Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[sfrench/cifs-2.6.git] / kernel / power / swap.c
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@suse.cz>
8  * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
9  *
10  * This file is released under the GPLv2.
11  *
12  */
13
14 #include <linux/module.h>
15 #include <linux/file.h>
16 #include <linux/utsname.h>
17 #include <linux/delay.h>
18 #include <linux/bitops.h>
19 #include <linux/genhd.h>
20 #include <linux/device.h>
21 #include <linux/buffer_head.h>
22 #include <linux/bio.h>
23 #include <linux/blkdev.h>
24 #include <linux/swap.h>
25 #include <linux/swapops.h>
26 #include <linux/pm.h>
27
28 #include "power.h"
29
30 #define SWSUSP_SIG      "S1SUSPEND"
31
32 struct swsusp_header {
33         char reserved[PAGE_SIZE - 20 - sizeof(sector_t) - sizeof(int)];
34         sector_t image;
35         unsigned int flags;     /* Flags to pass to the "boot" kernel */
36         char    orig_sig[10];
37         char    sig[10];
38 } __attribute__((packed));
39
40 static struct swsusp_header *swsusp_header;
41
42 /*
43  * General things
44  */
45
46 static unsigned short root_swap = 0xffff;
47 static struct block_device *resume_bdev;
48
49 /**
50  *      submit - submit BIO request.
51  *      @rw:    READ or WRITE.
52  *      @off    physical offset of page.
53  *      @page:  page we're reading or writing.
54  *      @bio_chain: list of pending biod (for async reading)
55  *
56  *      Straight from the textbook - allocate and initialize the bio.
57  *      If we're reading, make sure the page is marked as dirty.
58  *      Then submit it and, if @bio_chain == NULL, wait.
59  */
60 static int submit(int rw, pgoff_t page_off, struct page *page,
61                         struct bio **bio_chain)
62 {
63         const int bio_rw = rw | (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
64         struct bio *bio;
65
66         bio = bio_alloc(__GFP_WAIT | __GFP_HIGH, 1);
67         if (!bio)
68                 return -ENOMEM;
69         bio->bi_sector = page_off * (PAGE_SIZE >> 9);
70         bio->bi_bdev = resume_bdev;
71         bio->bi_end_io = end_swap_bio_read;
72
73         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
74                 printk(KERN_ERR "PM: Adding page to bio failed at %ld\n",
75                         page_off);
76                 bio_put(bio);
77                 return -EFAULT;
78         }
79
80         lock_page(page);
81         bio_get(bio);
82
83         if (bio_chain == NULL) {
84                 submit_bio(bio_rw, bio);
85                 wait_on_page_locked(page);
86                 if (rw == READ)
87                         bio_set_pages_dirty(bio);
88                 bio_put(bio);
89         } else {
90                 if (rw == READ)
91                         get_page(page); /* These pages are freed later */
92                 bio->bi_private = *bio_chain;
93                 *bio_chain = bio;
94                 submit_bio(bio_rw, bio);
95         }
96         return 0;
97 }
98
99 static int bio_read_page(pgoff_t page_off, void *addr, struct bio **bio_chain)
100 {
101         return submit(READ, page_off, virt_to_page(addr), bio_chain);
102 }
103
104 static int bio_write_page(pgoff_t page_off, void *addr, struct bio **bio_chain)
105 {
106         return submit(WRITE, page_off, virt_to_page(addr), bio_chain);
107 }
108
109 static int wait_on_bio_chain(struct bio **bio_chain)
110 {
111         struct bio *bio;
112         struct bio *next_bio;
113         int ret = 0;
114
115         if (bio_chain == NULL)
116                 return 0;
117
118         bio = *bio_chain;
119         if (bio == NULL)
120                 return 0;
121         while (bio) {
122                 struct page *page;
