Merge tag 'io_uring-5.11-2021-01-16' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / kernel / power / swap.c
blobc73f2e295167d141cdc68fd77cc8ac60eb371226
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/kernel/power/swap.c
5 * This file provides functions for reading the suspend image from
6 * and writing it to a swap partition.
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>
13 #define pr_fmt(fmt) "PM: " fmt
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>
35 #include "power.h"
37 #define HIBERNATE_SIG "S1SUSPEND"
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.
44 static bool clean_pages_on_read;
45 static bool clean_pages_on_decompress;
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.
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.
58 * During resume we pick up all swap_map_page structures into a list.
61 #define MAP_PAGE_ENTRIES (PAGE_SIZE / sizeof(sector_t) - 1)
64 * Number of free pages that are not high.
66 static inline unsigned long low_free_pages(void)
68 return nr_free_pages() - nr_free_highpages();
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.
75 static inline unsigned long reqd_free_pages(void)
77 return low_free_pages() / 2;
80 struct swap_map_page {
81 sector_t entries[MAP_PAGE_ENTRIES];
82 sector_t next_swap;
85 struct swap_map_page_list {
86 struct swap_map_page *map;
87 struct swap_map_page_list *next;
90 /**
91 * The swap_map_handle structure is used for handling swap in
92 * a file-alike way
95 struct 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;
105 struct 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;
115 static struct swsusp_header *swsusp_header;
118 * The following functions are used for tracing the allocated
119 * swap pages, so that they can be freed in case of an error.
122 struct swsusp_extent {
123 struct rb_node node;
124 unsigned long start;
125 unsigned long end;
128 static struct rb_root swsusp_extents = RB_ROOT;
130 static int swsusp_extents_insert(unsigned long swap_offset)
132 struct rb_node **new = &(swsusp_extents.rb_node);
133 struct rb_node *parent = NULL;
134 struct swsusp_extent *ext;
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;
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;
153 new = &((*new)->rb_right);
154 } else {
155 /* It already is in the tree */
156 return -EINVAL;
159 /* Add the new node and rebalance the tree. */
160 ext = kzalloc(sizeof(struct swsusp_extent), GFP_KERNEL);
161 if (!ext)
162 return -ENOMEM;
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;
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.
176 sector_t alloc_swapdev_block(int swap)
178 unsigned long offset;
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);
187 return 0;
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.
196 void free_all_swap_pages(int swap)
198 struct rb_node *node;
200 while ((node = swsusp_extents.rb_node)) {
201 struct swsusp_extent *ext;
202 unsigned long offset;
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));
209 kfree(ext);
213 int swsusp_swap_in_use(void)
215 return (swsusp_extents.rb_node != NULL);
219 * General things
222 static unsigned short root_swap = 0xffff;
223 static struct block_device *hib_resume_bdev;
225 struct hib_bio_batch {
226 atomic_t count;
227 wait_queue_head_t wait;
228 blk_status_t error;
229 struct blk_plug plug;
232 static void hib_init_batch(struct hib_bio_batch *hb)
234 atomic_set(&hb->count, 0);
235 init_waitqueue_head(&hb->wait);
236 hb->error = BLK_STS_OK;
237 blk_start_plug(&hb->plug);
240 static void hib_finish_batch(struct hib_bio_batch *hb)
242 blk_finish_plug(&hb->plug);
245 static void hib_end_io(struct bio *bio)
247 struct hib_bio_batch *hb = bio->bi_private;
248 struct page *page = bio_first_page_all(bio);
250 if (bio->bi_status) {
251 pr_alert("Read-error on swap-device (%u:%u:%Lu)\n",
252 MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
253 (unsigned long long)bio->bi_iter.bi_sector);
256 if (bio_data_dir(bio) == WRITE)
257 put_page(page);
258 else if (clean_pages_on_read)
259 flush_icache_range((unsigned long)page_address(page),
260 (unsigned long)page_address(page) + PAGE_SIZE);
262 if (bio->bi_status && !hb->error)
263 hb->error = bio->bi_status;
264 if (atomic_dec_and_test(&hb->count))
265 wake_up(&hb->wait);
267 bio_put(bio);
270 static int hib_submit_io(int op, int op_flags, pgoff_t page_off, void *addr,
271 struct hib_bio_batch *hb)
273 struct page *page = virt_to_page(addr);
274 struct bio *bio;
275 int error = 0;
277 bio = bio_alloc(GFP_NOIO | __GFP_HIGH, 1);
278 bio->bi_iter.bi_sector = page_off * (PAGE_SIZE >> 9);
279 bio_set_dev(bio, hib_resume_bdev);
280 bio_set_op_attrs(bio, op, op_flags);
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;
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);
299 return error;
302 static blk_status_t hib_wait_io(struct hib_bio_batch *hb)
305 * We are relying on the behavior of blk_plug that a thread with
306 * a plug will flush the plug list before sleeping.
