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@ucw.cz>
   8  * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
   9  * Copyright (C) 2010-2012 Bojan Smojver <bojan@rexursive.com>
  10  *
  11  * This file is released under the GPLv2.
  12  *
  13  */
  14
  15 #define pr_fmt(fmt) "PM: " fmt
  16
  17 #include <linux/module.h>
  18 #include <linux/file.h>
  19 #include <linux/delay.h>
  20 #include <linux/bitops.h>
  21 #include <linux/genhd.h>
  22 #include <linux/device.h>
  23 #include <linux/bio.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/swap.h>
  26 #include <linux/swapops.h>
  27 #include <linux/pm.h>
  28 #include <linux/slab.h>
  29 #include <linux/lzo.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/cpumask.h>
  32 #include <linux/atomic.h>
  33 #include <linux/kthread.h>
  34 #include <linux/crc32.h>
  35 #include <linux/ktime.h>
  36
  37 #include "power.h"
  38
  39 #define HIBERNATE_SIG   "S1SUSPEND"
  40
  41 /*
  42  * When reading an {un,}compressed image, we may restore pages in place,
  43  * in which case some architectures need these pages cleaning before they
  44  * can be executed. We don't know which pages these may be, so clean the lot.
  45  */
  46 static bool clean_pages_on_read;
  47 static bool clean_pages_on_decompress;
  48
  49 /*
  50  *      The swap map is a data structure used for keeping track of each page
  51  *      written to a swap partition.  It consists of many swap_map_page
  52  *      structures that contain each an array of MAP_PAGE_ENTRIES swap entries.
  53  *      These structures are stored on the swap and linked together with the
  54  *      help of the .next_swap member.
  55  *
  56  *      The swap map is created during suspend.  The swap map pages are
  57  *      allocated and populated one at a time, so we only need one memory
  58  *      page to set up the entire structure.
  59  *
  60  *      During resume we pick up all swap_map_page structures into a list.
  61  */
  62
  63 #define MAP_PAGE_ENTRIES        (PAGE_SIZE / sizeof(sector_t) - 1)
  64
  65 /*
  66  * Number of free pages that are not high.
  67  */
  68 static inline unsigned long low_free_pages(void)
  69 {
  70         return nr_free_pages() - nr_free_highpages();
  71 }
  72
  73 /*
  74  * Number of pages required to be kept free while writing the image. Always
  75  * half of all available low pages before the writing starts.
  76  */
  77 static inline unsigned long reqd_free_pages(void)
  78 {
  79         return low_free_pages() / 2;
  80 }
  81
  82 struct swap_map_page {
  83         sector_t entries[MAP_PAGE_ENTRIES];
  84         sector_t next_swap;
  85 };
  86
  87 struct swap_map_page_list {
  88         struct swap_map_page *map;
  89         struct swap_map_page_list *next;
  90 };
  91
  92 /**
  93  *      The swap_map_handle structure is used for handling swap in
  94  *      a file-alike way
  95  */
  96
  97 struct swap_map_handle {
  98         struct swap_map_page *cur;
  99         struct swap_map_page_list *maps;
 100         sector_t cur_swap;
 101         sector_t first_sector;
 102         unsigned int k;
 103         unsigned long reqd_free_pages;
 104         u32 crc32;
 105 };
 106
 107 struct swsusp_header {
 108         char reserved[PAGE_SIZE - 20 - sizeof(sector_t) - sizeof(int) -
 109                       sizeof(u32)];
 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 block_device *hib_resume_bdev;
 226
 227 struct hib_bio_batch {
 228         atomic_t                count;
 229         wait_queue_head_t       wait;
 230         blk_status_t            error;
 231 };
 232
 233 static void hib_init_batch(struct hib_bio_batch *hb)
 234 {
 235         atomic_set(&hb->count, 0);
 236         init_waitqueue_head(&hb->wait);
 237         hb->error = BLK_STS_OK;
 238 }
 239
 240 static void hib_end_io(struct bio *bio)
 241 {
 242         struct hib_bio_batch *hb = bio->bi_private;
 243         struct page *page = bio->bi_io_vec[0].bv_page;
 244
 245         if (bio->bi_status) {
 246                 pr_alert("Read-error on swap-device (%u:%u:%Lu)\n",
 247                          MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
 248                          (unsigned long long)bio->bi_iter.bi_sector);