123
124                 next_bio = bio->bi_private;
125                 page = bio->bi_io_vec[0].bv_page;
126                 wait_on_page_locked(page);
127                 if (!PageUptodate(page) || PageError(page))
128                         ret = -EIO;
129                 put_page(page);
130                 bio_put(bio);
131                 bio = next_bio;
132         }
133         *bio_chain = NULL;
134         return ret;
135 }
136
137 /*
138  * Saving part
139  */
140
141 static int mark_swapfiles(sector_t start, unsigned int flags)
142 {
143         int error;
144
145         bio_read_page(swsusp_resume_block, swsusp_header, NULL);
146         if (!memcmp("SWAP-SPACE",swsusp_header->sig, 10) ||
147             !memcmp("SWAPSPACE2",swsusp_header->sig, 10)) {
148                 memcpy(swsusp_header->orig_sig,swsusp_header->sig, 10);
149                 memcpy(swsusp_header->sig,SWSUSP_SIG, 10);
150                 swsusp_header->image = start;
151                 swsusp_header->flags = flags;
152                 error = bio_write_page(swsusp_resume_block,
153                                         swsusp_header, NULL);
154         } else {
155                 printk(KERN_ERR "PM: Swap header not found!\n");
156                 error = -ENODEV;
157         }
158         return error;
159 }
160
161 /**
162  *      swsusp_swap_check - check if the resume device is a swap device
163  *      and get its index (if so)
164  */
165
166 static int swsusp_swap_check(void) /* This is called before saving image */
167 {
168         int res;
169
170         res = swap_type_of(swsusp_resume_device, swsusp_resume_block,
171                         &resume_bdev);
172         if (res < 0)
173                 return res;
174
175         root_swap = res;
176         res = blkdev_get(resume_bdev, FMODE_WRITE);
177         if (res)
178                 return res;
179
180         res = set_blocksize(resume_bdev, PAGE_SIZE);
181         if (res < 0)
182                 blkdev_put(resume_bdev, FMODE_WRITE);
183
184         return res;
185 }
186
187 /**
188  *      write_page - Write one page to given swap location.
189  *      @buf:           Address we're writing.
190  *      @offset:        Offset of the swap page we're writing to.
191  *      @bio_chain:     Link the next write BIO here
192  */
193
194 static int write_page(void *buf, sector_t offset, struct bio **bio_chain)
195 {
196         void *src;
197
198         if (!offset)
199                 return -ENOSPC;
200
201         if (bio_chain) {
202                 src = (void *)__get_free_page(__GFP_WAIT | __GFP_HIGH);
203                 if (src) {
204                         memcpy(src, buf, PAGE_SIZE);
205                 } else {
206                         WARN_ON_ONCE(1);
207                         bio_chain = NULL;       /* Go synchronous */
208                         src = buf;
209                 }
210         } else {
211                 src = buf;
212         }
213         return bio_write_page(offset, src, bio_chain);
214 }
215
216 /*
217  *      The swap map is a data structure used for keeping track of each page
218  *      written to a swap partition.  It consists of many swap_map_page
219  *      structures that contain each an array of MAP_PAGE_SIZE swap entries.
220  *      These structures are stored on the swap and linked together with the
221  *      help of the .next_swap member.
222  *
223  *      The swap map is created during suspend.  The swap map pages are
224  *      allocated and populated one at a time, so we only need one memory
225  *      page to set up the entire structure.
226  *
227  *      During resume we also only need to use one swap_map_page structure
228  *      at a time.