308 wait_event(hb->wait, atomic_read(&hb->count) == 0);
309 return blk_status_to_errno(hb->error);
313 * Saving part
316 static int mark_swapfiles(struct swap_map_handle *handle, unsigned int flags)
318 int error;
320 hib_submit_io(REQ_OP_READ, 0, swsusp_resume_block,
321 swsusp_header, NULL);
322 if (!memcmp("SWAP-SPACE",swsusp_header->sig, 10) ||
323 !memcmp("SWAPSPACE2",swsusp_header->sig, 10)) {
324 memcpy(swsusp_header->orig_sig,swsusp_header->sig, 10);
325 memcpy(swsusp_header->sig, HIBERNATE_SIG, 10);
326 swsusp_header->image = handle->first_sector;
327 swsusp_header->flags = flags;
328 if (flags & SF_CRC32_MODE)
329 swsusp_header->crc32 = handle->crc32;
330 error = hib_submit_io(REQ_OP_WRITE, REQ_SYNC,
331 swsusp_resume_block, swsusp_header, NULL);
332 } else {
333 pr_err("Swap header not found!\n");
334 error = -ENODEV;
336 return error;
340 * swsusp_swap_check - check if the resume device is a swap device
341 * and get its index (if so)
343 * This is called before saving image
345 static int swsusp_swap_check(void)
347 int res;
349 if (swsusp_resume_device)
350 res = swap_type_of(swsusp_resume_device, swsusp_resume_block);
351 else
352 res = find_first_swap(&swsusp_resume_device);
353 if (res < 0)
354 return res;
355 root_swap = res;
357 hib_resume_bdev = blkdev_get_by_dev(swsusp_resume_device, FMODE_WRITE,
358 NULL);
359 if (IS_ERR(hib_resume_bdev))
360 return PTR_ERR(hib_resume_bdev);
362 res = set_blocksize(hib_resume_bdev, PAGE_SIZE);
363 if (res < 0)
364 blkdev_put(hib_resume_bdev, FMODE_WRITE);
366 return res;
370 * write_page - Write one page to given swap location.
371 * @buf: Address we're writing.
372 * @offset: Offset of the swap page we're writing to.
373 * @hb: bio completion batch
376 static int write_page(void *buf, sector_t offset, struct hib_bio_batch *hb)
378 void *src;
379 int ret;
381 if (!offset)
382 return -ENOSPC;
384 if (hb) {
385 src = (void *)__get_free_page(GFP_NOIO | __GFP_NOWARN |
386 __GFP_NORETRY);
387 if (src) {
388 copy_page(src, buf);
389 } else {
390 ret = hib_wait_io(hb); /* Free pages */
391 if (ret)
392 return ret;
393 src = (void *)__get_free_page(GFP_NOIO |
394 __GFP_NOWARN |
395 __GFP_NORETRY);
396 if (src) {
397 copy_page(src, buf);
398 } else {
399 WARN_ON_ONCE(1);
400 hb = NULL; /* Go synchronous */
401 src = buf;
404 } else {
405 src = buf;
407 return hib_submit_io(REQ_OP_WRITE, REQ_SYNC, offset, src, hb);
410 static void release_swap_writer(struct swap_map_handle *handle)
412 if (handle->cur)
413 free_page((unsigned long)handle->cur);
414 handle->cur = NULL;
417 static int get_swap_writer(struct swap_map_handle *handle)
419 int ret;
421 ret = swsusp_swap_check();
422 if (ret) {
423 if (ret != -ENOSPC)
424 pr_err("Cannot find swap device, try swapon -a\n");
425 return ret;
427 handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_KERNEL);
428 if (!handle->cur) {
429 ret = -ENOMEM;
430 goto err_close;
432 handle->cur_swap = alloc_swapdev_block(root_swap);
433 if (!handle->cur_swap) {
434 ret = -ENOSPC;
435 goto err_rel;
437 handle->k = 0;
438 handle->reqd_free_pages = reqd_free_pages();
439 handle->first_sector = handle->cur_swap;
440 return 0;
441 err_rel:
442 release_swap_writer(handle);
443 err_close:
444 swsusp_close(FMODE_WRITE);
445 return ret;
448 static int swap_write_page(struct swap_map_handle *handle, void *buf,
449 struct hib_bio_batch *hb)
451 int error = 0;
452 sector_t offset;
454 if (!handle->cur)
455 return -EINVAL;
456 offset = alloc_swapdev_block(root_swap);
457 error = write_page(buf, offset, hb);
458 if (error)
459 return error;
460 handle->cur->entries[handle->k++] = offset;
461 if (handle->k >= MAP_PAGE_ENTRIES) {
462 offset = alloc_swapdev_block(root_swap);
463 if (!offset)
464 return -ENOSPC;
465 handle->cur->next_swap = offset;
466 error = write_page(handle->cur, handle->cur_swap, hb);
467 if (error)
468 goto out;
469 clear_page(handle->cur);
470 handle->cur_swap = offset;
471 handle->k = 0;
473 if (hb && low_free_pages() <= handle->reqd_free_pages) {
474 error = hib_wait_io(hb);
475 if (error)
476 goto out;
478 * Recalculate the number of required free pages, to
479 * make sure we never take more than half.