 249         }
 250
 251         if (bio_data_dir(bio) == WRITE)
 252                 put_page(page);
 253         else if (clean_pages_on_read)
 254                 flush_icache_range((unsigned long)page_address(page),
 255                                    (unsigned long)page_address(page) + PAGE_SIZE);
 256
 257         if (bio->bi_status && !hb->error)
 258                 hb->error = bio->bi_status;
 259         if (atomic_dec_and_test(&hb->count))
 260                 wake_up(&hb->wait);
 261
 262         bio_put(bio);
 263 }
 264
 265 static int hib_submit_io(int op, int op_flags, pgoff_t page_off, void *addr,
 266                 struct hib_bio_batch *hb)
 267 {
 268         struct page *page = virt_to_page(addr);
 269         struct bio *bio;
 270         int error = 0;
 271
 272         bio = bio_alloc(__GFP_RECLAIM | __GFP_HIGH, 1);
 273         bio->bi_iter.bi_sector = page_off * (PAGE_SIZE >> 9);
 274         bio_set_dev(bio, hib_resume_bdev);
 275         bio_set_op_attrs(bio, op, op_flags);
 276
 277         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 278                 pr_err("Adding page to bio failed at %llu\n",
 279                        (unsigned long long)bio->bi_iter.bi_sector);
 280                 bio_put(bio);
 281                 return -EFAULT;
 282         }
 283
 284         if (hb) {
 285                 bio->bi_end_io = hib_end_io;
 286                 bio->bi_private = hb;
 287                 atomic_inc(&hb->count);
 288                 submit_bio(bio);
 289         } else {
 290                 error = submit_bio_wait(bio);
 291                 bio_put(bio);
 292         }
 293
 294         return error;
 295 }
 296
 297 static blk_status_t hib_wait_io(struct hib_bio_batch *hb)
 298 {
 299         wait_event(hb->wait, atomic_read(&hb->count) == 0);
 300         return blk_status_to_errno(hb->error);
 301 }
 302
 303 /*
 304  * Saving part
 305  */
 306
 307 static int mark_swapfiles(struct swap_map_handle *handle, unsigned int flags)
 308 {
 309         int error;
 310
 311         hib_submit_io(REQ_OP_READ, 0, swsusp_resume_block,
 312                       swsusp_header, NULL);
 313         if (!memcmp("SWAP-SPACE",swsusp_header->sig, 10) ||
 314             !memcmp("SWAPSPACE2",swsusp_header->sig, 10)) {
 315                 memcpy(swsusp_header->orig_sig,swsusp_header->sig, 10);
 316                 memcpy(swsusp_header->sig, HIBERNATE_SIG, 10);
 317                 swsusp_header->image = handle->first_sector;
 318                 swsusp_header->flags = flags;
 319                 if (flags & SF_CRC32_MODE)
 320                         swsusp_header->crc32 = handle->crc32;
 321                 error = hib_submit_io(REQ_OP_WRITE, REQ_SYNC,
 322                                       swsusp_resume_block, swsusp_header, NULL);
 323         } else {
 324                 pr_err("Swap header not found!\n");
 325                 error = -ENODEV;
 326         }
 327         return error;
 328 }
 329
 330 /**
 331  *      swsusp_swap_check - check if the resume device is a swap device
 332  *      and get its index (if so)
 333  *
 334  *      This is called before saving image
 335  */
 336 static int swsusp_swap_check(void)
 337 {
 338         int res;
 339
 340         res = swap_type_of(swsusp_resume_device, swsusp_resume_block,
 341                         &hib_resume_bdev);
 342         if (res < 0)
 343                 return res;
 344
 345         root_swap = res;
 346         res = blkdev_get(hib_resume_bdev, FMODE_WRITE, NULL);
 347         if (res)
 348                 return res;
 349
 350         res = set_blocksize(hib_resume_bdev, PAGE_SIZE);
 351         if (res < 0)
 352                 blkdev_put(hib_resume_bdev, FMODE_WRITE);
 353
 354         /*
 355          * Update the resume device to the one actually used,
 356          * so the test_resume mode can use it in case it is
 357          * invoked from hibernate() to test the snapshot.
 358          */
 359         swsusp_resume_device = hib_resume_bdev->bd_dev;
 360         return res;
 361 }
 362
 363 /**
 364  *      write_page - Write one page to given swap location.
 365  *      @buf:           Address we're writing.
 366  *      @offset:        Offset of the swap page we're writing to.
