Flags to pass to the "boot" kernel */
113 char orig_sig[10];
114 char sig[10];
115} __packed;
116
117static struct swsusp_header *swsusp_header;
118
119/*
120 * The following functions are used for tracing the allocated
121 * swap pages, so that they can be freed in case of an error.
122 */
123
124struct swsusp_extent {
125 struct rb_node node;
126 unsigned long start;
127 unsigned long end;
128};
129
130static struct rb_root swsusp_extents = RB_ROOT;
131
132static int swsusp_extents_insert(unsigned long swap_offset)
133{
134 struct rb_node **new = &(swsusp_extents.rb_node);
135 struct rb_node *parent = NULL;
136 struct swsusp_extent *ext;
137
138 /* Figure out where to put the new node */
139 while (*new) {
140 ext = rb_entry(*new, struct swsusp_extent, node);
141 parent = *new;
142 if (swap_offset < ext->start) {
143 /* Try to merge */
144 if (swap_offset == ext->start - 1) {
145 ext->start--;
146 return 0;
147 }
148 new = &((*new)->rb_left);
149 } else if (swap_offset > ext->end) {
150 /* Try to merge */
151 if (swap_offset == ext->end + 1) {
152 ext->end++;
153 return 0;
154 }
155 new = &((*new)->rb_right);
156 } else {
157 /* It already is in the tree */
158 return -EINVAL;
159 }
160 }
161 /* Add the new node and rebalance the tree. */
162 ext = kzalloc(sizeof(struct swsusp_extent), GFP_KERNEL);
163 if (!ext)
164 return -ENOMEM;
165
166 ext->start = swap_offset;
167 ext->end = swap_offset;
168 rb_link_node(&ext->node, parent, new);
169 rb_insert_color(&ext->node, &swsusp_extents);
170 return 0;
171}
172
173/*
174 * alloc_swapdev_block - allocate a swap page and register that it has
175 * been allocated, so that it can be freed in case of an error.
176 */
177
178sector_t alloc_swapdev_block(int swap)
179{
180 unsigned long offset;
181
182 offset = swp_offset(get_swap_page_of_type(swap));
183 if (offset) {
184 if (swsusp_extents_insert(offset))
185 swap_free(swp_entry(swap, offset));
186 else
187 return swapdev_block(swap, offset);
188 }
189 return 0;
190}
191
192/*
193 * free_all_swap_pages - free swap pages allocated for saving image data.
194 * It also frees the extents used to register which swap entries had been
195 * allocated.
196 */
197
198void free_all_swap_pages(int swap)
199{
200 struct rb_node *node;
201
202 while ((node = swsusp_extents.rb_node)) {
203 struct swsusp_extent *ext;
204 unsigned long offset;
205
206 ext = rb_entry(node, struct swsusp_extent, node);
207 rb_erase(node, &swsusp_extents);
208 for (offset = ext->start; offset <= ext->end; offset++)
209 swap_free(swp_entry(swap, offset));
210
211 kfree(ext);
212 }
213}
214
215int swsusp_swap_in_use(void)
216{
217 return (swsusp_extents.rb_node != NULL);
218}
219
220/*
221 * General things
222 */
223
224static unsigned short root_swap = 0xffff;
225static struct block_device *hib_resume_bdev;
226
227struct hib_bio_batch {
228 atomic_t count;
229 wait_queue_head_t wait;
230 blk_status_t error;
231 struct blk_plug plug;
232};
233
234static void hib_init_batch(struct hib_bio_batch *hb)
235{
236 atomic_set(&hb->count, 0);
237 init_waitqueue_head(&hb->wait);
238 hb->error = BLK_STS_OK;
239 blk_start_plug(&hb->plug);
240}
241
242static void hib_finish_batch(struct hib_bio_batch *hb)
243{
244 blk_finish_plug(&hb->plug);
245}
246
247static void hib_end_io(struct bio *bio)
248{
249 struct hib_bio_batch *hb = bio->bi_private;
250 struct page *page = bio_first_page_all(bio);
251
252 if (bio->bi_status) {
253 pr_alert("Read-error on swap-device (%u:%u:%Lu)\n",
254 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
255 (unsigned long long)bio->bi_iter.bi_sector);
256 }
257
258 if (bio_data_dir(bio) == WRITE)
259 put_page(page);
260 else if (clean_pages_on_read)
261 flush_icache_range((unsigned long)page_address(page),
262 (unsigned long)page_address(page) + PAGE_SIZE);
263
264 if (bio->bi_status && !hb->error)
265 hb->error = bio->bi_status;
266 if (atomic_dec_and_test(&hb->count))
267 wake_up(&hb->wait);
268
269 bio_put(bio);
270}
271
272static int hib_submit_io(blk_opf_t opf, pgoff_t page_off, void *addr,
273 struct hib_bio_batch *hb)
274{
275 struct page *page = virt_to_page(addr);
276 struct bio *bio;
277 int error = 0;
278
279 bio = bio_alloc(hib_resume_bdev, 1, opf, GFP_NOIO | __GFP_HIGH);
280 bio->bi_iter.bi_sector = page_off * (PAGE_SIZE >> 9);
281
282 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
283 pr_err("Adding page to bio failed at %llu\n",
284 (unsigned long long)bio->bi_iter.bi_sector);
285 bio_put(bio);
286 return -EFAULT;
287 }
288
289 if (hb) {
290 bio->bi_end_io = hib_end_io;
291 bio->bi_private = hb;
292 atomic_inc(&hb->count);
293 submit_bio(bio);
294 } else {
295 error = submit_bio_wait(bio);
296 bio_put(bio);
297 }
298
299 return error;
300}
301
302static int hib_wait_io(struct hib_bio_batch *hb)
303{
304 /*
305 * We are relying on the behavior of blk_plug that a thread with
306 * a plug will flush the plug list before sleeping.