481 handle->reqd_free_pages = reqd_free_pages();
484 out:
485 return error;
488 static int flush_swap_writer(struct swap_map_handle *handle)
490 if (handle->cur && handle->cur_swap)
491 return write_page(handle->cur, handle->cur_swap, NULL);
492 else
493 return -EINVAL;
496 static int swap_writer_finish(struct swap_map_handle *handle,
497 unsigned int flags, int error)
499 if (!error) {
500 flush_swap_writer(handle);
501 pr_info("S");
502 error = mark_swapfiles(handle, flags);
503 pr_cont("|\n");
506 if (error)
507 free_all_swap_pages(root_swap);
508 release_swap_writer(handle);
509 swsusp_close(FMODE_WRITE);
511 return error;
514 /* We need to remember how much compressed data we need to read. */
515 #define LZO_HEADER sizeof(size_t)
517 /* Number of pages/bytes we'll compress at one time. */
518 #define LZO_UNC_PAGES 32
519 #define LZO_UNC_SIZE (LZO_UNC_PAGES * PAGE_SIZE)
521 /* Number of pages/bytes we need for compressed data (worst case). */
522 #define LZO_CMP_PAGES DIV_ROUND_UP(lzo1x_worst_compress(LZO_UNC_SIZE) + \
523 LZO_HEADER, PAGE_SIZE)
524 #define LZO_CMP_SIZE (LZO_CMP_PAGES * PAGE_SIZE)
526 /* Maximum number of threads for compression/decompression. */
527 #define LZO_THREADS 3
529 /* Minimum/maximum number of pages for read buffering. */
530 #define LZO_MIN_RD_PAGES 1024
531 #define LZO_MAX_RD_PAGES 8192
535 * save_image - save the suspend image data
538 static int save_image(struct swap_map_handle *handle,
539 struct snapshot_handle *snapshot,
540 unsigned int nr_to_write)
542 unsigned int m;
543 int ret;
544 int nr_pages;
545 int err2;
546 struct hib_bio_batch hb;
547 ktime_t start;
548 ktime_t stop;
550 hib_init_batch(&hb);
552 pr_info("Saving image data pages (%u pages)...\n",
553 nr_to_write);
554 m = nr_to_write / 10;
555 if (!m)
556 m = 1;
557 nr_pages = 0;
558 start = ktime_get();
559 while (1) {
560 ret = snapshot_read_next(snapshot);
561 if (ret <= 0)
562 break;
563 ret = swap_write_page(handle, data_of(*snapshot), &hb);
564 if (ret)
565 break;
566 if (!(nr_pages % m))
567 pr_info("Image saving progress: %3d%%\n",
568 nr_pages / m * 10);
569 nr_pages++;
571 err2 = hib_wait_io(&hb);
572 hib_finish_batch(&hb);
573 stop = ktime_get();
574 if (!ret)
575 ret = err2;
576 if (!ret)
577 pr_info("Image saving done\n");
578 swsusp_show_speed(start, stop, nr_to_write, "Wrote");
579 return ret;
583 * Structure used for CRC32.
585 struct crc_data {
586 struct task_struct *thr; /* thread */
587 atomic_t ready; /* ready to start flag */
588 atomic_t stop; /* ready to stop flag */
589 unsigned run_threads; /* nr current threads */
590 wait_queue_head_t go; /* start crc update */
591 wait_queue_head_t done; /* crc update done */
592 u32 *crc32; /* points to handle's crc32 */
593 size_t *unc_len[LZO_THREADS]; /* uncompressed lengths */
594 unsigned char *unc[LZO_THREADS]; /* uncompressed data */
598 * CRC32 update function that runs in its own thread.
600 static int crc32_threadfn(void *data)
602 struct crc_data *d = data;
603 unsigned i;
605 while (1) {
606 wait_event(d->go, atomic_read(&d->ready) ||
607 kthread_should_stop());
608 if (kthread_should_stop()) {
609 d->thr = NULL;
610 atomic_set(&d->stop, 1);
611 wake_up(&d->done);
612 break;
614 atomic_set(&d->ready, 0);
616 for (i = 0; i < d->run_threads; i++)
617 *d->crc32 = crc32_le(*d->crc32,
618 d->unc[i], *d->unc_len[i]);
619 atomic_set(&d->stop, 1);
620 wake_up(&d->done);
622 return 0;
625 * Structure used for LZO data compression.
627 struct cmp_data {
628 struct task_struct *thr; /* thread */
629 atomic_t ready; /* ready to start flag */
630 atomic_t stop; /* ready to stop flag */
631 int ret; /* return code */
632 wait_queue_head_t go; /* start compression */
633 wait_queue_head_t done; /* compression done */
634 size_t unc_len; /* uncompressed length */
635 size_t cmp_len; /* compressed length */
636 unsigned char unc[LZO_UNC_SIZE]; /* uncompressed buffer */
637 unsigned char cmp[LZO_CMP_SIZE]; /* compressed buffer */
638 unsigned char wrk[LZO1X_1_MEM_COMPRESS]; /* compression workspace */
642 * Compression function that runs in its own thread.
644 static int lzo_compress_threadfn(void *data)
646 struct cmp_data *d = data;
648 while (1) {
649 wait_event(d->go, atomic_read(&d->ready) ||
650 kthread_should_stop());
651 if (kthread_should_stop()) {
652 d->thr = NULL;
653 d->ret = -1;
654 atomic_set(&d->stop, 1);
655 wake_up(&d->done);
656 break;
658 atomic_set(&d->ready, 0);
660 d->ret = lzo1x_1_compress(d->unc, d->unc_len,
661 d->cmp + LZO_HEADER, &d->cmp_len,
662 d->wrk);
663 atomic_set(&d->stop, 1);
664 wake_up(&d->done);
666 return 0;
670 * save_image_lzo - Save the suspend image data compressed with LZO.
671 * @handle: Swap map handle to use for saving the image.
672 * @snapshot: Image to read data from.
673 * @nr_to_write: Number of pages to save.
675 static int save_image_lzo(struct swap_map_handle *handle,
676 struct snapshot_handle *snapshot,
677 unsigned int nr_to_write)
679 unsigned int m;
680 int ret = 0;
681 int nr_pages;
682 int err2;
683 struct hib_bio_batch hb;
684 ktime_t start;
685 ktime_t stop;
686 size_t off;
687 unsigned thr, run_threads, nr_threads;
688 unsigned char *page = NULL;
689 struct cmp_data *data = NULL;
690 struct crc_data *crc = NULL;
692 hib_init_batch(&hb);
695 * We'll limit the number of threads for compression to limit memory
696 * footprint.