 367  *      @hb:            bio completion batch
 368  */
 369
 370 static int write_page(void *buf, sector_t offset, struct hib_bio_batch *hb)
 371 {
 372         void *src;
 373         int ret;
 374
 375         if (!offset)
 376                 return -ENOSPC;
 377
 378         if (hb) {
 379                 src = (void *)__get_free_page(__GFP_RECLAIM | __GFP_NOWARN |
 380                                               __GFP_NORETRY);
 381                 if (src) {
 382                         copy_page(src, buf);
 383                 } else {
 384                         ret = hib_wait_io(hb); /* Free pages */
 385                         if (ret)
 386                                 return ret;
 387                         src = (void *)__get_free_page(__GFP_RECLAIM |
 388                                                       __GFP_NOWARN |
 389                                                       __GFP_NORETRY);
 390                         if (src) {
 391                                 copy_page(src, buf);
 392                         } else {
 393                                 WARN_ON_ONCE(1);
 394                                 hb = NULL;      /* Go synchronous */
 395                                 src = buf;
 396                         }
 397                 }
 398         } else {
 399                 src = buf;
 400         }
 401         return hib_submit_io(REQ_OP_WRITE, REQ_SYNC, offset, src, hb);
 402 }
 403
 404 static void release_swap_writer(struct swap_map_handle *handle)
 405 {
 406         if (handle->cur)
 407                 free_page((unsigned long)handle->cur);
 408         handle->cur = NULL;
 409 }
 410
 411 static int get_swap_writer(struct swap_map_handle *handle)
 412 {
 413         int ret;
 414
 415         ret = swsusp_swap_check();
 416         if (ret) {
 417                 if (ret != -ENOSPC)
 418                         pr_err("Cannot find swap device, try swapon -a\n");
 419                 return ret;
 420         }
 421         handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_KERNEL);
 422         if (!handle->cur) {
 423                 ret = -ENOMEM;
 424                 goto err_close;
 425         }
 426         handle->cur_swap = alloc_swapdev_block(root_swap);
 427         if (!handle->cur_swap) {
 428                 ret = -ENOSPC;
 429                 goto err_rel;
 430         }
 431         handle->k = 0;
 432         handle->reqd_free_pages = reqd_free_pages();
 433         handle->first_sector = handle->cur_swap;
 434         return 0;
 435 err_rel:
 436         release_swap_writer(handle);
 437 err_close:
 438         swsusp_close(FMODE_WRITE);
 439         return ret;
 440 }
 441
 442 static int swap_write_page(struct swap_map_handle *handle, void *buf,
 443                 struct hib_bio_batch *hb)
 444 {
 445         int error = 0;
 446         sector_t offset;
 447
 448         if (!handle->cur)
 449                 return -EINVAL;
 450         offset = alloc_swapdev_block(root_swap);
 451         error = write_page(buf, offset, hb);
 452         if (error)
 453                 return error;
 454         handle->cur->entries[handle->k++] = offset;
 455         if (handle->k >= MAP_PAGE_ENTRIES) {
 456                 offset = alloc_swapdev_block(root_swap);
 457                 if (!offset)
 458                         return -ENOSPC;
 459                 handle->cur->next_swap = offset;
 460                 error = write_page(handle->cur, handle->cur_swap, hb);
 461                 if (error)
 462                         goto out;
 463                 clear_page(handle->cur);
 464                 handle->cur_swap = offset;
 465                 handle->k = 0;
 466
 467                 if (hb && low_free_pages() <= handle->reqd_free_pages) {
 468                         error = hib_wait_io(hb);
 469                         if (error)
 470                                 goto out;
 471                         /*
 472                          * Recalculate the number of required free pages, to
 473                          * make sure we never take more than half.
