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