698 nr_threads = num_online_cpus() - 1;
699 nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
701 page = (void *)__get_free_page(GFP_NOIO | __GFP_HIGH);
702 if (!page) {
703 pr_err("Failed to allocate LZO page\n");
704 ret = -ENOMEM;
705 goto out_clean;
708 data = vmalloc(array_size(nr_threads, sizeof(*data)));
709 if (!data) {
710 pr_err("Failed to allocate LZO data\n");
711 ret = -ENOMEM;
712 goto out_clean;
714 for (thr = 0; thr < nr_threads; thr++)
715 memset(&data[thr], 0, offsetof(struct cmp_data, go));
717 crc = kmalloc(sizeof(*crc), GFP_KERNEL);
718 if (!crc) {
719 pr_err("Failed to allocate crc\n");
720 ret = -ENOMEM;
721 goto out_clean;
723 memset(crc, 0, offsetof(struct crc_data, go));
726 * Start the compression threads.
728 for (thr = 0; thr < nr_threads; thr++) {
729 init_waitqueue_head(&data[thr].go);
730 init_waitqueue_head(&data[thr].done);
732 data[thr].thr = kthread_run(lzo_compress_threadfn,
733 &data[thr],
734 "image_compress/%u", thr);
735 if (IS_ERR(data[thr].thr)) {
736 data[thr].thr = NULL;
737 pr_err("Cannot start compression threads\n");
738 ret = -ENOMEM;
739 goto out_clean;
744 * Start the CRC32 thread.
746 init_waitqueue_head(&crc->go);
747 init_waitqueue_head(&crc->done);
749 handle->crc32 = 0;
750 crc->crc32 = &handle->crc32;
751 for (thr = 0; thr < nr_threads; thr++) {
752 crc->unc[thr] = data[thr].unc;
753 crc->unc_len[thr] = &data[thr].unc_len;
756 crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
757 if (IS_ERR(crc->thr)) {
758 crc->thr = NULL;
759 pr_err("Cannot start CRC32 thread\n");
760 ret = -ENOMEM;
761 goto out_clean;
765 * Adjust the number of required free pages after all allocations have
766 * been done. We don't want to run out of pages when writing.
768 handle->reqd_free_pages = reqd_free_pages();
770 pr_info("Using %u thread(s) for compression\n", nr_threads);
771 pr_info("Compressing and saving image data (%u pages)...\n",
772 nr_to_write);
773 m = nr_to_write / 10;
774 if (!m)
775 m = 1;
776 nr_pages = 0;
777 start = ktime_get();
778 for (;;) {
779 for (thr = 0; thr < nr_threads; thr++) {
780 for (off = 0; off < LZO_UNC_SIZE; off += PAGE_SIZE) {
781 ret = snapshot_read_next(snapshot);
782 if (ret < 0)
783 goto out_finish;
785 if (!ret)
786 break;
788 memcpy(data[thr].unc + off,
789 data_of(*snapshot), PAGE_SIZE);
791 if (!(nr_pages % m))
792 pr_info("Image saving progress: %3d%%\n",
793 nr_pages / m * 10);
794 nr_pages++;
796 if (!off)
797 break;
799 data[thr].unc_len = off;
801 atomic_set(&data[thr].ready, 1);
802 wake_up(&data[thr].go);
805 if (!thr)
806 break;
808 crc->run_threads = thr;
809 atomic_set(&crc->ready, 1);
810 wake_up(&crc->go);
812 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
813 wait_event(data[thr].done,
814 atomic_read(&data[thr].stop));
815 atomic_set(&data[thr].stop, 0);
817 ret = data[thr].ret;
819 if (ret < 0) {
820 pr_err("LZO compression failed\n");
821 goto out_finish;
824 if (unlikely(!data[thr].cmp_len ||
825 data[thr].cmp_len >
826 lzo1x_worst_compress(data[thr].unc_len))) {
827 pr_err("Invalid LZO compressed length\n");
828 ret = -1;
829 goto out_finish;
832 *(size_t *)data[thr].cmp = data[thr].cmp_len;
835 * Given we are writing one page at a time to disk, we
836 * copy that much from the buffer, although the last
837 * bit will likely be smaller than full page. This is
838 * OK - we saved the length of the compressed data, so
839 * any garbage at the end will be discarded when we
840 * read it.
842 for (off = 0;
843 off < LZO_HEADER + data[thr].cmp_len;
844 off += PAGE_SIZE) {
845 memcpy(page, data[thr].cmp + off, PAGE_SIZE);
847 ret = swap_write_page(handle, page, &hb);
848 if (ret)
849 goto out_finish;
853 wait_event(crc->done, atomic_read(&crc->stop));
854 atomic_set(&crc->stop, 0);
857 out_finish:
858 err2 = hib_wait_io(&hb);
859 stop = ktime_get();
860 if (!ret)
861 ret = err2;
862 if (!ret)
863 pr_info("Image saving done\n");
864 swsusp_show_speed(start, stop, nr_to_write, "Wrote");
865 out_clean:
866 hib_finish_batch(&hb);
867 if (crc) {
868 if (crc->thr)
869 kthread_stop(crc->thr);
870 kfree(crc);
872 if (data) {
873 for (thr = 0; thr < nr_threads; thr++)
874 if (data[thr].thr)
875 kthread_stop(data[thr].thr);
876 vfree(data);
878 if (page) free_page((unsigned long)page);
880 return ret;
884 * enough_swap - Make sure we have enough swap to save the image.
886 * Returns TRUE or FALSE after checking the total amount of swap
887 * space avaiable from the resume partition.