 474                          */
 475                         handle->reqd_free_pages = reqd_free_pages();
 476                 }
 477         }
 478  out:
 479         return error;
 480 }
 481
 482 static int flush_swap_writer(struct swap_map_handle *handle)
 483 {
 484         if (handle->cur && handle->cur_swap)
 485                 return write_page(handle->cur, handle->cur_swap, NULL);
 486         else
 487                 return -EINVAL;
 488 }
 489
 490 static int swap_writer_finish(struct swap_map_handle *handle,
 491                 unsigned int flags, int error)
 492 {
 493         if (!error) {
 494                 flush_swap_writer(handle);
 495                 pr_info("S");
 496                 error = mark_swapfiles(handle, flags);
 497                 pr_cont("|\n");
 498         }
 499
 500         if (error)
 501                 free_all_swap_pages(root_swap);
 502         release_swap_writer(handle);
 503         swsusp_close(FMODE_WRITE);
 504
 505         return error;
 506 }
 507
 508 /* We need to remember how much compressed data we need to read. */
 509 #define LZO_HEADER      sizeof(size_t)
 510
 511 /* Number of pages/bytes we'll compress at one time. */
 512 #define LZO_UNC_PAGES   32
 513 #define LZO_UNC_SIZE    (LZO_UNC_PAGES * PAGE_SIZE)
 514
 515 /* Number of pages/bytes we need for compressed data (worst case). */
 516 #define LZO_CMP_PAGES   DIV_ROUND_UP(lzo1x_worst_compress(LZO_UNC_SIZE) + \
 517                                      LZO_HEADER, PAGE_SIZE)
 518 #define LZO_CMP_SIZE    (LZO_CMP_PAGES * PAGE_SIZE)
 519
 520 /* Maximum number of threads for compression/decompression. */
 521 #define LZO_THREADS     3
 522
 523 /* Minimum/maximum number of pages for read buffering. */
 524 #define LZO_MIN_RD_PAGES        1024
 525 #define LZO_MAX_RD_PAGES        8192
 526
 527
 528 /**
 529  *      save_image - save the suspend image data
 530  */
 531
 532 static int save_image(struct swap_map_handle *handle,
 533                       struct snapshot_handle *snapshot,
 534                       unsigned int nr_to_write)
 535 {
 536         unsigned int m;
 537         int ret;
 538         int nr_pages;
 539         int err2;
 540         struct hib_bio_batch hb;
 541         ktime_t start;
 542         ktime_t stop;
 543
 544         hib_init_batch(&hb);
 545
 546         pr_info("Saving image data pages (%u pages)...\n",
 547                 nr_to_write);
 548         m = nr_to_write / 10;
 549         if (!m)
 550                 m = 1;
 551         nr_pages = 0;
 552         start = ktime_get();
 553         while (1) {
 554                 ret = snapshot_read_next(snapshot);
 555                 if (ret <= 0)
 556                         break;
 557                 ret = swap_write_page(handle, data_of(*snapshot), &hb);
 558                 if (ret)
 559                         break;
 560                 if (!(nr_pages % m))
 561                         pr_info("Image saving progress: %3d%%\n",
 562                                 nr_pages / m * 10);
 563                 nr_pages++;
 564         }
 565         err2 = hib_wait_io(&hb);
 566         stop = ktime_get();
 567         if (!ret)
 568                 ret = err2;
 569         if (!ret)
 570                 pr_info("Image saving done\n");
 571         swsusp_show_speed(start, stop, nr_to_write, "Wrote");
 572         return ret;
 573 }
 574
 575 /**
 576  * Structure used for CRC32.
 577  */
 578 struct crc_data {
 579         struct task_struct *thr;                  /* thread */
 580         atomic_t ready;                           /* ready to start flag */
 581         atomic_t stop;                            /* ready to stop flag */
 582         unsigned run_threads;                     /* nr current threads */
 583         wait_queue_head_t go;                     /* start crc update */
 584         wait_queue_head_t done;                   /* crc update done */
 585         u32 *crc32;                               /* points to handle's crc32 */
 586         size_t *unc_len[LZO_THREADS];             /* uncompressed lengths */
 587         unsigned char *unc[LZO_THREADS];          /* uncompressed data */
 588 };
 589
 590 /**
 591  * CRC32 update function that runs in its own thread.
 592  */
 593 static int crc32_threadfn(void *data)
 594 {
 595         struct crc_data *d = data;
 596         unsigned i;
 597
 598         while (1) {
 599                 wait_event(d->go, atomic_read(&d->ready) ||
 600                                   kthread_should_stop());
 601                 if (kthread_should_stop()) {
 602                         d->thr = NULL;
 603                         atomic_set(&d->stop, 1);
 604                         wake_up(&d->done);
 605                         break;
 606                 }
 607                 atomic_set(&d->ready, 0);
 608
 609                 for (i = 0; i < d->run_threads; i++)
 610                         *d->crc32 = crc32_le(*d->crc32,
 611                                              d->unc[i], *d->unc_len[i]);
 612                 atomic_set(&d->stop, 1);
 613                 wake_up(&d->done);
 614         }
 615         return 0;
 616 }
 617 /**
 618  * Structure used for LZO data compression.
