Merge tag 'x86-cleanups-2024-03-11' of git://git.kernel.org/pub/scm/linux/kernel...
[sfrench/cifs-2.6.git] / kernel / power / swap.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/kernel/power/swap.c
4  *
5  * This file provides functions for reading the suspend image from
6  * and writing it to a swap partition.
7  *
8  * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@ucw.cz>
9  * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
10  * Copyright (C) 2010-2012 Bojan Smojver <bojan@rexursive.com>
11  */
12
13 #define pr_fmt(fmt) "PM: " fmt
14
15 #include <linux/module.h>
16 #include <linux/file.h>
17 #include <linux/delay.h>
18 #include <linux/bitops.h>
19 #include <linux/device.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/swap.h>
23 #include <linux/swapops.h>
24 #include <linux/pm.h>
25 #include <linux/slab.h>
26 #include <linux/lzo.h>
27 #include <linux/vmalloc.h>
28 #include <linux/cpumask.h>
29 #include <linux/atomic.h>
30 #include <linux/kthread.h>
31 #include <linux/crc32.h>
32 #include <linux/ktime.h>
33
34 #include "power.h"
35
36 #define HIBERNATE_SIG   "S1SUSPEND"
37
38 u32 swsusp_hardware_signature;
39
40 /*
41  * When reading an {un,}compressed image, we may restore pages in place,
42  * in which case some architectures need these pages cleaning before they
43  * can be executed. We don't know which pages these may be, so clean the lot.
44  */
45 static bool clean_pages_on_read;
46 static bool clean_pages_on_decompress;
47
48 /*
49  *      The swap map is a data structure used for keeping track of each page
50  *      written to a swap partition.  It consists of many swap_map_page
51  *      structures that contain each an array of MAP_PAGE_ENTRIES swap entries.
52  *      These structures are stored on the swap and linked together with the
53  *      help of the .next_swap member.
54  *
55  *      The swap map is created during suspend.  The swap map pages are
56  *      allocated and populated one at a time, so we only need one memory
57  *      page to set up the entire structure.
58  *
59  *      During resume we pick up all swap_map_page structures into a list.
60  */
61
62 #define MAP_PAGE_ENTRIES        (PAGE_SIZE / sizeof(sector_t) - 1)
63
64 /*
65  * Number of free pages that are not high.
66  */
67 static inline unsigned long low_free_pages(void)
68 {
69         return nr_free_pages() - nr_free_highpages();
70 }
71
72 /*
73  * Number of pages required to be kept free while writing the image. Always
74  * half of all available low pages before the writing starts.
75  */
76 static inline unsigned long reqd_free_pages(void)
77 {
78         return low_free_pages() / 2;
79 }
80
81 struct swap_map_page {
82         sector_t entries[MAP_PAGE_ENTRIES];
83         sector_t next_swap;
84 };
85
86 struct swap_map_page_list {
87         struct swap_map_page *map;
88         struct swap_map_page_list *next;
89 };
90
91 /*
92  *      The swap_map_handle structure is used for handling swap in
93  *      a file-alike way
94  */
95
96 struct swap_map_handle {
97         struct swap_map_page *cur;
98         struct swap_map_page_list *maps;
99         sector_t cur_swap;
100         sector_t first_sector;
101         unsigned int k;
102         unsigned long reqd_free_pages;
103         u32 crc32;
104 };
105
106 struct swsusp_header {
107         char reserved[PAGE_SIZE - 20 - sizeof(sector_t) - sizeof(int) -
108                       sizeof(u32) - sizeof(u32)];
109         u32     hw_sig;
110         u32     crc32;
111         sector_t image;
112         unsigned int flags;     /* Flags to pass to the "boot" kernel */
113         char    orig_sig[10];
114         char    sig[10];
115 } __packed;
116
117 static struct swsusp_header *swsusp_header;
118
119 /*
120  *      The following functions are used for tracing the allocated
121  *      swap pages, so that they can be freed in case of an error.
122  */
123
124 struct swsusp_extent {
125         struct rb_node node;
126         unsigned long start;
127         unsigned long end;
128 };
129
130 static struct rb_root swsusp_extents = RB_ROOT;
131
132 static int swsusp_extents_insert(unsigned long swap_offset)
133 {
134         struct rb_node **new = &(swsusp_extents.rb_node);
135         struct rb_node *parent = NULL;
136         struct swsusp_extent *ext;
137
138         /* Figure out where to put the new node */
139         while (*new) {
140                 ext = rb_entry(*new, struct swsusp_extent, node);
141                 parent = *new;
142                 if (swap_offset < ext->start) {
143                         /* Try to merge */
144                         if (swap_offset == ext->start - 1) {
145                                 ext->start--;
146                                 return 0;
147                         }
148                         new = &((*new)->rb_left);
149                 } else if (swap_offset > ext->end) {
150                         /* Try to merge */
151                         if (swap_offset == ext->end + 1) {
152                                 ext->end++;
153                                 return 0;
154                         }
155                         new = &((*new)->rb_right);
156                 } else {
157                         /* It already is in the tree */
158                         return -EINVAL;
159                 }
160         }
161         /* Add the new node and rebalance the tree. */
162         ext = kzalloc(sizeof(struct swsusp_extent), GFP_KERNEL);
163         if (!ext)
164                 return -ENOMEM;
165
166         ext->start = swap_offset;
167         ext->end = swap_offset;
168         rb_link_node(&ext->node, parent, new);
169         rb_insert_color(&ext->node, &swsusp_extents);
170         return 0;
171 }
172
173 /*
174  *      alloc_swapdev_block - allocate a swap page and register that it has
175  *      been allocated, so that it can be freed in case of an error.