307 */
308 wait_event(hb->wait, atomic_read(&hb->count) == 0);
309 return blk_status_to_errno(hb->error);
310}
311
312/*
313 * Saving part
314 */
315static int mark_swapfiles(struct swap_map_handle *handle, unsigned int flags)
316{
317 int error;
318
319 hib_submit_io(REQ_OP_READ, swsusp_resume_block, swsusp_header, NULL);
320 if (!memcmp("SWAP-SPACE",swsusp_header->sig, 10) ||
321 !memcmp("SWAPSPACE2",swsusp_header->sig, 10)) {
322 memcpy(swsusp_header->orig_sig,swsusp_header->sig, 10);
323 memcpy(swsusp_header->sig, HIBERNATE_SIG, 10);
324 swsusp_header->image = handle->first_sector;
325 if (swsusp_hardware_signature) {
326 swsusp_header->hw_sig = swsusp_hardware_signature;
327 flags |= SF_HW_SIG;
328 }
329 swsusp_header->flags = flags;
330 if (flags & SF_CRC32_MODE)
331 swsusp_header->crc32 = handle->crc32;
332 error = hib_submit_io(REQ_OP_WRITE | REQ_SYNC,
333 swsusp_resume_block, swsusp_header, NULL);
334 } else {
335 pr_err("Swap header not found!\n");
336 error = -ENODEV;
337 }
338 return error;
339}
340
341/**
342 * swsusp_swap_check - check if the resume device is a swap device
343 * and get its index (if so)
344 *
345 * This is called before saving image
346 */
347static int swsusp_swap_check(void)
348{
349 int res;
350
351 if (swsusp_resume_device)
352 res = swap_type_of(swsusp_resume_device, swsusp_resume_block);
353 else
354 res = find_first_swap(&swsusp_resume_device);
355 if (res < 0)
356 return res;
357 root_swap = res;
358
359 hib_resume_bdev = blkdev_get_by_dev(swsusp_resume_device, FMODE_WRITE,
360 NULL);
361 if (IS_ERR(hib_resume_bdev))
362 return PTR_ERR(hib_resume_bdev);
363
364 res = set_blocksize(hib_resume_bdev, PAGE_SIZE);
365 if (res < 0)
366 blkdev_put(hib_resume_bdev, FMODE_WRITE);
367
368 return res;
369}
370
371/**
372 * write_page - Write one page to given swap location.
373 * @buf: Address we're writing.
374 * @offset: Offset of the swap page we're writing to.
375 * @hb: bio completion batch
376 */
377
378static int write_page(void *buf, sector_t offset, struct hib_bio_batch *hb)
379{
380 void *src;
381 int ret;
382
383 if (!offset)
384 return -ENOSPC;
385
386 if (hb) {
387 src = (void *)__get_free_page(GFP_NOIO | __GFP_NOWARN |
388 __GFP_NORETRY);
389 if (src) {
390 copy_page(src, buf);
391 } else {
392 ret = hib_wait_io(hb); /* Free pages */
393 if (ret)
394 return ret;
395 src = (void *)__get_free_page(GFP_NOIO |
396 __GFP_NOWARN |
397 __GFP_NORETRY);
398 if (src) {
399 copy_page(src, buf);
400 } else {
401 WARN_ON_ONCE(1);
402 hb = NULL; /* Go synchronous */
403 src = buf;
404 }
405 }
406 } else {
407 src = buf;
408 }
409 return hib_submit_io(REQ_OP_WRITE | REQ_SYNC, offset, src, hb);
410}
411
412static void release_swap_writer(struct swap_map_handle *handle)
413{
414 if (handle->cur)
415 free_page((unsigned long)handle->cur);
416 handle->cur = NULL;
417}
418
419static int get_swap_writer(struct swap_map_handle *handle)
420{
421 int ret;
422
423 ret = swsusp_swap_check();
424 if (ret) {
425 if (ret != -ENOSPC)
426 pr_err("Cannot find swap device, try swapon -a\n");
427 return ret;
428 }
429 handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_KERNEL);
430 if (!handle->cur) {
431 ret = -ENOMEM;
432 goto err_close;
433 }
434 handle->cur_swap = alloc_swapdev_block(root_swap);
435 if (!handle->cur_swap) {
436 ret = -ENOSPC;
437 goto err_rel;
438 }
439 handle->k = 0;
440 handle->reqd_free_pages = reqd_free_pages();
441 handle->first_sector = handle->cur_swap;
442 return 0;
443err_rel:
444 release_swap_writer(handle);
445err_close:
446 swsusp_close(FMODE_WRITE);
447 return ret;
448}
449
450static int swap_write_page(struct swap_map_handle *handle, void *buf,
451 struct hib_bio_batch *hb)
452{
453 int error = 0;
454 sector_t offset;
455
456 if (!handle->cur)
457 return -EINVAL;
458 offset = alloc_swapdev_block(root_swap);
459 error = write_page(buf, offset, hb);
460 if (error)
461 return error;
462 handle->cur->entries[handle->k++] = offset;
463 if (handle->k >= MAP_PAGE_ENTRIES) {
464 offset = alloc_swapdev_block(root_swap);
465 if (!offset)
466 return -ENOSPC;
467 handle->cur->next_swap = offset;
468 error = write_page(handle->cur, handle->cur_swap, hb);
469 if (error)
470 goto out;
471 clear_page(handle->cur);
472 handle->cur_swap = offset;
473 handle->k = 0;
474
475 if (hb && low_free_pages() <= handle->reqd_free_pages) {
476 error = hib_wait_io(hb);
477 if (error)
478 goto out;
479 /*
480 * Recalculate the number of required free pages, to
481 * make sure we never take more than half.