229  */
230
231 #define MAP_PAGE_ENTRIES        (PAGE_SIZE / sizeof(sector_t) - 1)
232
233 struct swap_map_page {
234         sector_t entries[MAP_PAGE_ENTRIES];
235         sector_t next_swap;
236 };
237
238 /**
239  *      The swap_map_handle structure is used for handling swap in
240  *      a file-alike way
241  */
242
243 struct swap_map_handle {
244         struct swap_map_page *cur;
245         sector_t cur_swap;
246         unsigned int k;
247 };
248
249 static void release_swap_writer(struct swap_map_handle *handle)
250 {
251         if (handle->cur)
252                 free_page((unsigned long)handle->cur);
253         handle->cur = NULL;
254 }
255
256 static int get_swap_writer(struct swap_map_handle *handle)
257 {
258         handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_KERNEL);
259         if (!handle->cur)
260                 return -ENOMEM;
261         handle->cur_swap = alloc_swapdev_block(root_swap);
262         if (!handle->cur_swap) {
263                 release_swap_writer(handle);
264                 return -ENOSPC;
265         }
266         handle->k = 0;
267         return 0;
268 }
269
270 static int swap_write_page(struct swap_map_handle *handle, void *buf,
271                                 struct bio **bio_chain)
272 {
273         int error = 0;
274         sector_t offset;
275
276         if (!handle->cur)
277                 return -EINVAL;
278         offset = alloc_swapdev_block(root_swap);
279         error = write_page(buf, offset, bio_chain);
280         if (error)
281                 return error;
282         handle->cur->entries[handle->k++] = offset;
283         if (handle->k >= MAP_PAGE_ENTRIES) {
284                 error = wait_on_bio_chain(bio_chain);
285                 if (error)
286                         goto out;
287                 offset = alloc_swapdev_block(root_swap);
288                 if (!offset)
289                         return -ENOSPC;
290                 handle->cur->next_swap = offset;
291                 error = write_page(handle->cur, handle->cur_swap, NULL);
292                 if (error)
293                         goto out;
294                 memset(handle->cur, 0, PAGE_SIZE);
295                 handle->cur_swap = offset;
296                 handle->k = 0;
297         }
298  out:
299         return error;
300 }
301
302 static int flush_swap_writer(struct swap_map_handle *handle)
303 {
304         if (handle->cur && handle->cur_swap)
305                 return write_page(handle->cur, handle->cur_swap, NULL);
306         else
307                 return -EINVAL;
308 }
309
310 /**
311  *      save_image - save the suspend image data
312  */
313
314 static int save_image(struct swap_map_handle *handle,
315                       struct snapshot_handle *snapshot,
316                       unsigned int nr_to_write)
317 {
318         unsigned int m;
319         int ret;
320         int error = 0;
321         int nr_pages;
322         int err2;
323         struct bio *bio;
324         struct timeval start;
325         struct timeval stop;
326
327         printk(KERN_INFO "PM: Saving image data pages (%u pages) ...     ",
328                 nr_to_write);
329         m = nr_to_write / 100;
330         if (!m)
331                 m = 1;
332         nr_pages = 0;
333         bio = NULL;
334         do_gettimeofday(&start);
335         do {
336                 ret = snapshot_read_next(snapshot, PAGE_SIZE);
337                 if (ret > 0) {
338                         error = swap_write_page(handle, data_of(*snapshot),
339                                                 &bio);
340                         if (error)
341                                 break;
342                         if (!(nr_pages % m))
343                                 printk("\b\b\b\b%3d%%", nr_pages / m);
344                         nr_pages++;
345                 }
346         } while (ret > 0);
347         err2 = wait_on_bio_chain(&bio);
348         do_gettimeofday(&stop);
349         if (!error)
350                 error = err2;
351         if (!error)
352                 printk("\b\b\b\bdone\n");
353         swsusp_show_speed(&start, &stop, nr_to_write, "Wrote");
354         return error;
355 }
356
357 /**
358  *      enough_swap - Make sure we have enough swap to save the image.
359  *
360  *      Returns TRUE or FALSE after checking the total amount of swap
361  *      space avaiable from the resume partition.
362  */
363
364 static int enough_swap(unsigned int nr_pages)
365 {
366         unsigned int free_swap = count_swap_pages(root_swap, 1);
367
368         pr_debug("PM: Free swap pages: %u\n", free_swap);
369         return free_swap > nr_pages + PAGES_FOR_IO;
370 }
371
372 /**
373  *      swsusp_write - Write entire image and metadata.
374  *      @flags: flags to pass to the "boot" kernel in the image header
375  *
376  *      It is important _NOT_ to umount filesystems at this point. We want
377  *      them synced (in case something goes wrong) but we DO not want to mark
378  *      filesystem clean: it is not. (And it does not matter, if we resume
379  *      correctly, we'll mark system clean, anyway.)