176  */
177
178 sector_t alloc_swapdev_block(int swap)
179 {
180         unsigned long offset;
181
182         offset = swp_offset(get_swap_page_of_type(swap));
183         if (offset) {
184                 if (swsusp_extents_insert(offset))
185                         swap_free(swp_entry(swap, offset));
186                 else
187                         return swapdev_block(swap, offset);
188         }
189         return 0;
190 }
191
192 /*
193  *      free_all_swap_pages - free swap pages allocated for saving image data.
194  *      It also frees the extents used to register which swap entries had been
195  *      allocated.
196  */
197
198 void free_all_swap_pages(int swap)
199 {
200         struct rb_node *node;
201
202         while ((node = swsusp_extents.rb_node)) {
203                 struct swsusp_extent *ext;
204                 unsigned long offset;
205
206                 ext = rb_entry(node, struct swsusp_extent, node);
207                 rb_erase(node, &swsusp_extents);
208                 for (offset = ext->start; offset <= ext->end; offset++)
209                         swap_free(swp_entry(swap, offset));
210
211                 kfree(ext);
212         }
213 }
214
215 int swsusp_swap_in_use(void)
216 {
217         return (swsusp_extents.rb_node != NULL);
218 }
219
220 /*
221  * General things
222  */
223
224 static unsigned short root_swap = 0xffff;
225 static struct file *hib_resume_bdev_file;
226
227 struct hib_bio_batch {
228         atomic_t                count;
229         wait_queue_head_t       wait;
230         blk_status_t            error;
231         struct blk_plug         plug;
232 };
233
234 static void hib_init_batch(struct hib_bio_batch *hb)
235 {
236         atomic_set(&hb->count, 0);
237         init_waitqueue_head(&hb->wait);
238         hb->error = BLK_STS_OK;
239         blk_start_plug(&hb->plug);
240 }
241
242 static void hib_finish_batch(struct hib_bio_batch *hb)
243 {
244         blk_finish_plug(&hb->plug);
245 }
246
247 static void hib_end_io(struct bio *bio)
248 {
249         struct hib_bio_batch *hb = bio->bi_private;
250         struct page *page = bio_first_page_all(bio);
251
252         if (bio->bi_status) {
253                 pr_alert("Read-error on swap-device (%u:%u:%Lu)\n",
254                          MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
255                          (unsigned long long)bio->bi_iter.bi_sector);
256         }
257
258         if (bio_data_dir(bio) == WRITE)
259                 put_page(page);
260         else if (clean_pages_on_read)
261                 flush_icache_range((unsigned long)page_address(page),
262                                    (unsigned long)page_address(page) + PAGE_SIZE);
263
264         if (bio->bi_status && !hb->error)
265                 hb->error = bio->bi_status;
266         if (atomic_dec_and_test(&hb->count))
267                 wake_up(&hb->wait);
268
269         bio_put(bio);
270 }
271
272 static int hib_submit_io(blk_opf_t opf, pgoff_t page_off, void *addr,
273                          struct hib_bio_batch *hb)
274 {
275         struct page *page = virt_to_page(addr);
276         struct bio *bio;
277         int error = 0;
278
279         bio = bio_alloc(file_bdev(hib_resume_bdev_file), 1, opf,
280                         GFP_NOIO | __GFP_HIGH);
281         bio->bi_iter.bi_sector = page_off * (PAGE_SIZE >> 9);
282
283         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
284                 pr_err("Adding page to bio failed at %llu\n",
285                        (unsigned long long)bio->bi_iter.bi_sector);
286                 bio_put(bio);
287                 return -EFAULT;
288         }
289
290         if (hb) {
291                 bio->bi_end_io = hib_end_io;
292                 bio->bi_private = hb;
293                 atomic_inc(&hb->count);
294                 submit_bio(bio);
295         } else {
296                 error = submit_bio_wait(bio);
297                 bio_put(bio);
298         }
299
300         return error;
301 }
302
303 static int hib_wait_io(struct hib_bio_batch *hb)
304 {
305         /*
306          * We are relying on the behavior of blk_plug that a thread with
307          * a plug will flush the plug list before sleeping.
308          */
309         wait_event(hb->wait, atomic_read(&hb->count) == 0);
310         return blk_status_to_errno(hb->error);
311 }
312
313 /*
314  * Saving part
315  */
316 static int mark_swapfiles(struct swap_map_handle *handle, unsigned int flags)
317 {
318         int error;
319
320         hib_submit_io(REQ_OP_READ, swsusp_resume_block, swsusp_header, NULL);
321         if (!memcmp("SWAP-SPACE",swsusp_header->sig, 10) ||
322             !memcmp("SWAPSPACE2",swsusp_header->sig, 10)) {
323                 memcpy(swsusp_header->orig_sig,swsusp_header->sig, 10);
324                 memcpy(swsusp_header->sig, HIBERNATE_SIG, 10);
325                 swsusp_header->image = handle->first_sector;
326                 if (swsusp_hardware_signature) {
327                         swsusp_header->hw_sig = swsusp_hardware_signature;
328                         flags |= SF_HW_SIG;
329                 }
330                 swsusp_header->flags = flags;
331                 if (flags & SF_CRC32_MODE)
332                         swsusp_header->crc32 = handle->crc32;
333                 error = hib_submit_io(REQ_OP_WRITE | REQ_SYNC,
334                                       swsusp_resume_block, swsusp_header, NULL);
335         } else {
336                 pr_err("Swap header not found!\n");
337                 error = -ENODEV;
338         }
339         return error;
340 }
341
342 /**
343  *      swsusp_swap_check - check if the resume device is a swap device
344  *      and get its index (if so)
345  *
346  *      This is called before saving image
347  */
348 static int swsusp_swap_check(void)
349 {
350         int res;
351
352         if (swsusp_resume_device)
353                 res = swap_type_of(swsusp_resume_device, swsusp_resume_block);
354         else
355                 res = find_first_swap(&swsusp_resume_device);
356         if (res < 0)
357                 return res;
358         root_swap = res;
359
360         hib_resume_bdev_file = bdev_file_open_by_dev(swsusp_resume_device,
361                         BLK_OPEN_WRITE, NULL, NULL);
362         if (IS_ERR(hib_resume_bdev_file))
363                 return PTR_ERR(hib_resume_bdev_file);
364
365         res = set_blocksize(file_bdev(hib_resume_bdev_file), PAGE_SIZE);
366         if (res < 0)
367                 fput(hib_resume_bdev_file);
368
369         return res;
370 }
371
372 /**
373  *      write_page - Write one page to given swap location.