482 */
483 handle->reqd_free_pages = reqd_free_pages();
484 }
485 }
486 out:
487 return error;
488}
489
490static int flush_swap_writer(struct swap_map_handle *handle)
491{
492 if (handle->cur && handle->cur_swap)
493 return write_page(handle->cur, handle->cur_swap, NULL);
494 else
495 return -EINVAL;
496}
497
498static int swap_writer_finish(struct swap_map_handle *handle,
499 unsigned int flags, int error)
500{
501 if (!error) {
502 pr_info("S");
503 error = mark_swapfiles(handle, flags);
504 pr_cont("|\n");
505 flush_swap_writer(handle);
506 }
507
508 if (error)
509 free_all_swap_pages(root_swap);
510 release_swap_writer(handle);
511 swsusp_close(FMODE_WRITE);
512
513 return error;
514}
515
516/* We need to remember how much compressed data we need to read. */
517#define LZO_HEADER sizeof(size_t)
518
519/* Number of pages/bytes we'll compress at one time. */
520#define LZO_UNC_PAGES 32
521#define LZO_UNC_SIZE (LZO_UNC_PAGES * PAGE_SIZE)
522
523/* Number of pages/bytes we need for compressed data (worst case). */
524#define LZO_CMP_PAGES DIV_ROUND_UP(lzo1x_worst_compress(LZO_UNC_SIZE) + \
525 LZO_HEADER, PAGE_SIZE)
526#define LZO_CMP_SIZE (LZO_CMP_PAGES * PAGE_SIZE)
527
528/* Maximum number of threads for compression/decompression. */
529#define LZO_THREADS 3
530
531/* Minimum/maximum number of pages for read buffering. */
532#define LZO_MIN_RD_PAGES 1024
533#define LZO_MAX_RD_PAGES 8192
534
535
536/**
537 * save_image - save the suspend image data
538 */
539
540static int save_image(struct swap_map_handle *handle,
541 struct snapshot_handle *snapshot,
542 unsigned int nr_to_write)
543{
544 unsigned int m;
545 int ret;
546 int nr_pages;
547 int err2;
548 struct hib_bio_batch hb;
549 ktime_t start;
550 ktime_t stop;
551
552 hib_init_batch(&hb);
553
554 pr_info("Saving image data pages (%u pages)...\n",
555 nr_to_write);
556 m = nr_to_write / 10;
557 if (!m)
558 m = 1;
559 nr_pages = 0;
560 start = ktime_get();
561 while (1) {
562 ret = snapshot_read_next(snapshot);
563 if (ret <= 0)
564 break;
565 ret = swap_write_page(handle, data_of(*snapshot), &hb);
566 if (ret)
567 break;
568 if (!(nr_pages % m))
569 pr_info("Image saving progress: %3d%%\n",
570 nr_pages / m * 10);
571 nr_pages++;
572 }
573 err2 = hib_wait_io(&hb);
574 hib_finish_batch(&hb);
575 stop = ktime_get();
576 if (!ret)
577 ret = err2;
578 if (!ret)
579 pr_info("Image saving done\n");
580 swsusp_show_speed(start, stop, nr_to_write, "Wrote");
581 return ret;
582}
583
584/**
585 * Structure used for CRC32.
586 */
587struct crc_data {
588 struct task_struct *thr; /* thread */
589 atomic_t ready; /* ready to start flag */
590 atomic_t stop; /* ready to stop flag */
591 unsigned run_threads; /* nr current threads */
592 wait_queue_head_t go; /* start crc update */
593 wait_queue_head_t done; /* crc update done */
594 u32 *crc32; /* points to handle's crc32 */
595 size_t *unc_len[LZO_THREADS]; /* uncompressed lengths */
596 unsigned char *unc[LZO_THREADS]; /* uncompressed data */
597};
598
599/**
600 * CRC32 update function that runs in its own thread.
601 */
602static int crc32_threadfn(void *data)
603{
604 struct crc_data *d = data;
605 unsigned i;
606
607 while (1) {
608 wait_event(d->go, atomic_read(&d->ready) ||
609 kthread_should_stop());
610 if (kthread_should_stop()) {
611 d->thr = NULL;
612 atomic_set(&d->stop, 1);
613 wake_up(&d->done);
614 break;
615 }
616 atomic_set(&d->ready, 0);
617
618 for (i = 0; i < d->run_threads; i++)
619 *d->crc32 = crc32_le(*d->crc32,
620 d->unc[i], *d->unc_len[i]);
621 atomic_set(&d->stop, 1);
622 wake_up(&d->done);
623 }
624 return 0;
625}
626/**
627 * Structure used for LZO data compression.
628 */
629struct cmp_data {
630 struct task_struct *thr; /* thread */
631 atomic_t ready; /* ready to start flag */
632 atomic_t stop; /* ready to stop flag */
633 int ret; /* return code */
634 wait_queue_head_t go; /* start compression */
635 wait_queue_head_t done; /* compression done */
636 size_t unc_len; /* uncompressed length */
637 size_t cmp_len; /* compressed length */
638 unsigned char unc[LZO_UNC_SIZE]; /* uncompressed buffer */
639 unsigned char cmp[LZO_CMP_SIZE]; /* compressed buffer */
640 unsigned char wrk[LZO1X_1_MEM_COMPRESS]; /* compression workspace */
641};
642
643/**
644 * Compression function that runs in its own thread.
645 */
646static int lzo_compress_threadfn(void *data)
647{
648 struct cmp_data *d = data;
649
650 while (1) {
651 wait_event(d->go, atomic_read(&d->ready) ||
652 kthread_should_stop());
653 if (kthread_should_stop()) {
654 d->thr = NULL;
655 d->ret = -1;
656 atomic_set(&d->stop, 1);
657 wake_up(&d->done);
658 break;
659 }
660 atomic_set(&d->ready, 0);
661
662 d->ret = lzo1x_1_compress(d->unc, d->unc_len,
663 d->cmp + LZO_HEADER, &d->cmp_len,
664 d->wrk);
665 atomic_set(&d->stop, 1);
666 wake_up(&d->done);
667 }
668 return 0;
669}
670
671/**
672 * save_image_lzo - Save the suspend image data compressed with LZO.
673 * @handle: Swap map handle to use for saving the image.
674 * @snapshot: Image to read data from.
675 * @nr_to_write: Number of pages to save.