890 static int enough_swap(unsigned int nr_pages)
892 unsigned int free_swap = count_swap_pages(root_swap, 1);
893 unsigned int required;
895 pr_debug("Free swap pages: %u\n", free_swap);
897 required = PAGES_FOR_IO + nr_pages;
898 return free_swap > required;
902 * swsusp_write - Write entire image and metadata.
903 * @flags: flags to pass to the "boot" kernel in the image header
905 * It is important _NOT_ to umount filesystems at this point. We want
906 * them synced (in case something goes wrong) but we DO not want to mark
907 * filesystem clean: it is not. (And it does not matter, if we resume
908 * correctly, we'll mark system clean, anyway.)
911 int swsusp_write(unsigned int flags)
913 struct swap_map_handle handle;
914 struct snapshot_handle snapshot;
915 struct swsusp_info *header;
916 unsigned long pages;
917 int error;
919 pages = snapshot_get_image_size();
920 error = get_swap_writer(&handle);
921 if (error) {
922 pr_err("Cannot get swap writer\n");
923 return error;
925 if (flags & SF_NOCOMPRESS_MODE) {
926 if (!enough_swap(pages)) {
927 pr_err("Not enough free swap\n");
928 error = -ENOSPC;
929 goto out_finish;
932 memset(&snapshot, 0, sizeof(struct snapshot_handle));
933 error = snapshot_read_next(&snapshot);
934 if (error < (int)PAGE_SIZE) {
935 if (error >= 0)
936 error = -EFAULT;
938 goto out_finish;
940 header = (struct swsusp_info *)data_of(snapshot);
941 error = swap_write_page(&handle, header, NULL);
942 if (!error) {
943 error = (flags & SF_NOCOMPRESS_MODE) ?
944 save_image(&handle, &snapshot, pages - 1) :
945 save_image_lzo(&handle, &snapshot, pages - 1);
947 out_finish:
948 error = swap_writer_finish(&handle, flags, error);
949 return error;
953 * The following functions allow us to read data using a swap map
954 * in a file-alike way
957 static void release_swap_reader(struct swap_map_handle *handle)
959 struct swap_map_page_list *tmp;
961 while (handle->maps) {
962 if (handle->maps->map)
963 free_page((unsigned long)handle->maps->map);
964 tmp = handle->maps;
965 handle->maps = handle->maps->next;
966 kfree(tmp);
968 handle->cur = NULL;
971 static int get_swap_reader(struct swap_map_handle *handle,
972 unsigned int *flags_p)
974 int error;
975 struct swap_map_page_list *tmp, *last;
976 sector_t offset;
978 *flags_p = swsusp_header->flags;
980 if (!swsusp_header->image) /* how can this happen? */
981 return -EINVAL;
983 handle->cur = NULL;
984 last = handle->maps = NULL;
985 offset = swsusp_header->image;
986 while (offset) {
987 tmp = kzalloc(sizeof(*handle->maps), GFP_KERNEL);
988 if (!tmp) {
989 release_swap_reader(handle);
990 return -ENOMEM;
992 if (!handle->maps)
993 handle->maps = tmp;
994 if (last)
995 last->next = tmp;
996 last = tmp;
998 tmp->map = (struct swap_map_page *)
999 __get_free_page(GFP_NOIO | __GFP_HIGH);
1000 if (!tmp->map) {
1001 release_swap_reader(handle);
1002 return -ENOMEM;
1005 error = hib_submit_io(REQ_OP_READ, 0, offset, tmp->map, NULL);
1006 if (error) {
1007 release_swap_reader(handle);
1008 return error;
1010 offset = tmp->map->next_swap;
1012 handle->k = 0;
1013 handle->cur = handle->maps->map;
1014 return 0;
1017 static int swap_read_page(struct swap_map_handle *handle, void *buf,
1018 struct hib_bio_batch *hb)
1020 sector_t offset;
1021 int error;
1022 struct swap_map_page_list *tmp;
1024 if (!handle->cur)
1025 return -EINVAL;
1026 offset = handle->cur->entries[handle->k];
1027 if (!offset)
1028 return -EFAULT;
1029 error = hib_submit_io(REQ_OP_READ, 0, offset, buf, hb);
1030 if (error)
1031 return error;
1032 if (++handle->k >= MAP_PAGE_ENTRIES) {
1033 handle->k = 0;
1034 free_page((unsigned long)handle->maps->map);
1035 tmp = handle->maps;
1036 handle->maps = handle->maps->next;
1037 kfree(tmp);
1038 if (!handle->maps)
1039 release_swap_reader(handle);
1040 else
1041 handle->cur = handle->maps->map;
1043 return error;
1046 static int swap_reader_finish(struct swap_map_handle *handle)
1048 release_swap_reader(handle);
1050 return 0;
1054 * load_image - load the image using the swap map handle
1055 * @handle and the snapshot handle @snapshot
1056 * (assume there are @nr_pages pages to load)
1059 static int load_image(struct swap_map_handle *handle,
1060 struct snapshot_handle *snapshot,
1061 unsigned int nr_to_read)
1063 unsigned int m;
1064 int ret = 0;
1065 ktime_t start;
1066 ktime_t stop;
1067 struct hib_bio_batch hb;
1068 int err2;
1069 unsigned nr_pages;
1071 hib_init_batch(&hb);
1073 clean_pages_on_read = true;
1074 pr_info("Loading image data pages (%u pages)...\n", nr_to_read);
1075 m = nr_to_read / 10;
1076 if (!m)
1077 m = 1;
1078 nr_pages = 0;
1079 start = ktime_get();
1080 for ( ; ; ) {
1081 ret = snapshot_write_next(snapshot);
1082 if (ret <= 0)
1083 break;
1084 ret = swap_read_page(handle, data_of(*snapshot), &hb);
1085 if (ret)
1086 break;
1087 if (snapshot->sync_read)
1088 ret = hib_wait_io(&hb);
1089 if (ret)
1090 break;
1091 if (!(nr_pages % m))
1092 pr_info("Image loading progress: %3d%%\n",
1093 nr_pages / m * 10);
1094 nr_pages++;
1096 err2 = hib_wait_io(&hb);
1097 hib_finish_batch(&hb);
1098 stop = ktime_get();
1099 if (!ret)
1100 ret = err2;
1101 if (!ret) {
1102 pr_info("Image loading done\n");
1103 snapshot_write_finalize(snapshot);
1104 if (!snapshot_image_loaded(snapshot))
1105 ret = -ENODATA;
1107 swsusp_show_speed(start, stop, nr_to_read, "Read");
1108 return ret;
1112 * Structure used for LZO data decompression.