 619  */
 620 struct cmp_data {
 621         struct task_struct *thr;                  /* thread */
 622         atomic_t ready;                           /* ready to start flag */
 623         atomic_t stop;                            /* ready to stop flag */
 624         int ret;                                  /* return code */
 625         wait_queue_head_t go;                     /* start compression */
 626         wait_queue_head_t done;                   /* compression done */
 627         size_t unc_len;                           /* uncompressed length */
 628         size_t cmp_len;                           /* compressed length */
 629         unsigned char unc[LZO_UNC_SIZE];          /* uncompressed buffer */
 630         unsigned char cmp[LZO_CMP_SIZE];          /* compressed buffer */
 631         unsigned char wrk[LZO1X_1_MEM_COMPRESS];  /* compression workspace */
 632 };
 633
 634 /**
 635  * Compression function that runs in its own thread.
 636  */
 637 static int lzo_compress_threadfn(void *data)
 638 {
 639         struct cmp_data *d = data;
 640
 641         while (1) {
 642                 wait_event(d->go, atomic_read(&d->ready) ||
 643                                   kthread_should_stop());
 644                 if (kthread_should_stop()) {
 645                         d->thr = NULL;
 646                         d->ret = -1;
 647                         atomic_set(&d->stop, 1);
 648                         wake_up(&d->done);
 649                         break;
 650                 }
 651                 atomic_set(&d->ready, 0);
 652
 653                 d->ret = lzo1x_1_compress(d->unc, d->unc_len,
 654                                           d->cmp + LZO_HEADER, &d->cmp_len,
 655                                           d->wrk);
 656                 atomic_set(&d->stop, 1);
 657                 wake_up(&d->done);
 658         }
 659         return 0;
 660 }
 661
 662 /**
 663  * save_image_lzo - Save the suspend image data compressed with LZO.
 664  * @handle: Swap map handle to use for saving the image.
 665  * @snapshot: Image to read data from.
 666  * @nr_to_write: Number of pages to save.
 667  */
 668 static int save_image_lzo(struct swap_map_handle *handle,
 669                           struct snapshot_handle *snapshot,
 670                           unsigned int nr_to_write)
 671 {
 672         unsigned int m;
 673         int ret = 0;
 674         int nr_pages;
 675         int err2;
 676         struct hib_bio_batch hb;
 677         ktime_t start;
 678         ktime_t stop;
 679         size_t off;
 680         unsigned thr, run_threads, nr_threads;
 681         unsigned char *page = NULL;
 682         struct cmp_data *data = NULL;
 683         struct crc_data *crc = NULL;
 684
 685         hib_init_batch(&hb);
 686
 687         /*
 688          * We'll limit the number of threads for compression to limit memory
 689          * footprint.
 690          */
 691         nr_threads = num_online_cpus() - 1;
 692         nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
 693
 694         page = (void *)__get_free_page(__GFP_RECLAIM | __GFP_HIGH);
 695         if (!page) {
 696                 pr_err("Failed to allocate LZO page\n");
 697                 ret = -ENOMEM;
 698                 goto out_clean;
 699         }
 700
 701         data = vmalloc(sizeof(*data) * nr_threads);
 702         if (!data) {
 703                 pr_err("Failed to allocate LZO data\n");
 704                 ret = -ENOMEM;
 705                 goto out_clean;
 706         }
 707         for (thr = 0; thr < nr_threads; thr++)
 708                 memset(&data[thr], 0, offsetof(struct cmp_data, go));
 709
 710         crc = kmalloc(sizeof(*crc), GFP_KERNEL);
 711         if (!crc) {
 712                 pr_err("Failed to allocate crc\n");
 713                 ret = -ENOMEM;
 714                 goto out_clean;
 715         }
 716         memset(crc, 0, offsetof(struct crc_data, go));
 717
 718         /*
 719          * Start the compression threads.
 720          */
 721         for (thr = 0; thr < nr_threads; thr++) {
 722                 init_waitqueue_head(&data[thr].go);
 723                 init_waitqueue_head(&data[thr].done);
 724
 725                 data[thr].thr = kthread_run(lzo_compress_threadfn,
 726                                             &data[thr],
 727                                             "image_compress/%u", thr);
 728                 if (IS_ERR(data[thr].thr)) {
 729                         data[thr].thr = NULL;
 730                         pr_err("Cannot start compression threads\n");
 731                         ret = -ENOMEM;
 732                         goto out_clean;
 733                 }
 734         }
 735
 736         /*
 737          * Start the CRC32 thread.