676 */
677static int save_image_lzo(struct swap_map_handle *handle,
678 struct snapshot_handle *snapshot,
679 unsigned int nr_to_write)
680{
681 unsigned int m;
682 int ret = 0;
683 int nr_pages;
684 int err2;
685 struct hib_bio_batch hb;
686 ktime_t start;
687 ktime_t stop;
688 size_t off;
689 unsigned thr, run_threads, nr_threads;
690 unsigned char *page = NULL;
691 struct cmp_data *data = NULL;
692 struct crc_data *crc = NULL;
693
694 hib_init_batch(&hb);
695
696 /*
697 * We'll limit the number of threads for compression to limit memory
698 * footprint.
699 */
700 nr_threads = num_online_cpus() - 1;
701 nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
702
703 page = (void *)__get_free_page(GFP_NOIO | __GFP_HIGH);
704 if (!page) {
705 pr_err("Failed to allocate LZO page\n");
706 ret = -ENOMEM;
707 goto out_clean;
708 }
709
710 data = vzalloc(array_size(nr_threads, sizeof(*data)));
711 if (!data) {
712 pr_err("Failed to allocate LZO data\n");
713 ret = -ENOMEM;
714 goto out_clean;
715 }
716
717 crc = kzalloc(sizeof(*crc), GFP_KERNEL);
718 if (!crc) {
719 pr_err("Failed to allocate crc\n");
720 ret = -ENOMEM;
721 goto out_clean;
722 }
723
724 /*
725 * Start the compression threads.
726 */
727 for (thr = 0; thr < nr_threads; thr++) {
728 init_waitqueue_head(&data[thr].go);
729 init_waitqueue_head(&data[thr].done);
730
731 data[thr].thr = kthread_run(lzo_compress_threadfn,
732 &data[thr],
733 "image_compress/%u", thr);
734 if (IS_ERR(data[thr].thr)) {
735 data[thr].thr = NULL;
736 pr_err("Cannot start compression threads\n");
737 ret = -ENOMEM;
738 goto out_clean;
739 }
740 }
741
742 /*
743 * Start the CRC32 thread.
744 */
745 init_waitqueue_head(&crc->go);
746 init_waitqueue_head(&crc->done);
747
748 handle->crc32 = 0;
749 crc->crc32 = &handle->crc32;
750 for (thr = 0; thr < nr_threads; thr++) {
751 crc->unc[thr] = data[thr].unc;
752 crc->unc_len[thr] = &data[thr].unc_len;
753 }
754
755 crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
756 if (IS_ERR(crc->thr)) {
757 crc->thr = NULL;
758 pr_err("Cannot start CRC32 thread\n");
759 ret = -ENOMEM;
760 goto out_clean;
761 }
762
763 /*
764 * Adjust the number of required free pages after all allocations have
765 * been done. We don't want to run out of pages when writing.
766 */
767 handle->reqd_free_pages = reqd_free_pages();
768
769 pr_info("Using %u thread(s) for compression\n", nr_threads);
770 pr_info("Compressing and saving image data (%u pages)...\n",
771 nr_to_write);
772 m = nr_to_write / 10;
773 if (!m)
774 m = 1;
775 nr_pages = 0;
776 start = ktime_get();
777 for (;;) {
778 for (thr = 0; thr < nr_threads; thr++) {
779 for (off = 0; off < LZO_UNC_SIZE; off += PAGE_SIZE) {
780 ret = snapshot_read_next(snapshot);
781 if (ret < 0)
782 goto out_finish;
783
784 if (!ret)
785 break;
786
787 memcpy(data[thr].unc + off,
788 data_of(*snapshot), PAGE_SIZE);
789
790 if (!(nr_pages % m))
791 pr_info("Image saving progress: %3d%%\n",
792 nr_pages / m * 10);
793 nr_pages++;
794 }
795 if (!off)
796 break;
797
798 data[thr].unc_len = off;
799
800 atomic_set(&data[thr].ready, 1);
801 wake_up(&data[thr].go);
802 }
803
804 if (!thr)
805 break;
806
807 crc->run_threads = thr;
808 atomic_set(&crc->ready, 1);
809 wake_up(&crc->go);
810
811 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
812 wait_event(data[thr].done,
813 atomic_read(&data[thr].stop));
814 atomic_set(&data[thr].stop, 0);
815
816 ret = data[thr].ret;
817
818 if (ret < 0) {
819 pr_err("LZO compression failed\n");
820 goto out_finish;
821 }
822
823 if (unlikely(!data[thr].cmp_len ||
824 data[thr].cmp_len >
825 lzo1x_worst_compress(data[thr].unc_len))) {
826 pr_err("Invalid LZO compressed length\n");
827 ret = -1;
828 goto out_finish;
829 }
830
831 *(size_t *)data[thr].cmp = data[thr].cmp_len;
832
833 /*
834 * Given we are writing one page at a time to disk, we
835 * copy that much from the buffer, although the last
836 * bit will likely be smaller than full page. This is
837 * OK - we saved the length of the compressed data, so
838 * any garbage at the end will be discarded when we
839 * read it.
840 */
841 for (off = 0;
842 off < LZO_HEADER + data[thr].cmp_len;
843 off += PAGE_SIZE) {
844 memcpy(page, data[thr].cmp + off, PAGE_SIZE);
845
846 ret = swap_write_page(handle, page, &hb);
847 if (ret)
848 goto out_finish;
849 }
850 }
851
852 wait_event(crc->done, atomic_read(&crc->stop));
853 atomic_set(&crc->stop, 0);
854 }
855
856out_finish:
857 err2 = hib_wait_io(&hb);
858 stop = ktime_get();
859 if (!ret)
860 ret = err2;
861 if (!ret)
862 pr_info("Image saving done\n");
863 swsusp_show_speed(start, stop, nr_to_write, "Wrote");
864out_clean:
865 hib_finish_batch(&hb);
866 if (crc) {
867 if (crc->thr)
868 kthread_stop(crc->thr);
869 kfree(crc);
870 }
871 if (data) {
872 for (thr = 0; thr < nr_threads; thr++)
873 if (data[thr].thr)
874 kthread_stop(data[thr].thr);
875 vfree(data);
876 }
877 if (page) free_page((unsigned long)page);
878
879 return ret;
880}
881
882/**
883 * enough_swap - Make sure we have enough swap to save the image.