374  *      @buf:           Address we're writing.
375  *      @offset:        Offset of the swap page we're writing to.
376  *      @hb:            bio completion batch
377  */
378
379 static int write_page(void *buf, sector_t offset, struct hib_bio_batch *hb)
380 {
381         void *src;
382         int ret;
383
384         if (!offset)
385                 return -ENOSPC;
386
387         if (hb) {
388                 src = (void *)__get_free_page(GFP_NOIO | __GFP_NOWARN |
389                                               __GFP_NORETRY);
390                 if (src) {
391                         copy_page(src, buf);
392                 } else {
393                         ret = hib_wait_io(hb); /* Free pages */
394                         if (ret)
395                                 return ret;
396                         src = (void *)__get_free_page(GFP_NOIO |
397                                                       __GFP_NOWARN |
398                                                       __GFP_NORETRY);
399                         if (src) {
400                                 copy_page(src, buf);
401                         } else {
402                                 WARN_ON_ONCE(1);
403                                 hb = NULL;      /* Go synchronous */
404                                 src = buf;
405                         }
406                 }
407         } else {
408                 src = buf;
409         }
410         return hib_submit_io(REQ_OP_WRITE | REQ_SYNC, offset, src, hb);
411 }
412
413 static void release_swap_writer(struct swap_map_handle *handle)
414 {
415         if (handle->cur)
416                 free_page((unsigned long)handle->cur);
417         handle->cur = NULL;
418 }
419
420 static int get_swap_writer(struct swap_map_handle *handle)
421 {
422         int ret;
423
424         ret = swsusp_swap_check();
425         if (ret) {
426                 if (ret != -ENOSPC)
427                         pr_err("Cannot find swap device, try swapon -a\n");
428                 return ret;
429         }
430         handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_KERNEL);
431         if (!handle->cur) {
432                 ret = -ENOMEM;
433                 goto err_close;
434         }
435         handle->cur_swap = alloc_swapdev_block(root_swap);
436         if (!handle->cur_swap) {
437                 ret = -ENOSPC;
438                 goto err_rel;
439         }
440         handle->k = 0;
441         handle->reqd_free_pages = reqd_free_pages();
442         handle->first_sector = handle->cur_swap;
443         return 0;
444 err_rel:
445         release_swap_writer(handle);
446 err_close:
447         swsusp_close();
448         return ret;
449 }
450
451 static int swap_write_page(struct swap_map_handle *handle, void *buf,
452                 struct hib_bio_batch *hb)
453 {
454         int error;
455         sector_t offset;
456
457         if (!handle->cur)
458                 return -EINVAL;
459         offset = alloc_swapdev_block(root_swap);
460         error = write_page(buf, offset, hb);
461         if (error)
462                 return error;
463         handle->cur->entries[handle->k++] = offset;
464         if (handle->k >= MAP_PAGE_ENTRIES) {
465                 offset = alloc_swapdev_block(root_swap);
466                 if (!offset)
467                         return -ENOSPC;
468                 handle->cur->next_swap = offset;
469                 error = write_page(handle->cur, handle->cur_swap, hb);
470                 if (error)
471                         goto out;
472                 clear_page(handle->cur);
473                 handle->cur_swap = offset;
474                 handle->k = 0;
475
476                 if (hb && low_free_pages() <= handle->reqd_free_pages) {
477                         error = hib_wait_io(hb);
478                         if (error)
479                                 goto out;
480                         /*
481                          * Recalculate the number of required free pages, to
482                          * make sure we never take more than half.