380  */
381
382 int swsusp_write(unsigned int flags)
383 {
384         struct swap_map_handle handle;
385         struct snapshot_handle snapshot;
386         struct swsusp_info *header;
387         int error;
388
389         error = swsusp_swap_check();
390         if (error) {
391                 printk(KERN_ERR "PM: Cannot find swap device, try "
392                                 "swapon -a.\n");
393                 return error;
394         }
395         memset(&snapshot, 0, sizeof(struct snapshot_handle));
396         error = snapshot_read_next(&snapshot, PAGE_SIZE);
397         if (error < PAGE_SIZE) {
398                 if (error >= 0)
399                         error = -EFAULT;
400
401                 goto out;
402         }
403         header = (struct swsusp_info *)data_of(snapshot);
404         if (!enough_swap(header->pages)) {
405                 printk(KERN_ERR "PM: Not enough free swap\n");
406                 error = -ENOSPC;
407                 goto out;
408         }
409         error = get_swap_writer(&handle);
410         if (!error) {
411                 sector_t start = handle.cur_swap;
412
413                 error = swap_write_page(&handle, header, NULL);
414                 if (!error)
415                         error = save_image(&handle, &snapshot,
416                                         header->pages - 1);
417
418                 if (!error) {
419                         flush_swap_writer(&handle);
420                         printk(KERN_INFO "PM: S");
421                         error = mark_swapfiles(start, flags);
422                         printk("|\n");
423                 }
424         }
425         if (error)
426                 free_all_swap_pages(root_swap);
427
428         release_swap_writer(&handle);
429  out:
430         swsusp_close(FMODE_WRITE);
431         return error;
432 }
433
434 /**
435  *      The following functions allow us to read data using a swap map
436  *      in a file-alike way
437  */
438
439 static void release_swap_reader(struct swap_map_handle *handle)
440 {
441         if (handle->cur)
442                 free_page((unsigned long)handle->cur);
443         handle->cur = NULL;
444 }
445
446 static int get_swap_reader(struct swap_map_handle *handle, sector_t start)
447 {
448         int error;
449
450         if (!start)
451                 return -EINVAL;
452
453         handle->cur = (struct swap_map_page *)get_zeroed_page(__GFP_WAIT | __GFP_HIGH);
454         if (!handle->cur)
455                 return -ENOMEM;
456
457         error = bio_read_page(start, handle->cur, NULL);
458         if (error) {
459                 release_swap_reader(handle);
460                 return error;
461         }
462         handle->k = 0;
463         return 0;
464 }
465
466 static int swap_read_page(struct swap_map_handle *handle, void *buf,
467                                 struct bio **bio_chain)
468 {
469         sector_t offset;
470         int error;
471
472         if (!handle->cur)
473                 return -EINVAL;
474         offset = handle->cur->entries[handle->k];
475         if (!offset)
476                 return -EFAULT;
477         error = bio_read_page(offset, buf, bio_chain);
478         if (error)
479                 return error;
480         if (++handle->k >= MAP_PAGE_ENTRIES) {
481                 error = wait_on_bio_chain(bio_chain);
482                 handle->k = 0;
483                 offset = handle->cur->next_swap;
484                 if (!offset)
485                         release_swap_reader(handle);
486                 else if (!error)
487                         error = bio_read_page(offset, handle->cur, NULL);
488         }
489         return error;
490 }
491
492 /**
493  *      load_image - load the image using the swap map handle
494  *      @handle and the snapshot handle @snapshot
495  *      (assume there are @nr_pages pages to load)
496  */
497
498 static int load_image(struct swap_map_handle *handle,
499                       struct snapshot_handle *snapshot,
500                       unsigned int nr_to_read)
501 {
502         unsigned int m;
503         int error = 0;
504         struct timeval start;
505         struct timeval stop;
506         struct bio *bio;
507         int err2;
508         unsigned nr_pages;
509
510         printk(KERN_INFO "PM: Loading image data pages (%u pages) ...     ",
511                 nr_to_read);
512         m = nr_to_read / 100;
513         if (!m)
514                 m = 1;
515         nr_pages = 0;
516         bio = NULL;
517         do_gettimeofday(&start);
518         for ( ; ; ) {
519                 error = snapshot_write_next(snapshot, PAGE_SIZE);
520                 if (error <= 0)
521                         break;
522                 error = swap_read_page(handle, data_of(*snapshot), &bio);
523                 if (error)
524                         break;
525                 if (snapshot->sync_read)
526                         error = wait_on_bio_chain(&bio);
527                 if (error)
528                         break;
529                 if (!(nr_pages % m))
530                         printk("\b\b\b\b%3d%%", nr_pages / m);
531                 nr_pages++;
532         }
533         err2 = wait_on_bio_chain(&bio);
534         do_gettimeofday(&stop);
535         if (!error)
536                 error = err2;
537         if (!error) {
538                 printk("\b\b\b\bdone\n");
539                 snapshot_write_finalize(snapshot);
540                 if (!snapshot_image_loaded(snapshot))
541                         error = -ENODATA;
542         }
543         swsusp_show_speed(&start, &stop, nr_to_read, "Read");
544         return error;
545 }
546
547 /**
548  *      swsusp_read - read the hibernation image.