884 *
885 * Returns TRUE or FALSE after checking the total amount of swap
886 * space available from the resume partition.
887 */
888
889static int enough_swap(unsigned int nr_pages)
890{
891 unsigned int free_swap = count_swap_pages(root_swap, 1);
892 unsigned int required;
893
894 pr_debug("Free swap pages: %u\n", free_swap);
895
896 required = PAGES_FOR_IO + nr_pages;
897 return free_swap > required;
898}
899
900/**
901 * swsusp_write - Write entire image and metadata.
902 * @flags: flags to pass to the "boot" kernel in the image header
903 *
904 * It is important _NOT_ to umount filesystems at this point. We want
905 * them synced (in case something goes wrong) but we DO not want to mark
906 * filesystem clean: it is not. (And it does not matter, if we resume
907 * correctly, we'll mark system clean, anyway.)
908 */
909
910int swsusp_write(unsigned int flags)
911{
912 struct swap_map_handle handle;
913 struct snapshot_handle snapshot;
914 struct swsusp_info *header;
915 unsigned long pages;
916 int error;
917
918 pages = snapshot_get_image_size();
919 error = get_swap_writer(&handle);
920 if (error) {
921 pr_err("Cannot get swap writer\n");
922 return error;
923 }
924 if (flags & SF_NOCOMPRESS_MODE) {
925 if (!enough_swap(pages)) {
926 pr_err("Not enough free swap\n");
927 error = -ENOSPC;
928 goto out_finish;
929 }
930 }
931 memset(&snapshot, 0, sizeof(struct snapshot_handle));
932 error = snapshot_read_next(&snapshot);
933 if (error < (int)PAGE_SIZE) {
934 if (error >= 0)
935 error = -EFAULT;
936
937 goto out_finish;
938 }
939 header = (struct swsusp_info *)data_of(snapshot);
940 error = swap_write_page(&handle, header, NULL);
941 if (!error) {
942 error = (flags & SF_NOCOMPRESS_MODE) ?
943 save_image(&handle, &snapshot, pages - 1) :
944 save_image_lzo(&handle, &snapshot, pages - 1);
945 }
946out_finish:
947 error = swap_writer_finish(&handle, flags, error);
948 return error;
949}
950
951/**
952 * The following functions allow us to read data using a swap map
953 * in a file-alike way
954 */
955
956static void release_swap_reader(struct swap_map_handle *handle)
957{
958 struct swap_map_page_list *tmp;
959
960 while (handle->maps) {
961 if (handle->maps->map)
962 free_page((unsigned long)handle->maps->map);
963 tmp = handle->maps;
964 handle->maps = handle->maps->next;
965 kfree(tmp);
966 }
967 handle->cur = NULL;
968}
969
970static int get_swap_reader(struct swap_map_handle *handle,
971 unsigned int *flags_p)
972{
973 int error;
974 struct swap_map_page_list *tmp, *last;
975 sector_t offset;
976
977 *flags_p = swsusp_header->flags;
978
979 if (!swsusp_header->image) /* how can this happen? */
980 return -EINVAL;
981
982 handle->cur = NULL;
983 last = handle->maps = NULL;
984 offset = swsusp_header->image;
985 while (offset) {
986 tmp = kzalloc(sizeof(*handle->maps), GFP_KERNEL);
987 if (!tmp) {
988 release_swap_reader(handle);
989 return -ENOMEM;
990 }
991 if (!handle->maps)
992 handle->maps = tmp;
993 if (last)
994 last->next = tmp;
995 last = tmp;
996
997 tmp->map = (struct swap_map_page *)
998 __get_free_page(GFP_NOIO | __GFP_HIGH);
999 if (!tmp->map) {
1000 release_swap_reader(handle);
1001 return -ENOMEM;
1002 }
1003
1004 error = hib_submit_io(REQ_OP_READ, offset, tmp->map, NULL);
1005 if (error) {
1006 release_swap_reader(handle);
1007 return error;
1008 }
1009 offset = tmp->map->next_swap;
1010 }
1011 handle->k = 0;
1012 handle->cur = handle->maps->map;
1013 return 0;
1014}
1015
1016static int swap_read_page(struct swap_map_handle *handle, void *buf,
1017 struct hib_bio_batch *hb)
1018{
1019 sector_t offset;
1020 int error;
1021 struct swap_map_page_list *tmp;
1022
1023 if (!handle->cur)
1024 return -EINVAL;
1025 offset = handle->cur->entries[handle->k];
1026 if (!offset)
1027 return -EFAULT;
1028 error = hib_submit_io(REQ_OP_READ, offset, buf, hb);
1029 if (error)
1030 return error;
1031 if (++handle->k >= MAP_PAGE_ENTRIES) {
1032 handle->k = 0;
1033 free_page((unsigned long)handle->maps->map);
1034 tmp = handle->maps;
1035 handle->maps = handle->maps->next;
1036 kfree(tmp);
1037 if (!handle->maps)
1038 release_swap_reader(handle);
1039 else
1040 handle->cur = handle->maps->map;
1041 }
1042 return error;
1043}
1044
1045static int swap_reader_finish(struct swap_map_handle *handle)
1046{
1047 release_swap_reader(handle);
1048
1049 return 0;
1050}
1051
1052/**
1053 * load_image - load the image using the swap map handle
1054 * @handle and the snapshot handle @snapshot
1055 * (assume there are @nr_pages pages to load)
1056 */
1057
1058static int load_image(struct swap_map_handle *handle,
1059 struct snapshot_handle *snapshot,
1060 unsigned int nr_to_read)
1061{
1062 unsigned int m;
1063 int ret = 0;
1064 ktime_t start;
1065 ktime_t stop;
1066 struct hib_bio_batch hb;
1067 int err2;
1068 unsigned nr_pages;
1069
1070 hib_init_batch(&hb);
1071
1072 clean_pages_on_read = true;
1073 pr_info("Loading image data pages (%u pages)...\n", 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 pr_info("Image loading progress: %3d%%\n",
1092 nr_pages / m * 10);
1093 nr_pages++;
1094 }
1095 err2 = hib_wait_io(&hb);
1096 hib_finish_batch(&hb);
1097 stop = ktime_get();
1098 if (!ret)
1099 ret = err2;
1100 if (!ret) {
1101 pr_info("Image loading done\n");
1102 snapshot_write_finalize(snapshot);
1103 if (!snapshot_image_loaded(snapshot))
1104 ret = -ENODATA;
1105 }
1106 swsusp_show_speed(start, stop, nr_to_read, "Read");
1107 return ret;
1108}
1109
1110/**
1111 * Structure used for LZO data decompression.