 738          */
 739         init_waitqueue_head(&crc->go);
 740         init_waitqueue_head(&crc->done);
 741
 742         handle->crc32 = 0;
 743         crc->crc32 = &handle->crc32;
 744         for (thr = 0; thr < nr_threads; thr++) {
 745                 crc->unc[thr] = data[thr].unc;
 746                 crc->unc_len[thr] = &data[thr].unc_len;
 747         }
 748
 749         crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
 750         if (IS_ERR(crc->thr)) {
 751                 crc->thr = NULL;
 752                 pr_err("Cannot start CRC32 thread\n");
 753                 ret = -ENOMEM;
 754                 goto out_clean;
 755         }
 756
 757         /*
 758          * Adjust the number of required free pages after all allocations have
 759          * been done. We don't want to run out of pages when writing.
 760          */
 761         handle->reqd_free_pages = reqd_free_pages();
 762
 763         pr_info("Using %u thread(s) for compression\n", nr_threads);
 764         pr_info("Compressing and saving image data (%u pages)...\n",
 765                 nr_to_write);
 766         m = nr_to_write / 10;
 767         if (!m)
 768                 m = 1;
 769         nr_pages = 0;
 770         start = ktime_get();
 771         for (;;) {
 772                 for (thr = 0; thr < nr_threads; thr++) {
 773                         for (off = 0; off < LZO_UNC_SIZE; off += PAGE_SIZE) {
 774                                 ret = snapshot_read_next(snapshot);
 775                                 if (ret < 0)
 776                                         goto out_finish;
 777
 778                                 if (!ret)
 779                                         break;
 780
 781                                 memcpy(data[thr].unc + off,
 782                                        data_of(*snapshot), PAGE_SIZE);
 783
 784                                 if (!(nr_pages % m))
 785                                         pr_info("Image saving progress: %3d%%\n",
 786                                                 nr_pages / m * 10);
 787                                 nr_pages++;
 788                         }
 789                         if (!off)
 790                                 break;
 791
 792                         data[thr].unc_len = off;
 793
 794                         atomic_set(&data[thr].ready, 1);
 795                         wake_up(&data[thr].go);
 796                 }
 797
 798                 if (!thr)
 799                         break;
 800
 801                 crc->run_threads = thr;
 802                 atomic_set(&crc->ready, 1);
 803                 wake_up(&crc->go);
 804
 805                 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
 806                         wait_event(data[thr].done,
 807                                    atomic_read(&data[thr].stop));
 808                         atomic_set(&data[thr].stop, 0);
 809
 810                         ret = data[thr].ret;
 811
 812                         if (ret < 0) {
 813                                 pr_err("LZO compression failed\n");
 814                                 goto out_finish;
 815                         }
 816
 817                         if (unlikely(!data[thr].cmp_len ||
 818                                      data[thr].cmp_len >
 819                                      lzo1x_worst_compress(data[thr].unc_len))) {
 820                                 pr_err("Invalid LZO compressed length\n");
 821                                 ret = -1;
 822                                 goto out_finish;
 823                         }
 824
 825                         *(size_t *)data[thr].cmp = data[thr].cmp_len;
 826
 827                         /*
 828                          * Given we are writing one page at a time to disk, we
 829                          * copy that much from the buffer, although the last
 830                          * bit will likely be smaller than full page. This is
 831                          * OK - we saved the length of the compressed data, so
 832                          * any garbage at the end will be discarded when we
 833                          * read it.