1114 struct dec_data {
1115 struct task_struct *thr; /* thread */
1116 atomic_t ready; /* ready to start flag */
1117 atomic_t stop; /* ready to stop flag */
1118 int ret; /* return code */
1119 wait_queue_head_t go; /* start decompression */
1120 wait_queue_head_t done; /* decompression done */
1121 size_t unc_len; /* uncompressed length */
1122 size_t cmp_len; /* compressed length */
1123 unsigned char unc[LZO_UNC_SIZE]; /* uncompressed buffer */
1124 unsigned char cmp[LZO_CMP_SIZE]; /* compressed buffer */
1128 * Deompression function that runs in its own thread.
1130 static int lzo_decompress_threadfn(void *data)
1132 struct dec_data *d = data;
1134 while (1) {
1135 wait_event(d->go, atomic_read(&d->ready) ||
1136 kthread_should_stop());
1137 if (kthread_should_stop()) {
1138 d->thr = NULL;
1139 d->ret = -1;
1140 atomic_set(&d->stop, 1);
1141 wake_up(&d->done);
1142 break;
1144 atomic_set(&d->ready, 0);
1146 d->unc_len = LZO_UNC_SIZE;
1147 d->ret = lzo1x_decompress_safe(d->cmp + LZO_HEADER, d->cmp_len,
1148 d->unc, &d->unc_len);
1149 if (clean_pages_on_decompress)
1150 flush_icache_range((unsigned long)d->unc,
1151 (unsigned long)d->unc + d->unc_len);
1153 atomic_set(&d->stop, 1);
1154 wake_up(&d->done);
1156 return 0;
1160 * load_image_lzo - Load compressed image data and decompress them with LZO.
1161 * @handle: Swap map handle to use for loading data.
1162 * @snapshot: Image to copy uncompressed data into.
1163 * @nr_to_read: Number of pages to load.
1165 static int load_image_lzo(struct swap_map_handle *handle,
1166 struct snapshot_handle *snapshot,
1167 unsigned int nr_to_read)
1169 unsigned int m;
1170 int ret = 0;
1171 int eof = 0;
1172 struct hib_bio_batch hb;
1173 ktime_t start;
1174 ktime_t stop;
1175 unsigned nr_pages;
1176 size_t off;
1177 unsigned i, thr, run_threads, nr_threads;
1178 unsigned ring = 0, pg = 0, ring_size = 0,
1179 have = 0, want, need, asked = 0;
1180 unsigned long read_pages = 0;
1181 unsigned char **page = NULL;
1182 struct dec_data *data = NULL;
1183 struct crc_data *crc = NULL;
1185 hib_init_batch(&hb);
1188 * We'll limit the number of threads for decompression to limit memory
1189 * footprint.
1191 nr_threads = num_online_cpus() - 1;
1192 nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
1194 page = vmalloc(array_size(LZO_MAX_RD_PAGES, sizeof(*page)));
1195 if (!page) {
1196 pr_err("Failed to allocate LZO page\n");
1197 ret = -ENOMEM;
1198 goto out_clean;
1201 data = vmalloc(array_size(nr_threads, sizeof(*data)));
1202 if (!data) {
1203 pr_err("Failed to allocate LZO data\n");
1204 ret = -ENOMEM;
1205 goto out_clean;
1207 for (thr = 0; thr < nr_threads; thr++)
1208 memset(&data[thr], 0, offsetof(struct dec_data, go));
1210 crc = kmalloc(sizeof(*crc), GFP_KERNEL);
1211 if (!crc) {
1212 pr_err("Failed to allocate crc\n");
1213 ret = -ENOMEM;
1214 goto out_clean;
1216 memset(crc, 0, offsetof(struct crc_data, go));
1218 clean_pages_on_decompress = true;
1221 * Start the decompression threads.
1223 for (thr = 0; thr < nr_threads; thr++) {
1224 init_waitqueue_head(&data[thr].go);
1225 init_waitqueue_head(&data[thr].done);
1227 data[thr].thr = kthread_run(lzo_decompress_threadfn,
1228 &data[thr],
1229 "image_decompress/%u", thr);
1230 if (IS_ERR(data[thr].thr)) {
1231 data[thr].thr = NULL;
1232 pr_err("Cannot start decompression threads\n");
1233 ret = -ENOMEM;
1234 goto out_clean;
1239 * Start the CRC32 thread.