1112 */
1113struct dec_data {
1114 struct task_struct *thr; /* thread */
1115 atomic_t ready; /* ready to start flag */
1116 atomic_t stop; /* ready to stop flag */
1117 int ret; /* return code */
1118 wait_queue_head_t go; /* start decompression */
1119 wait_queue_head_t done; /* decompression done */
1120 size_t unc_len; /* uncompressed length */
1121 size_t cmp_len; /* compressed length */
1122 unsigned char unc[LZO_UNC_SIZE]; /* uncompressed buffer */
1123 unsigned char cmp[LZO_CMP_SIZE]; /* compressed buffer */
1124};
1125
1126/**
1127 * Decompression function that runs in its own thread.
1128 */
1129static int lzo_decompress_threadfn(void *data)
1130{
1131 struct dec_data *d = data;
1132
1133 while (1) {
1134 wait_event(d->go, atomic_read(&d->ready) ||
1135 kthread_should_stop());
1136 if (kthread_should_stop()) {
1137 d->thr = NULL;
1138 d->ret = -1;
1139 atomic_set(&d->stop, 1);
1140 wake_up(&d->done);
1141 break;
1142 }
1143 atomic_set(&d->ready, 0);
1144
1145 d->unc_len = LZO_UNC_SIZE;
1146 d->ret = lzo1x_decompress_safe(d->cmp + LZO_HEADER, d->cmp_len,
1147 d->unc, &d->unc_len);
1148 if (clean_pages_on_decompress)
1149 flush_icache_range((unsigned long)d->unc,
1150 (unsigned long)d->unc + d->unc_len);
1151
1152 atomic_set(&d->stop, 1);
1153 wake_up(&d->done);
1154 }
1155 return 0;
1156}
1157
1158/**
1159 * load_image_lzo - Load compressed image data and decompress them with LZO.
1160 * @handle: Swap map handle to use for loading data.
1161 * @snapshot: Image to copy uncompressed data into.
1162 * @nr_to_read: Number of pages to load.
1163 */
1164static int load_image_lzo(struct swap_map_handle *handle,
1165 struct snapshot_handle *snapshot,
1166 unsigned int nr_to_read)
1167{
1168 unsigned int m;
1169 int ret = 0;
1170 int eof = 0;
1171 struct hib_bio_batch hb;
1172 ktime_t start;
1173 ktime_t stop;
1174 unsigned nr_pages;
1175 size_t off;
1176 unsigned i, thr, run_threads, nr_threads;
1177 unsigned ring = 0, pg = 0, ring_size = 0,
1178 have = 0, want, need, asked = 0;
1179 unsigned long read_pages = 0;
1180 unsigned char **page = NULL;
1181 struct dec_data *data = NULL;
1182 struct crc_data *crc = NULL;
1183
1184 hib_init_batch(&hb);
1185
1186 /*
1187 * We'll limit the number of threads for decompression to limit memory
1188 * footprint.
1189 */
1190 nr_threads = num_online_cpus() - 1;
1191 nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
1192
1193 page = vmalloc(array_size(LZO_MAX_RD_PAGES, sizeof(*page)));
1194 if (!page) {
1195 pr_err("Failed to allocate LZO page\n");
1196 ret = -ENOMEM;
1197 goto out_clean;
1198 }
1199
1200 data = vzalloc(array_size(nr_threads, sizeof(*data)));
1201 if (!data) {
1202 pr_err("Failed to allocate LZO data\n");
1203 ret = -ENOMEM;
1204 goto out_clean;
1205 }
1206
1207 crc = kzalloc(sizeof(*crc), GFP_KERNEL);
1208 if (!crc) {
1209 pr_err("Failed to allocate crc\n");
1210 ret = -ENOMEM;
1211 goto out_clean;
1212 }
1213
1214 clean_pages_on_decompress = true;
1215
1216 /*
1217 * Start the decompression threads.
1218 */
1219 for (thr = 0; thr < nr_threads; thr++) {
1220 init_waitqueue_head(&data[thr].go);
1221 init_waitqueue_head(&data[thr].done);
1222
1223 data[thr].thr = kthread_run(lzo_decompress_threadfn,
1224 &data[thr],
1225 "image_decompress/%u", thr);
1226 if (IS_ERR(data[thr].thr)) {
1227 data[thr].thr = NULL;
1228 pr_err("Cannot start decompression threads\n");
1229 ret = -ENOMEM;
1230 goto out_clean;
1231 }
1232 }
1233
1234 /*
1235 * Start the CRC32 thread.
1236 */
1237 init_waitqueue_head(&crc->go);
1238 init_waitqueue_head(&crc->done);
1239
1240 handle->crc32 = 0;
1241 crc->crc32 = &handle->crc32;
1242 for (thr = 0; thr < nr_threads; thr++) {
1243 crc->unc[thr] = data[thr].unc;
1244 crc->unc_len[thr] = &data[thr].unc_len;
1245 }
1246
1247 crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
1248 if (IS_ERR(crc->thr)) {
1249 crc->thr = NULL;
1250 pr_err("Cannot start CRC32 thread\n");
1251 ret = -ENOMEM;
1252 goto out_clean;
1253 }
1254
1255 /*
1256 * Set the number of pages for read buffering.
1257 * This is complete guesswork, because we'll only know the real
1258 * picture once prepare_image() is called, which is much later on
1259 * during the image load phase. We'll assume the worst case and
1260 * say that none of the image pages are from high memory.