483                          */
484                         handle->reqd_free_pages = reqd_free_pages();
485                 }
486         }
487  out:
488         return error;
489 }
490
491 static int flush_swap_writer(struct swap_map_handle *handle)
492 {
493         if (handle->cur && handle->cur_swap)
494                 return write_page(handle->cur, handle->cur_swap, NULL);
495         else
496                 return -EINVAL;
497 }
498
499 static int swap_writer_finish(struct swap_map_handle *handle,
500                 unsigned int flags, int error)
501 {
502         if (!error) {
503                 pr_info("S");
504                 error = mark_swapfiles(handle, flags);
505                 pr_cont("|\n");
506                 flush_swap_writer(handle);
507         }
508
509         if (error)
510                 free_all_swap_pages(root_swap);
511         release_swap_writer(handle);
512         swsusp_close();
513
514         return error;
515 }
516
517 /* We need to remember how much compressed data we need to read. */
518 #define LZO_HEADER      sizeof(size_t)
519
520 /* Number of pages/bytes we'll compress at one time. */
521 #define LZO_UNC_PAGES   32
522 #define LZO_UNC_SIZE    (LZO_UNC_PAGES * PAGE_SIZE)
523
524 /* Number of pages/bytes we need for compressed data (worst case). */
525 #define LZO_CMP_PAGES   DIV_ROUND_UP(lzo1x_worst_compress(LZO_UNC_SIZE) + \
526                                      LZO_HEADER, PAGE_SIZE)
527 #define LZO_CMP_SIZE    (LZO_CMP_PAGES * PAGE_SIZE)
528
529 /* Maximum number of threads for compression/decompression. */
530 #define LZO_THREADS     3
531
532 /* Minimum/maximum number of pages for read buffering. */
533 #define LZO_MIN_RD_PAGES        1024
534 #define LZO_MAX_RD_PAGES        8192
535
536
537 /**
538  *      save_image - save the suspend image data
539  */
540
541 static int save_image(struct swap_map_handle *handle,
542                       struct snapshot_handle *snapshot,
543                       unsigned int nr_to_write)
544 {
545         unsigned int m;
546         int ret;
547         int nr_pages;
548         int err2;
549         struct hib_bio_batch hb;
550         ktime_t start;
551         ktime_t stop;
552
553         hib_init_batch(&hb);
554
555         pr_info("Saving image data pages (%u pages)...\n",
556                 nr_to_write);
557         m = nr_to_write / 10;
558         if (!m)
559                 m = 1;
560         nr_pages = 0;
561         start = ktime_get();
562         while (1) {
563                 ret = snapshot_read_next(snapshot);
564                 if (ret <= 0)
565                         break;
566                 ret = swap_write_page(handle, data_of(*snapshot), &hb);
567                 if (ret)
568                         break;
569                 if (!(nr_pages % m))
570                         pr_info("Image saving progress: %3d%%\n",
571                                 nr_pages / m * 10);
572                 nr_pages++;
573         }
574         err2 = hib_wait_io(&hb);
575         hib_finish_batch(&hb);
576         stop = ktime_get();
577         if (!ret)
578                 ret = err2;
579         if (!ret)
580                 pr_info("Image saving done\n");
581         swsusp_show_speed(start, stop, nr_to_write, "Wrote");
582         return ret;
583 }
584
585 /*
586  * Structure used for CRC32.
587  */
588 struct crc_data {
589         struct task_struct *thr;                  /* thread */
590         atomic_t ready;                           /* ready to start flag */
591         atomic_t stop;                            /* ready to stop flag */
592         unsigned run_threads;                     /* nr current threads */
593         wait_queue_head_t go;                     /* start crc update */
594         wait_queue_head_t done;                   /* crc update done */
595         u32 *crc32;                               /* points to handle's crc32 */
596         size_t *unc_len[LZO_THREADS];             /* uncompressed lengths */
597         unsigned char *unc[LZO_THREADS];          /* uncompressed data */
598 };
599
600 /*
601  * CRC32 update function that runs in its own thread.
602  */
603 static int crc32_threadfn(void *data)
604 {
605         struct crc_data *d = data;
606         unsigned i;
607
608         while (1) {
609                 wait_event(d->go, atomic_read_acquire(&d->ready) ||
610                                   kthread_should_stop());
611                 if (kthread_should_stop()) {
612                         d->thr = NULL;
613                         atomic_set_release(&d->stop, 1);
614                         wake_up(&d->done);
615                         break;
616                 }
617                 atomic_set(&d->ready, 0);
618
619                 for (i = 0; i < d->run_threads; i++)
620                         *d->crc32 = crc32_le(*d->crc32,
621                                              d->unc[i], *d->unc_len[i]);
622                 atomic_set_release(&d->stop, 1);
623                 wake_up(&d->done);
624         }
625         return 0;
626 }
627 /*
628  * Structure used for LZO data compression.
629  */
630 struct cmp_data {
631         struct task_struct *thr;                  /* thread */
632         atomic_t ready;                           /* ready to start flag */
633         atomic_t stop;                            /* ready to stop flag */
634         int ret;                                  /* return code */
635         wait_queue_head_t go;                     /* start compression */
636         wait_queue_head_t done;                   /* compression done */
637         size_t unc_len;                           /* uncompressed length */
638         size_t cmp_len;                           /* compressed length */
639         unsigned char unc[LZO_UNC_SIZE];          /* uncompressed buffer */
640         unsigned char cmp[LZO_CMP_SIZE];          /* compressed buffer */
641         unsigned char wrk[LZO1X_1_MEM_COMPRESS];  /* compression workspace */
642 };
643
644 /*
645  * Compression function that runs in its own thread.
646  */
647 static int lzo_compress_threadfn(void *data)
648 {
649         struct cmp_data *d = data;
650
651         while (1) {
652                 wait_event(d->go, atomic_read_acquire(&d->ready) ||
653                                   kthread_should_stop());
654                 if (kthread_should_stop()) {
655                         d->thr = NULL;
656                         d->ret = -1;
657                         atomic_set_release(&d->stop, 1);
658                         wake_up(&d->done);
659                         break;
660                 }
661                 atomic_set(&d->ready, 0);
662
663                 d->ret = lzo1x_1_compress(d->unc, d->unc_len,
664                                           d->cmp + LZO_HEADER, &d->cmp_len,
665                                           d->wrk);
666                 atomic_set_release(&d->stop, 1);
667                 wake_up(&d->done);
668         }
669         return 0;
670 }
671
672 /**
673  * save_image_lzo - Save the suspend image data compressed with LZO.
674  * @handle: Swap map handle to use for saving the image.
675  * @snapshot: Image to read data from.
676  * @nr_to_write: Number of pages to save.