1241 init_waitqueue_head(&crc->go);
1242 init_waitqueue_head(&crc->done);
1244 handle->crc32 = 0;
1245 crc->crc32 = &handle->crc32;
1246 for (thr = 0; thr < nr_threads; thr++) {
1247 crc->unc[thr] = data[thr].unc;
1248 crc->unc_len[thr] = &data[thr].unc_len;
1251 crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
1252 if (IS_ERR(crc->thr)) {
1253 crc->thr = NULL;
1254 pr_err("Cannot start CRC32 thread\n");
1255 ret = -ENOMEM;
1256 goto out_clean;
1260 * Set the number of pages for read buffering.
1261 * This is complete guesswork, because we'll only know the real
1262 * picture once prepare_image() is called, which is much later on
1263 * during the image load phase. We'll assume the worst case and
1264 * say that none of the image pages are from high memory.
1266 if (low_free_pages() > snapshot_get_image_size())
1267 read_pages = (low_free_pages() - snapshot_get_image_size()) / 2;
1268 read_pages = clamp_val(read_pages, LZO_MIN_RD_PAGES, LZO_MAX_RD_PAGES);
1270 for (i = 0; i < read_pages; i++) {
1271 page[i] = (void *)__get_free_page(i < LZO_CMP_PAGES ?
1272 GFP_NOIO | __GFP_HIGH :
1273 GFP_NOIO | __GFP_NOWARN |
1274 __GFP_NORETRY);
1276 if (!page[i]) {
1277 if (i < LZO_CMP_PAGES) {
1278 ring_size = i;
1279 pr_err("Failed to allocate LZO pages\n");
1280 ret = -ENOMEM;
1281 goto out_clean;
1282 } else {
1283 break;
1287 want = ring_size = i;
1289 pr_info("Using %u thread(s) for decompression\n", nr_threads);
1290 pr_info("Loading and decompressing image data (%u pages)...\n",
1291 nr_to_read);
1292 m = nr_to_read / 10;
1293 if (!m)
1294 m = 1;
1295 nr_pages = 0;
1296 start = ktime_get();
1298 ret = snapshot_write_next(snapshot);
1299 if (ret <= 0)
1300 goto out_finish;
1302 for(;;) {
1303 for (i = 0; !eof && i < want; i++) {
1304 ret = swap_read_page(handle, page[ring], &hb);
1305 if (ret) {
1307 * On real read error, finish. On end of data,
1308 * set EOF flag and just exit the read loop.
1310 if (handle->cur &&
1311 handle->cur->entries[handle->k]) {
1312 goto out_finish;
1313 } else {
1314 eof = 1;
1315 break;
1318 if (++ring >= ring_size)
1319 ring = 0;
1321 asked += i;
1322 want -= i;
1325 * We are out of data, wait for some more.
1327 if (!have) {
1328 if (!asked)
1329 break;
1331 ret = hib_wait_io(&hb);
1332 if (ret)
1333 goto out_finish;
1334 have += asked;
1335 asked = 0;
1336 if (eof)
1337 eof = 2;
1340 if (crc->run_threads) {
1341 wait_event(crc->done, atomic_read(&crc->stop));
1342 atomic_set(&crc->stop, 0);
1343 crc->run_threads = 0;
1346 for (thr = 0; have && thr < nr_threads; thr++) {
1347 data[thr].cmp_len = *(size_t *)page[pg];
1348 if (unlikely(!data[thr].cmp_len ||
1349 data[thr].cmp_len >
1350 lzo1x_worst_compress(LZO_UNC_SIZE))) {
1351 pr_err("Invalid LZO compressed length\n");
1352 ret = -1;
1353 goto out_finish;
1356 need = DIV_ROUND_UP(data[thr].cmp_len + LZO_HEADER,
1357 PAGE_SIZE);
1358 if (need > have) {
1359 if (eof > 1) {
1360 ret = -1;
1361 goto out_finish;
1363 break;
1366 for (off = 0;
1367 off < LZO_HEADER + data[thr].cmp_len;
1368 off += PAGE_SIZE) {
1369 memcpy(data[thr].cmp + off,
1370 page[pg], PAGE_SIZE);
1371 have--;
1372 want++;
1373 if (++pg >= ring_size)
1374 pg = 0;
1377 atomic_set(&data[thr].ready, 1);
1378 wake_up(&data[thr].go);
1382 * Wait for more data while we are decompressing.
1384 if (have < LZO_CMP_PAGES && asked) {
1385 ret = hib_wait_io(&hb);
1386 if (ret)
1387 goto out_finish;
1388 have += asked;
1389 asked = 0;
1390 if (eof)
1391 eof = 2;
1394 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
1395 wait_event(data[thr].done,
1396 atomic_read(&data[thr].stop));
1397 atomic_set(&data[thr].stop, 0);
1399 ret = data[thr].ret;
1401 if (ret < 0) {
1402 pr_err("LZO decompression failed\n");
1403 goto out_finish;
1406 if (unlikely(!data[thr].unc_len ||
1407 data[thr].unc_len > LZO_UNC_SIZE ||
1408 data[thr].unc_len & (PAGE_SIZE - 1))) {
1409 pr_err("Invalid LZO uncompressed length\n");
1410 ret = -1;
1411 goto out_finish;
1414 for (off = 0;
1415 off < data[thr].unc_len; off += PAGE_SIZE) {
1416 memcpy(data_of(*snapshot),
1417 data[thr].unc + off, PAGE_SIZE);
1419 if (!(nr_pages % m))
1420 pr_info("Image loading progress: %3d%%\n",
1421 nr_pages / m * 10);
1422 nr_pages++;
1424 ret = snapshot_write_next(snapshot);
1425 if (ret <= 0) {
1426 crc->run_threads = thr + 1;
1427 atomic_set(&crc->ready, 1);
1428 wake_up(&crc->go);
1429 goto out_finish;
1434 crc->run_threads = thr;
1435 atomic_set(&crc->ready, 1);
1436 wake_up(&crc->go);
1439 out_finish:
1440 if (crc->run_threads) {
1441 wait_event(crc->done, atomic_read(&crc->stop));
1442 atomic_set(&crc->stop, 0);
1444 stop = ktime_get();
1445 if (!ret) {
1446 pr_info("Image loading done\n");
1447 snapshot_write_finalize(snapshot);
1448 if (!snapshot_image_loaded(snapshot))
1449 ret = -ENODATA;
1450 if (!ret) {
1451 if (swsusp_header->flags & SF_CRC32_MODE) {
1452 if(handle->crc32 != swsusp_header->crc32) {
1453 pr_err("Invalid image CRC32!\n");
1454 ret = -ENODATA;
1459 swsusp_show_speed(start, stop, nr_to_read, "Read");
1460 out_clean:
1461 hib_finish_batch(&hb);
1462 for (i = 0; i < ring_size; i++)
1463 free_page((unsigned long)page[i]);
1464 if (crc) {
1465 if (crc->thr)
1466 kthread_stop(crc->thr);
1467 kfree(crc);
1469 if (data) {
1470 for (thr = 0; thr < nr_threads; thr++)
1471 if (data[thr].thr)
1472 kthread_stop(data[thr].thr);
1473 vfree(data);
1475 vfree(page);
1477 return ret;
1481 * swsusp_read - read the hibernation image.