 834                          */
 835                         for (off = 0;
 836                              off < LZO_HEADER + data[thr].cmp_len;
 837                              off += PAGE_SIZE) {
 838                                 memcpy(page, data[thr].cmp + off, PAGE_SIZE);
 839
 840                                 ret = swap_write_page(handle, page, &hb);
 841                                 if (ret)
 842                                         goto out_finish;
 843                         }
 844                 }
 845
 846                 wait_event(crc->done, atomic_read(&crc->stop));
 847                 atomic_set(&crc->stop, 0);
 848         }
 849
 850 out_finish:
 851         err2 = hib_wait_io(&hb);
 852         stop = ktime_get();
 853         if (!ret)
 854                 ret = err2;
 855         if (!ret)
 856                 pr_info("Image saving done\n");
 857         swsusp_show_speed(start, stop, nr_to_write, "Wrote");
 858 out_clean:
 859         if (crc) {
 860                 if (crc->thr)
 861                         kthread_stop(crc->thr);
 862                 kfree(crc);
 863         }
 864         if (data) {
 865                 for (thr = 0; thr < nr_threads; thr++)
 866                         if (data[thr].thr)
 867                                 kthread_stop(data[thr].thr);
 868                 vfree(data);
 869         }
 870         if (page) free_page((unsigned long)page);
 871
 872         return ret;
 873 }
 874
 875 /**
 876  *      enough_swap - Make sure we have enough swap to save the image.
 877  *
 878  *      Returns TRUE or FALSE after checking the total amount of swap
 879  *      space avaiable from the resume partition.
 880  */
 881
 882 static int enough_swap(unsigned int nr_pages)
 883 {
 884         unsigned int free_swap = count_swap_pages(root_swap, 1);
 885         unsigned int required;
 886
 887         pr_debug("Free swap pages: %u\n", free_swap);
 888
 889         required = PAGES_FOR_IO + nr_pages;
 890         return free_swap > required;
 891 }
 892
 893 /**
 894  *      swsusp_write - Write entire image and metadata.
 895  *      @flags: flags to pass to the "boot" kernel in the image header
 896  *
 897  *      It is important _NOT_ to umount filesystems at this point. We want
 898  *      them synced (in case something goes wrong) but we DO not want to mark
 899  *      filesystem clean: it is not. (And it does not matter, if we resume
 900  *      correctly, we'll mark system clean, anyway.)
 901  */
 902
 903 int swsusp_write(unsigned int flags)
 904 {
 905         struct swap_map_handle handle;
 906         struct snapshot_handle snapshot;
 907         struct swsusp_info *header;
 908         unsigned long pages;
 909         int error;
 910
 911         pages = snapshot_get_image_size();
 912         error = get_swap_writer(&handle);
 913         if (error) {
 914                 pr_err("Cannot get swap writer\n");
 915                 return error;
 916         }
 917         if (flags & SF_NOCOMPRESS_MODE) {
 918                 if (!enough_swap(pages)) {
 919                         pr_err("Not enough free swap\n");
 920                         error = -ENOSPC;
 921                         goto out_finish;
 922                 }
 923         }
 924         memset(&snapshot, 0, sizeof(struct snapshot_handle));
 925         error = snapshot_read_next(&snapshot);
 926         if (error < PAGE_SIZE) {
 927                 if (error >= 0)
 928                         error = -EFAULT;
 929
 930                 goto out_finish;
 931         }
 932         header = (struct swsusp_info *)data_of(snapshot);
 933         error = swap_write_page(&handle, header, NULL);
 934         if (!error) {
 935                 error = (flags & SF_NOCOMPRESS_MODE) ?
 936                         save_image(&handle, &snapshot, pages - 1) :
 937                         save_image_lzo(&handle, &snapshot, pages - 1);
 938         }
 939 out_finish:
 940         error = swap_writer_finish(&handle, flags, error);
 941         return error;
 942 }
 943
 944 /**
 945  *      The following functions allow us to read data using a swap map
 946  *      in a file-alike way
 947  */
 948
 949 static void release_swap_reader(struct swap_map_handle *handle)
 950 {
 951         struct swap_map_page_list *tmp;
 952
 953         while (handle->maps) {
 954                 if (handle->maps->map)
 955                         free_page((unsigned long)handle->maps->map);
 956                 tmp = handle->maps;
 957                 handle->maps = handle->maps->next;
 958                 kfree(tmp);
 959         }
 960         handle->cur = NULL;
 961 }
 962
 963 static int get_swap_reader(struct swap_map_handle *handle,
 964                 unsigned int *flags_p)
 965 {
 966         int error;
 967         struct swap_map_page_list *tmp, *last;
 968         sector_t offset;
 969