677  */
678 static int save_image_lzo(struct swap_map_handle *handle,
679                           struct snapshot_handle *snapshot,
680                           unsigned int nr_to_write)
681 {
682         unsigned int m;
683         int ret = 0;
684         int nr_pages;
685         int err2;
686         struct hib_bio_batch hb;
687         ktime_t start;
688         ktime_t stop;
689         size_t off;
690         unsigned thr, run_threads, nr_threads;
691         unsigned char *page = NULL;
692         struct cmp_data *data = NULL;
693         struct crc_data *crc = NULL;
694
695         hib_init_batch(&hb);
696
697         /*
698          * We'll limit the number of threads for compression to limit memory
699          * footprint.
700          */
701         nr_threads = num_online_cpus() - 1;
702         nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
703
704         page = (void *)__get_free_page(GFP_NOIO | __GFP_HIGH);
705         if (!page) {
706                 pr_err("Failed to allocate LZO page\n");
707                 ret = -ENOMEM;
708                 goto out_clean;
709         }
710
711         data = vzalloc(array_size(nr_threads, sizeof(*data)));
712         if (!data) {
713                 pr_err("Failed to allocate LZO data\n");
714                 ret = -ENOMEM;
715                 goto out_clean;
716         }
717
718         crc = kzalloc(sizeof(*crc), GFP_KERNEL);
719         if (!crc) {
720                 pr_err("Failed to allocate crc\n");
721                 ret = -ENOMEM;
722                 goto out_clean;
723         }
724
725         /*
726          * Start the compression threads.
727          */
728         for (thr = 0; thr < nr_threads; thr++) {
729                 init_waitqueue_head(&data[thr].go);
730                 init_waitqueue_head(&data[thr].done);
731
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;
740                 }
741         }
742
743         /*
744          * Start the CRC32 thread.
745          */
746         init_waitqueue_head(&crc->go);
747         init_waitqueue_head(&crc->done);
748
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;
754         }
755
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;
762         }
763
764         /*
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.
767          */
768         handle->reqd_free_pages = reqd_free_pages();
769
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;
784
785                                 if (!ret)
786                                         break;
787
788                                 memcpy(data[thr].unc + off,
789                                        data_of(*snapshot), PAGE_SIZE);
790
791                                 if (!(nr_pages % m))
792                                         pr_info("Image saving progress: %3d%%\n",
793                                                 nr_pages / m * 10);
794                                 nr_pages++;
795                         }
796                         if (!off)
797                                 break;
798
799                         data[thr].unc_len = off;
800
801                         atomic_set_release(&data[thr].ready, 1);
802                         wake_up(&data[thr].go);
803                 }
804
805                 if (!thr)
806                         break;
807
808                 crc->run_threads = thr;
809                 atomic_set_release(&crc->ready, 1);
810                 wake_up(&crc->go);
811
812                 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
813                         wait_event(data[thr].done,
814                                 atomic_read_acquire(&data[thr].stop));
815                         atomic_set(&data[thr].stop, 0);
816
817                         ret = data[thr].ret;
818
819                         if (ret < 0) {
820                                 pr_err("LZO compression failed\n");
821                                 goto out_finish;
822                         }
823
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;
830                         }
831
832                         *(size_t *)data[thr].cmp = data[thr].cmp_len;
833
834                         /*
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.
841                          */
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);
846
847                                 ret = swap_write_page(handle, page, &hb);
848                                 if (ret)
849                                         goto out_finish;
850                         }
851                 }
852
853                 wait_event(crc->done, atomic_read_acquire(&crc->stop));
854                 atomic_set(&crc->stop, 0);
855         }
856
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);
871         }
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);
877         }
878         if (page) free_page((unsigned long)page);
879
880         return ret;
881 }
882
883 /**
884  *      enough_swap - Make sure we have enough swap to save the image.
885  *
886  *      Returns TRUE or FALSE after checking the total amount of swap
887  *      space available from the resume partition.
888  */
889
890 static int enough_swap(unsigned int nr_pages)
891 {
892         unsigned int free_swap = count_swap_pages(root_swap, 1);
893         unsigned int required;
894
895         pr_debug("Free swap pages: %u\n", free_swap);
896
897         required = PAGES_FOR_IO + nr_pages;
898         return free_swap > required;
899 }
900
901 /**
902  *      swsusp_write - Write entire image and metadata.
903  *      @flags: flags to pass to the "boot" kernel in the image header
904  *
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.)
909  */
910
911 int swsusp_write(unsigned int flags)
912 {
913         struct swap_map_handle handle;
914         struct snapshot_handle snapshot;
915         struct swsusp_info *header;
916         unsigned long pages;
917         int error;
918
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;
924         }
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;
930                 }
931         }
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;
937
938                 goto out_finish;
939         }
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);
946         }
947 out_finish:
948         error = swap_writer_finish(&handle, flags, error);
949         return error;
950 }
951
952 /*
953  *      The following functions allow us to read data using a swap map
954  *      in a file-like way.