1482 * @flags_p: flags passed by the "frozen" kernel in the image header should
1483 * be written into this memory location
1486 int swsusp_read(unsigned int *flags_p)
1488 int error;
1489 struct swap_map_handle handle;
1490 struct snapshot_handle snapshot;
1491 struct swsusp_info *header;
1493 memset(&snapshot, 0, sizeof(struct snapshot_handle));
1494 error = snapshot_write_next(&snapshot);
1495 if (error < (int)PAGE_SIZE)
1496 return error < 0 ? error : -EFAULT;
1497 header = (struct swsusp_info *)data_of(snapshot);
1498 error = get_swap_reader(&handle, flags_p);
1499 if (error)
1500 goto end;
1501 if (!error)
1502 error = swap_read_page(&handle, header, NULL);
1503 if (!error) {
1504 error = (*flags_p & SF_NOCOMPRESS_MODE) ?
1505 load_image(&handle, &snapshot, header->pages - 1) :
1506 load_image_lzo(&handle, &snapshot, header->pages - 1);
1508 swap_reader_finish(&handle);
1509 end:
1510 if (!error)
1511 pr_debug("Image successfully loaded\n");
1512 else
1513 pr_debug("Error %d resuming\n", error);
1514 return error;
1518 * swsusp_check - Check for swsusp signature in the resume device
1521 int swsusp_check(void)
1523 int error;
1525 hib_resume_bdev = blkdev_get_by_dev(swsusp_resume_device,
1526 FMODE_READ, NULL);
1527 if (!IS_ERR(hib_resume_bdev)) {
1528 set_blocksize(hib_resume_bdev, PAGE_SIZE);
1529 clear_page(swsusp_header);
1530 error = hib_submit_io(REQ_OP_READ, 0,
1531 swsusp_resume_block,
1532 swsusp_header, NULL);
1533 if (error)
1534 goto put;
1536 if (!memcmp(HIBERNATE_SIG, swsusp_header->sig, 10)) {
1537 memcpy(swsusp_header->sig, swsusp_header->orig_sig, 10);
1538 /* Reset swap signature now */
1539 error = hib_submit_io(REQ_OP_WRITE, REQ_SYNC,
1540 swsusp_resume_block,
1541 swsusp_header, NULL);
1542 } else {
1543 error = -EINVAL;
1546 put:
1547 if (error)
1548 blkdev_put(hib_resume_bdev, FMODE_READ);
1549 else
1550 pr_debug("Image signature found, resuming\n");
1551 } else {
1552 error = PTR_ERR(hib_resume_bdev);
1555 if (error)
1556 pr_debug("Image not found (code %d)\n", error);
1558 return error;
1562 * swsusp_close - close swap device.
1565 void swsusp_close(fmode_t mode)
1567 if (IS_ERR(hib_resume_bdev)) {
1568 pr_debug("Image device not initialised\n");
1569 return;
1572 blkdev_put(hib_resume_bdev, mode);
1576 * swsusp_unmark - Unmark swsusp signature in the resume device
1579 #ifdef CONFIG_SUSPEND
1580 int swsusp_unmark(void)
1582 int error;
1584 hib_submit_io(REQ_OP_READ, 0, swsusp_resume_block,
1585 swsusp_header, NULL);
1586 if (!memcmp(HIBERNATE_SIG,swsusp_header->sig, 10)) {
1587 memcpy(swsusp_header->sig,swsusp_header->orig_sig, 10);
1588 error = hib_submit_io(REQ_OP_WRITE, REQ_SYNC,
1589 swsusp_resume_block,
1590 swsusp_header, NULL);
1591 } else {
1592 pr_err("Cannot find swsusp signature!\n");
1593 error = -ENODEV;
1597 * We just returned from suspend, we don't need the image any more.
1599 free_all_swap_pages(root_swap);
1601 return error;
1603 #endif
1605 static int __init swsusp_header_init(void)
1607 swsusp_header = (struct swsusp_header*) __get_free_page(GFP_KERNEL);
1608 if (!swsusp_header)
1609 panic("Could not allocate memory for swsusp_header\n");
1610 return 0;
1613 core_initcall(swsusp_header_init);