549  *      @flags_p: flags passed by the "frozen" kernel in the image header should
550  *                be written into this memeory location
551  */
552
553 int swsusp_read(unsigned int *flags_p)
554 {
555         int error;
556         struct swap_map_handle handle;
557         struct snapshot_handle snapshot;
558         struct swsusp_info *header;
559
560         *flags_p = swsusp_header->flags;
561         if (IS_ERR(resume_bdev)) {
562                 pr_debug("PM: Image device not initialised\n");
563                 return PTR_ERR(resume_bdev);
564         }
565
566         memset(&snapshot, 0, sizeof(struct snapshot_handle));
567         error = snapshot_write_next(&snapshot, PAGE_SIZE);
568         if (error < PAGE_SIZE)
569                 return error < 0 ? error : -EFAULT;
570         header = (struct swsusp_info *)data_of(snapshot);
571         error = get_swap_reader(&handle, swsusp_header->image);
572         if (!error)
573                 error = swap_read_page(&handle, header, NULL);
574         if (!error)
575                 error = load_image(&handle, &snapshot, header->pages - 1);
576         release_swap_reader(&handle);
577
578         blkdev_put(resume_bdev, FMODE_READ);
579
580         if (!error)
581                 pr_debug("PM: Image successfully loaded\n");
582         else
583                 pr_debug("PM: Error %d resuming\n", error);
584         return error;
585 }
586
587 /**
588  *      swsusp_check - Check for swsusp signature in the resume device
589  */
590
591 int swsusp_check(void)
592 {
593         int error;
594
595         resume_bdev = open_by_devnum(swsusp_resume_device, FMODE_READ);
596         if (!IS_ERR(resume_bdev)) {
597                 set_blocksize(resume_bdev, PAGE_SIZE);
598                 memset(swsusp_header, 0, PAGE_SIZE);
599                 error = bio_read_page(swsusp_resume_block,
600                                         swsusp_header, NULL);
601                 if (error)
602                         return error;
603
604                 if (!memcmp(SWSUSP_SIG, swsusp_header->sig, 10)) {
605                         memcpy(swsusp_header->sig, swsusp_header->orig_sig, 10);
606                         /* Reset swap signature now */
607                         error = bio_write_page(swsusp_resume_block,
608                                                 swsusp_header, NULL);
609                 } else {
610                         return -EINVAL;
611                 }
612                 if (error)
613                         blkdev_put(resume_bdev, FMODE_READ);
614                 else
615                         pr_debug("PM: Signature found, resuming\n");
616         } else {
617                 error = PTR_ERR(resume_bdev);
618         }
619
620         if (error)
621                 pr_debug("PM: Error %d checking image file\n", error);
622
623         return error;
624 }
625
626 /**
627  *      swsusp_close - close swap device.
628  */
629
630 void swsusp_close(fmode_t mode)
631 {
632         if (IS_ERR(resume_bdev)) {
633                 pr_debug("PM: Image device not initialised\n");
634                 return;
635         }
636
637         blkdev_put(resume_bdev, mode);
638 }
639
640 static int swsusp_header_init(void)
641 {
642         swsusp_header = (struct swsusp_header*) __get_free_page(GFP_KERNEL);
643         if (!swsusp_header)
644                 panic("Could not allocate memory for swsusp_header\n");
645         return 0;
646 }
647
648 core_initcall(swsusp_header_init);