955  */
956
957 static void release_swap_reader(struct swap_map_handle *handle)
958 {
959         struct swap_map_page_list *tmp;
960
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);
967         }
968         handle->cur = NULL;
969 }
970
971 static int get_swap_reader(struct swap_map_handle *handle,
972                 unsigned int *flags_p)
973 {
974         int error;
975         struct swap_map_page_list *tmp, *last;
976         sector_t offset;
977
978         *flags_p = swsusp_header->flags;
979
980         if (!swsusp_header->image) /* how can this happen? */
981                 return -EINVAL;
982
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;
991                 }
992                 if (!handle->maps)
993                         handle->maps = tmp;
994                 if (last)
995                         last->next = tmp;
996                 last = tmp;
997
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;
1003                 }
1004
1005                 error = hib_submit_io(REQ_OP_READ, offset, tmp->map, NULL);
1006                 if (error) {
1007                         release_swap_reader(handle);
1008                         return error;
1009                 }
1010                 offset = tmp->map->next_swap;
1011         }
1012         handle->k = 0;
1013         handle->cur = handle->maps->map;
1014         return 0;
1015 }
1016
1017 static int swap_read_page(struct swap_map_handle *handle, void *buf,
1018                 struct hib_bio_batch *hb)
1019 {
1020         sector_t offset;
1021         int error;
1022         struct swap_map_page_list *tmp;
1023
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, 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;
1042         }
1043         return error;
1044 }
1045
1046 static int swap_reader_finish(struct swap_map_handle *handle)
1047 {
1048         release_swap_reader(handle);
1049
1050         return 0;
1051 }
1052
1053 /**
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)
1057  */
1058
1059 static int load_image(struct swap_map_handle *handle,
1060                       struct snapshot_handle *snapshot,
1061                       unsigned int nr_to_read)
1062 {
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;
1070
1071         hib_init_batch(&hb);
1072
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++;
1095         }
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;
1106         }
1107         swsusp_show_speed(start, stop, nr_to_read, "Read");
1108         return ret;
1109 }
1110
1111 /*
1112  * Structure used for LZO data decompression.
1113  */
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 */
1125 };
1126
1127 /*
1128  * Decompression function that runs in its own thread.
1129  */
1130 static int lzo_decompress_threadfn(void *data)
1131 {
1132         struct dec_data *d = data;
1133
1134         while (1) {
1135                 wait_event(d->go, atomic_read_acquire(&d->ready) ||
1136                                   kthread_should_stop());
1137                 if (kthread_should_stop()) {
1138                         d->thr = NULL;
1139                         d->ret = -1;
1140                         atomic_set_release(&d->stop, 1);
1141                         wake_up(&d->done);
1142                         break;
1143                 }
1144                 atomic_set(&d->ready, 0);
1145
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);
1152
1153                 atomic_set_release(&d->stop, 1);
1154                 wake_up(&d->done);
1155         }
1156         return 0;
1157 }
1158
1159 /**
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.
1164  */
1165 static int load_image_lzo(struct swap_map_handle *handle,
1166                           struct snapshot_handle *snapshot,
1167                           unsigned int nr_to_read)
1168 {
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;
1184
1185         hib_init_batch(&hb);
1186
1187         /*
1188          * We'll limit the number of threads for decompression to limit memory
1189          * footprint.
1190          */
1191         nr_threads = num_online_cpus() - 1;
1192         nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
1193
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;
1199         }
1200
1201         data = vzalloc(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;
1206         }
1207
1208         crc = kzalloc(sizeof(*crc), GFP_KERNEL);
1209         if (!crc) {
1210                 pr_err("Failed to allocate crc\n");
1211                 ret = -ENOMEM;
1212                 goto out_clean;
1213         }
1214
1215         clean_pages_on_decompress = true;
1216
1217         /*
1218          * Start the decompression threads.
1219          */
1220         for (thr = 0; thr < nr_threads; thr++) {
1221                 init_waitqueue_head(&data[thr].go);
1222                 init_waitqueue_head(&data[thr].done);
1223
1224                 data[thr].thr = kthread_run(lzo_decompress_threadfn,
1225                                             &data[thr],
1226                                             "image_decompress/%u", thr);
1227                 if (IS_ERR(data[thr].thr)) {
1228                         data[thr].thr = NULL;
1229                         pr_err("Cannot start decompression threads\n");
1230                         ret = -ENOMEM;
1231                         goto out_clean;
1232                 }
1233         }
1234
1235         /*
1236          * Start the CRC32 thread.
1237          */
1238         init_waitqueue_head(&crc->go);
1239         init_waitqueue_head(&crc->done);
1240
1241         handle->crc32 = 0;
1242         crc->crc32 = &handle->crc32;
1243         for (thr = 0; thr < nr_threads; thr++) {
1244                 crc->unc[thr] = data[thr].unc;
1245                 crc->unc_len[thr] = &data[thr].unc_len;
1246         }
1247
1248         crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
1249         if (IS_ERR(crc->thr)) {
1250                 crc->thr = NULL;
1251                 pr_err("Cannot start CRC32 thread\n");
1252                 ret = -ENOMEM;
1253                 goto out_clean;
1254         }
1255
1256         /*
1257          * Set the number of pages for read buffering.
1258          * This is complete guesswork, because we'll only know the real
1259          * picture once prepare_image() is called, which is much later on
1260          * during the image load phase. We'll assume the worst case and
1261          * say that none of the image pages are from high memory.
1262          */
1263         if (low_free_pages() > snapshot_get_image_size())
1264                 read_pages = (low_free_pages() - snapshot_get_image_size()) / 2;
1265         read_pages = clamp_val(read_pages, LZO_MIN_RD_PAGES, LZO_MAX_RD_PAGES);
1266
1267         for (i = 0; i < read_pages; i++) {
1268                 page[i] = (void *)__get_free_page(i < LZO_CMP_PAGES ?
1269                                                   GFP_NOIO | __GFP_HIGH :
1270                                                   GFP_NOIO | __GFP_NOWARN |
1271                                                   __GFP_NORETRY);
1272
1273                 if (!page[i]) {
1274                         if (i < LZO_CMP_PAGES) {
1275                                 ring_size = i;
1276                                 pr_err("Failed to allocate LZO pages\n");
1277                                 ret = -ENOMEM;
1278                                 goto out_clean;
1279                         } else {
1280                                 break;
1281                         }
1282                 }
1283         }
1284         want = ring_size = i;
1285
1286         pr_info("Using %u thread(s) for decompression\n", nr_threads);
1287         pr_info("Loading and decompressing image data (%u pages)...\n",
1288                 nr_to_read);
1289         m = nr_to_read / 10;
1290         if (!m)
1291                 m = 1;
1292         nr_pages = 0;
1293         start = ktime_get();
1294
1295         ret = snapshot_write_next(snapshot);
1296         if (ret <= 0)
1297                 goto out_finish;
1298
1299         for(;;) {
1300                 for (i = 0; !eof && i < want; i++) {
1301                         ret = swap_read_page(handle, page[ring], &hb);
1302                         if (ret) {
1303                                 /*
1304                                  * On real read error, finish. On end of data,
1305                                  * set EOF flag and just exit the read loop.
