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