1261 */
1262 if (low_free_pages() > snapshot_get_image_size())
1263 read_pages = (low_free_pages() - snapshot_get_image_size()) / 2;
1264 read_pages = clamp_val(read_pages, LZO_MIN_RD_PAGES, LZO_MAX_RD_PAGES);
1265
1266 for (i = 0; i < read_pages; i++) {
1267 page[i] = (void *)__get_free_page(i < LZO_CMP_PAGES ?
1268 GFP_NOIO | __GFP_HIGH :
1269 GFP_NOIO | __GFP_NOWARN |
1270 __GFP_NORETRY);
1271
1272 if (!page[i]) {
1273 if (i < LZO_CMP_PAGES) {
1274 ring_size = i;
1275 pr_err("Failed to allocate LZO pages\n");
1276 ret = -ENOMEM;
1277 goto out_clean;
1278 } else {
1279 break;
1280 }
1281 }
1282 }
1283 want = ring_size = i;
1284
1285 pr_info("Using %u thread(s) for decompression\n", nr_threads);
1286 pr_info("Loading and decompressing image data (%u pages)...\n",
1287 nr_to_read);
1288 m = nr_to_read / 10;
1289 if (!m)
1290 m = 1;
1291 nr_pages = 0;
1292 start = ktime_get();
1293
1294 ret = snapshot_write_next(snapshot);
1295 if (ret <= 0)
1296 goto out_finish;
1297
1298 for(;;) {
1299 for (i = 0; !eof && i < want; i++) {
1300 ret = swap_read_page(handle, page[ring], &hb);
1301 if (ret) {
1302 /*
1303 * On real read error, finish. On end of data,
1304 * set EOF flag and just exit the read loop.
1305 */
1306 if (handle->cur &&
1307 handle->cur->entries[handle->k]) {
1308 goto out_finish;
1309 } else {
1310 eof = 1;
1311 break;
1312 }
1313 }
1314 if (++ring >= ring_size)
1315 ring = 0;
1316 }
1317 asked += i;
1318 want -= i;
1319
1320 /*
1321 * We are out of data, wait for some more.
1322 */
1323 if (!have) {
1324 if (!asked)
1325 break;
1326
1327 ret = hib_wait_io(&hb);
1328 if (ret)
1329 goto out_finish;
1330 have += asked;
1331 asked = 0;
1332 if (eof)
1333 eof = 2;
1334 }
1335
1336 if (crc->run_threads) {
1337 wait_event(crc->done, atomic_read(&crc->stop));
1338 atomic_set(&crc->stop, 0);
1339 crc->run_threads = 0;
1340 }
1341
1342 for (thr = 0; have && thr < nr_threads; thr++) {
1343 data[thr].cmp_len = *(size_t *)page[pg];
1344 if (unlikely(!data[thr].cmp_len ||
1345 data[thr].cmp_len >
1346 lzo1x_worst_compress(LZO_UNC_SIZE))) {
1347 pr_err("Invalid LZO compressed length\n");
1348 ret = -1;
1349 goto out_finish;
1350 }
1351
1352 need = DIV_ROUND_UP(data[thr].cmp_len + LZO_HEADER,
1353 PAGE_SIZE);
1354 if (need > have) {
1355 if (eof > 1) {
1356 ret = -1;
1357 goto out_finish;
1358 }
1359 break;
1360 }
1361
1362 for (off = 0;
1363 off < LZO_HEADER + data[thr].cmp_len;
1364 off += PAGE_SIZE) {
1365 memcpy(data[thr].cmp + off,
1366 page[pg], PAGE_SIZE);
1367 have--;
1368 want++;
1369 if (++pg >= ring_size)
1370 pg = 0;
1371 }
1372
1373 atomic_set(&data[thr].ready, 1);
1374 wake_up(&data[thr].go);
1375 }
1376
1377 /*
1378 * Wait for more data while we are decompressing.
1379 */
1380 if (have < LZO_CMP_PAGES && asked) {
1381 ret = hib_wait_io(&hb);
1382 if (ret)
1383 goto out_finish;
1384 have += asked;
1385 asked = 0;
1386 if (eof)
1387 eof = 2;
1388 }
1389
1390 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
1391 wait_event(data[thr].done,
1392 atomic_read(&data[thr].stop));
1393 atomic_set(&data[thr].stop, 0);
1394
1395 ret = data[thr].ret;
1396
1397 if (ret < 0) {
1398 pr_err("LZO decompression failed\n");
1399 goto out_finish;
1400 }
1401
1402 if (unlikely(!data[thr].unc_len ||
1403 data[thr].unc_len > LZO_UNC_SIZE ||
1404 data[thr].unc_len & (PAGE_SIZE - 1))) {
1405 pr_err("Invalid LZO uncompressed length\n");
1406 ret = -1;
1407 goto out_finish;
1408 }
1409
1410 for (off = 0;
1411 off < data[thr].unc_len; off += PAGE_SIZE) {
1412 memcpy(data_of(*snapshot),
1413 data[thr].unc + off, PAGE_SIZE);
1414
1415 if (!(nr_pages % m))
1416 pr_info("Image loading progress: %3d%%\n",
1417 nr_pages / m * 10);
1418 nr_pages++;
1419
1420 ret = snapshot_write_next(snapshot);
1421 if (ret <= 0) {
1422 crc->run_threads = thr + 1;
1423 atomic_set(&crc->ready, 1);
1424 wake_up(&crc->go);
1425 goto out_finish;
1426 }
1427 }
1428 }
1429
1430 crc->run_threads = thr;
1431 atomic_set(&crc->ready, 1);
1432 wake_up(&crc->go);
1433 }
1434
1435out_finish:
1436 if (crc->run_threads) {
1437 wait_event(crc->done, atomic_read(&crc->stop));
1438 atomic_set(&crc->stop, 0);
1439 }
1440 stop = ktime_get();
1441 if (!ret) {
1442 pr_info("Image loading done\n");
1443 snapshot_write_finalize(snapshot);
1444 if (!snapshot_image_loaded(snapshot))
1445 ret = -ENODATA;
1446 if (!ret) {
1447 if (swsusp_header->flags & SF_CRC32_MODE) {
1448 if(handle->crc32 != swsusp_header->crc32) {
1449 pr_err("Invalid image CRC32!\n");
1450 ret = -ENODATA;
1451 }
1452 }
1453 }
1454 }
1455 swsusp_show_speed(start, stop, nr_to_read, "Read");
1456out_clean:
1457 hib_finish_batch(&hb);
1458 for (i = 0; i < ring_size; i++)
1459 free_page((unsigned long)page[i]);
1460 if (crc) {
1461 if (crc->thr)
1462 kthread_stop(crc->thr);
1463 kfree(crc);
1464 }
1465 if (data) {
1466 for (thr = 0; thr < nr_threads; thr++)
1467 if (data[thr].thr)
1468 kthread_stop(data[thr].thr);
1469 vfree(data);
1470 }
1471 vfree(page);
1472
1473 return ret;
1474}
1475
1476/**
1477 * swsusp_read - read the hibernation image.