1306                                  */
1307                                 if (handle->cur &&
1308                                     handle->cur->entries[handle->k]) {
1309                                         goto out_finish;
1310                                 } else {
1311                                         eof = 1;
1312                                         break;
1313                                 }
1314                         }
1315                         if (++ring >= ring_size)
1316                                 ring = 0;
1317                 }
1318                 asked += i;
1319                 want -= i;
1320
1321                 /*
1322                  * We are out of data, wait for some more.
1323                  */
1324                 if (!have) {
1325                         if (!asked)
1326                                 break;
1327
1328                         ret = hib_wait_io(&hb);
1329                         if (ret)
1330                                 goto out_finish;
1331                         have += asked;
1332                         asked = 0;
1333                         if (eof)
1334                                 eof = 2;
1335                 }
1336
1337                 if (crc->run_threads) {
1338                         wait_event(crc->done, atomic_read_acquire(&crc->stop));
1339                         atomic_set(&crc->stop, 0);
1340                         crc->run_threads = 0;
1341                 }
1342
1343                 for (thr = 0; have && thr < nr_threads; thr++) {
1344                         data[thr].cmp_len = *(size_t *)page[pg];
1345                         if (unlikely(!data[thr].cmp_len ||
1346                                      data[thr].cmp_len >
1347                                      lzo1x_worst_compress(LZO_UNC_SIZE))) {
1348                                 pr_err("Invalid LZO compressed length\n");
1349                                 ret = -1;
1350                                 goto out_finish;
1351                         }
1352
1353                         need = DIV_ROUND_UP(data[thr].cmp_len + LZO_HEADER,
1354                                             PAGE_SIZE);
1355                         if (need > have) {
1356                                 if (eof > 1) {
1357                                         ret = -1;
1358                                         goto out_finish;
1359                                 }
1360                                 break;
1361                         }
1362
1363                         for (off = 0;
1364                              off < LZO_HEADER + data[thr].cmp_len;
1365                              off += PAGE_SIZE) {
1366                                 memcpy(data[thr].cmp + off,
1367                                        page[pg], PAGE_SIZE);
1368                                 have--;
1369                                 want++;
1370                                 if (++pg >= ring_size)
1371                                         pg = 0;
1372                         }
1373
1374                         atomic_set_release(&data[thr].ready, 1);
1375                         wake_up(&data[thr].go);
1376                 }
1377
1378                 /*
1379                  * Wait for more data while we are decompressing.
1380                  */
1381                 if (have < LZO_CMP_PAGES && asked) {
1382                         ret = hib_wait_io(&hb);
1383                         if (ret)
1384                                 goto out_finish;
1385                         have += asked;
1386                         asked = 0;
1387                         if (eof)
1388                                 eof = 2;
1389                 }
1390
1391                 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
1392                         wait_event(data[thr].done,
1393                                 atomic_read_acquire(&data[thr].stop));
1394                         atomic_set(&data[thr].stop, 0);
1395
1396                         ret = data[thr].ret;
1397
1398                         if (ret < 0) {
1399                                 pr_err("LZO decompression failed\n");
1400                                 goto out_finish;
1401                         }
1402
1403                         if (unlikely(!data[thr].unc_len ||
1404                                      data[thr].unc_len > LZO_UNC_SIZE ||
1405                                      data[thr].unc_len & (PAGE_SIZE - 1))) {
1406                                 pr_err("Invalid LZO uncompressed length\n");
1407                                 ret = -1;
1408                                 goto out_finish;
1409                         }
1410
1411                         for (off = 0;
1412                              off < data[thr].unc_len; off += PAGE_SIZE) {
1413                                 memcpy(data_of(*snapshot),
1414                                        data[thr].unc + off, PAGE_SIZE);
1415
1416                                 if (!(nr_pages % m))
1417                                         pr_info("Image loading progress: %3d%%\n",
1418                                                 nr_pages / m * 10);
1419                                 nr_pages++;
1420
1421                                 ret = snapshot_write_next(snapshot);
1422                                 if (ret <= 0) {
1423                                         crc->run_threads = thr + 1;
1424                                         atomic_set_release(&crc->ready, 1);
1425                                         wake_up(&crc->go);
1426                                         goto out_finish;
1427                                 }
1428                         }
1429                 }
1430
1431                 crc->run_threads = thr;
1432                 atomic_set_release(&crc->ready, 1);
1433                 wake_up(&crc->go);
1434         }
1435
1436 out_finish:
1437         if (crc->run_threads) {
1438                 wait_event(crc->done, atomic_read_acquire(&crc->stop));
1439                 atomic_set(&crc->stop, 0);
1440         }
1441         stop = ktime_get();
1442         if (!ret) {
1443                 pr_info("Image loading done\n");
1444                 snapshot_write_finalize(snapshot);
1445                 if (!snapshot_image_loaded(snapshot))
1446                         ret = -ENODATA;
1447                 if (!ret) {
1448                         if (swsusp_header->flags & SF_CRC32_MODE) {
1449                                 if(handle->crc32 != swsusp_header->crc32) {
1450                                         pr_err("Invalid image CRC32!\n");
1451                                         ret = -ENODATA;
1452                                 }
1453                         }
1454                 }
1455         }
1456         swsusp_show_speed(start, stop, nr_to_read, "Read");
1457 out_clean:
1458         hib_finish_batch(&hb);
1459         for (i = 0; i < ring_size; i++)
1460                 free_page((unsigned long)page[i]);
1461         if (crc) {
1462                 if (crc->thr)
1463                         kthread_stop(crc->thr);
1464                 kfree(crc);
1465         }
1466         if (data) {
1467                 for (thr = 0; thr < nr_threads; thr++)
1468                         if (data[thr].thr)
1469                                 kthread_stop(data[thr].thr);
1470                 vfree(data);
1471         }
1472         vfree(page);
1473
1474         return ret;
1475 }
1476
1477 /**
1478  *      swsusp_read - read the hibernation image.
1479  *      @flags_p: flags passed by the "frozen" kernel in the image header should
1480  *                be written into this memory location
1481  */
1482
1483 int swsusp_read(unsigned int *flags_p)
1484 {
1485         int error;
1486         struct swap_map_handle handle;
1487         struct snapshot_handle snapshot;
1488         struct swsusp_info *header;
1489
1490         memset(&snapshot, 0, sizeof(struct snapshot_handle));
1491         error = snapshot_write_next(&snapshot);
1492         if (error < (int)PAGE_SIZE)
1493                 return error < 0 ? error : -EFAULT;
1494         header = (struct swsusp_info *)data_of(snapshot);
1495         error = get_swap_reader(&handle, flags_p);
1496         if (error)
1497                 goto end;
1498         if (!error)
1499                 error = swap_read_page(&handle, header, NULL);
1500         if (!error) {
1501                 error = (*flags_p & SF_NOCOMPRESS_MODE) ?
1502                         load_image(&handle, &snapshot, header->pages - 1) :
1503                         load_image_lzo(&handle, &snapshot, header->pages - 1);
1504         }
1505         swap_reader_finish(&handle);
1506 end:
1507         if (!error)
1508                 pr_debug("Image successfully loaded\n");
1509         else
1510                 pr_debug("Error %d resuming\n", error);
1511         return error;
1512 }
1513
1514 static void *swsusp_holder;
1515
1516 /**
1517  * swsusp_check - Open the resume device and check for the swsusp signature.
1518  * @exclusive: Open the resume device exclusively.
1519  */
1520
1521 int swsusp_check(bool exclusive)
1522 {
1523         void *holder = exclusive ? &swsusp_holder : NULL;
1524         int error;
1525
1526         hib_resume_bdev_file = bdev_file_open_by_dev(swsusp_resume_device,
1527                                 BLK_OPEN_READ, holder, NULL);
1528         if (!IS_ERR(hib_resume_bdev_file)) {
1529                 set_blocksize(file_bdev(hib_resume_bdev_file), PAGE_SIZE);
1530                 clear_page(swsusp_header);
1531                 error = hib_submit_io(REQ_OP_READ, swsusp_resume_block,
1532                                         swsusp_header, NULL);
1533                 if (error)
1534                         goto put;
1535
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;
1544                 }
1545                 if (!error && swsusp_header->flags & SF_HW_SIG &&
1546                     swsusp_header->hw_sig != swsusp_hardware_signature) {
1547                         pr_info("Suspend image hardware signature mismatch (%08x now %08x); aborting resume.\n",
1548                                 swsusp_header->hw_sig, swsusp_hardware_signature);
1549                         error = -EINVAL;
1550                 }
1551
1552 put:
1553                 if (error)
1554                         fput(hib_resume_bdev_file);
1555                 else
1556                         pr_debug("Image signature found, resuming\n");
1557         } else {
1558                 error = PTR_ERR(hib_resume_bdev_file);
1559         }
1560
1561         if (error)
1562                 pr_debug("Image not found (code %d)\n", error);
1563
1564         return error;
1565 }
1566
1567 /**
1568  * swsusp_close - close resume device.
1569  */
1570
1571 void swsusp_close(void)
1572 {
1573         if (IS_ERR(hib_resume_bdev_file)) {
1574                 pr_debug("Image device not initialised\n");
1575                 return;
1576         }
1577
1578         fput(hib_resume_bdev_file);
1579 }
1580
1581 /**
1582  *      swsusp_unmark - Unmark swsusp signature in the resume device
1583  */
1584
1585 #ifdef CONFIG_SUSPEND
1586 int swsusp_unmark(void)
1587 {
1588         int error;
1589
1590         hib_submit_io(REQ_OP_READ, swsusp_resume_block,
1591                         swsusp_header, NULL);
1592         if (!memcmp(HIBERNATE_SIG,swsusp_header->sig, 10)) {
1593                 memcpy(swsusp_header->sig,swsusp_header->orig_sig, 10);
1594                 error = hib_submit_io(REQ_OP_WRITE | REQ_SYNC,
1595                                         swsusp_resume_block,
1596                                         swsusp_header, NULL);
1597         } else {
1598                 pr_err("Cannot find swsusp signature!\n");
1599                 error = -ENODEV;
1600         }
1601
1602         /*
1603          * We just returned from suspend, we don't need the image any more.
1604          */
1605         free_all_swap_pages(root_swap);
1606
1607         return error;
1608 }
1609 #endif
1610
1611 static int __init swsusp_header_init(void)
1612 {
1613         swsusp_header = (struct swsusp_header*) __get_free_page(GFP_KERNEL);
1614         if (!swsusp_header)
1615                 panic("Could not allocate memory for swsusp_header\n");
1616         return 0;
1617 }
1618
1619 core_initcall(swsusp_header_init);