1478 * @flags_p: flags passed by the "frozen" kernel in the image header should
1479 * be written into this memory location
1480 */
1481
1482int swsusp_read(unsigned int *flags_p)
1483{
1484 int error;
1485 struct swap_map_handle handle;
1486 struct snapshot_handle snapshot;
1487 struct swsusp_info *header;
1488
1489 memset(&snapshot, 0, sizeof(struct snapshot_handle));
1490 error = snapshot_write_next(&snapshot);
1491 if (error < (int)PAGE_SIZE)
1492 return error < 0 ? error : -EFAULT;
1493 header = (struct swsusp_info *)data_of(snapshot);
1494 error = get_swap_reader(&handle, flags_p);
1495 if (error)
1496 goto end;
1497 if (!error)
1498 error = swap_read_page(&handle, header, NULL);
1499 if (!error) {
1500 error = (*flags_p & SF_NOCOMPRESS_MODE) ?
1501 load_image(&handle, &snapshot, header->pages - 1) :
1502 load_image_lzo(&handle, &snapshot, header->pages - 1);
1503 }
1504 swap_reader_finish(&handle);
1505end:
1506 if (!error)
1507 pr_debug("Image successfully loaded\n");
1508 else
1509 pr_debug("Error %d resuming\n", error);
1510 return error;
1511}
1512
1513/**
1514 * swsusp_check - Check for swsusp signature in the resume device
1515 */
1516
1517int swsusp_check(void)
1518{
1519 int error;
1520 void *holder;
1521
1522 hib_resume_bdev = blkdev_get_by_dev(swsusp_resume_device,
1523 FMODE_READ | FMODE_EXCL, &holder);
1524 if (!IS_ERR(hib_resume_bdev)) {
1525 set_blocksize(hib_resume_bdev, PAGE_SIZE);
1526 clear_page(swsusp_header);
1527 error = hib_submit_io(REQ_OP_READ, swsusp_resume_block,
1528 swsusp_header, NULL);
1529 if (error)
1530 goto put;
1531
1532 if (!memcmp(HIBERNATE_SIG, swsusp_header->sig, 10)) {
1533 memcpy(swsusp_header->sig, swsusp_header->orig_sig, 10);
1534 /* Reset swap signature now */
1535 error = hib_submit_io(REQ_OP_WRITE | REQ_SYNC,
1536 swsusp_resume_block,
1537 swsusp_header, NULL);
1538 } else {
1539 error = -EINVAL;
1540 }
1541 if (!error && swsusp_header->flags & SF_HW_SIG &&
1542 swsusp_header->hw_sig != swsusp_hardware_signature) {
1543 pr_info("Suspend image hardware signature mismatch (%08x now %08x); aborting resume.\n",
1544 swsusp_header->hw_sig, swsusp_hardware_signature);
1545 error = -EINVAL;
1546 }
1547
1548put:
1549 if (error)
1550 blkdev_put(hib_resume_bdev, FMODE_READ | FMODE_EXCL);
1551 else
1552 pr_debug("Image signature found, resuming\n");
1553 } else {
1554 error = PTR_ERR(hib_resume_bdev);
1555 }
1556
1557 if (error)
1558 pr_debug("Image not found (code %d)\n", error);
1559
1560 return error;
1561}
1562
1563/**
1564 * swsusp_close - close swap device.
1565 */
1566
1567void swsusp_close(fmode_t mode)
1568{
1569 if (IS_ERR(hib_resume_bdev)) {
1570 pr_debug("Image device not initialised\n");
1571 return;
1572 }
1573
1574 blkdev_put(hib_resume_bdev, mode);
1575}
1576
1577/**
1578 * swsusp_unmark - Unmark swsusp signature in the resume device
1579 */
1580
1581#ifdef CONFIG_SUSPEND
1582int swsusp_unmark(void)
1583{
1584 int error;
1585
1586 hib_submit_io(REQ_OP_READ, swsusp_resume_block,
1587 swsusp_header, NULL);
1588 if (!memcmp(HIBERNATE_SIG,swsusp_header->sig, 10)) {
1589 memcpy(swsusp_header->sig,swsusp_header->orig_sig, 10);
1590 error = hib_submit_io(REQ_OP_WRITE | REQ_SYNC,
1591 swsusp_resume_block,
1592 swsusp_header, NULL);
1593 } else {
1594 pr_err("Cannot find swsusp signature!\n");
1595 error = -ENODEV;
1596 }
1597
1598 /*
1599 * We just returned from suspend, we don't need the image any more.
1600 */
1601 free_all_swap_pages(root_swap);
1602
1603 return error;
1604}
1605#endif
1606
1607static int __init swsusp_header_init(void)
1608{
1609 swsusp_header = (struct swsusp_header*) __get_free_page(GFP_KERNEL);
1610 if (!swsusp_header)
1611 panic("Could not allocate memory for swsusp_header\n");
1612 return 0;
1613}
1614
1615core_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
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);