 970         *flags_p = swsusp_header->flags;
 971
 972         if (!swsusp_header->image) /* how can this happen? */
 973                 return -EINVAL;
 974
 975         handle->cur = NULL;
 976         last = handle->maps = NULL;
 977         offset = swsusp_header->image;
 978         while (offset) {
 979                 tmp = kmalloc(sizeof(*handle->maps), GFP_KERNEL);
 980                 if (!tmp) {
 981                         release_swap_reader(handle);
 982                         return -ENOMEM;
 983                 }
 984                 memset(tmp, 0, sizeof(*tmp));
 985                 if (!handle->maps)
 986                         handle->maps = tmp;
 987                 if (last)
 988                         last->next = tmp;
 989                 last = tmp;
 990
 991                 tmp->map = (struct swap_map_page *)
 992                            __get_free_page(__GFP_RECLAIM | __GFP_HIGH);
 993                 if (!tmp->map) {
 994                         release_swap_reader(handle);
 995                         return -ENOMEM;
 996                 }
 997
 998                 error = hib_submit_io(REQ_OP_READ, 0, offset, tmp->map, NULL);
 999                 if (error) {
1000                         release_swap_reader(handle);
1001                         return error;
1002                 }
1003                 offset = tmp->map->next_swap;
1004         }
1005         handle->k = 0;
1006         handle->cur = handle->maps->map;
1007         return 0;
1008 }
1009
1010 static int swap_read_page(struct swap_map_handle *handle, void *buf,
1011                 struct hib_bio_batch *hb)
1012 {
1013         sector_t offset;
1014         int error;
1015         struct swap_map_page_list *tmp;
1016
1017         if (!handle->cur)
1018                 return -EINVAL;
1019         offset = handle->cur->entries[handle->k];
1020         if (!offset)
1021                 return -EFAULT;
1022         error = hib_submit_io(REQ_OP_READ, 0, offset, buf, hb);
1023         if (error)
1024                 return error;
1025         if (++handle->k >= MAP_PAGE_ENTRIES) {
1026                 handle->k = 0;
1027                 free_page((unsigned long)handle->maps->map);
1028                 tmp = handle->maps;
1029                 handle->maps = handle->maps->next;
1030                 kfree(tmp);
1031                 if (!handle->maps)
1032                         release_swap_reader(handle);
1033                 else
1034                         handle->cur = handle->maps->map;
1035         }
1036         return error;
1037 }
1038
1039 static int swap_reader_finish(struct swap_map_handle *handle)
1040 {
1041         release_swap_reader(handle);
1042
1043         return 0;
1044 }
1045
1046 /**
1047  *      load_image - load the image using the swap map handle
1048  *      @handle and the snapshot handle @snapshot
1049  *      (assume there are @nr_pages pages to load)
1050  */
1051
1052 static int load_image(struct swap_map_handle *handle,
1053                       struct snapshot_handle *snapshot,
1054                       unsigned int nr_to_read)
1055 {
1056         unsigned int m;
1057         int ret = 0;
1058         ktime_t start;
1059         ktime_t stop;
1060         struct hib_bio_batch hb;
1061         int err2;
1062         unsigned nr_pages;
1063
1064         hib_init_batch(&hb);
1065
1066         clean_pages_on_read = true;
1067         pr_info("Loading image data pages (%u pages)...\n", nr_to_read);
1068         m = nr_to_read / 10;
1069         if (!m)
1070                 m = 1;
1071         nr_pages = 0;
1072         start = ktime_get();
1073         for ( ; ; ) {
1074                 ret = snapshot_write_next(snapshot);
1075                 if (ret <= 0)
1076                         break;
1077                 ret = swap_read_page(handle, data_of(*snapshot), &hb);
1078                 if (ret)
1079                         break;
1080                 if (snapshot->sync_read)
1081                         ret = hib_wait_io(&hb);
1082                 if (ret)
1083                         break;
1084                 if (!(nr_pages % m))
1085                         pr_info("Image loading progress: %3d%%\n",
1086                                 nr_pages / m * 10);
1087                 nr_pages++;
1088         }
1089         err2 = hib_wait_io(&hb);
1090         stop = ktime_get();
1091         if (!ret)
1092                 ret = err2;
1093         if (!ret) {
1094                 pr_info("Image loading done\n");
1095                 snapshot_write_finalize(snapshot);
1096                 if (!snapshot_image_loaded(snapshot))
1097                         ret = -ENODATA;
1098         }
1099         swsusp_show_speed(start, stop, nr_to_read, "Read");
1100         return ret;
1101 }
1102
1103 /**
1104  * Structure used for LZO data decompression.
1105  */
1106 struct dec_data {
1107         struct task_struct *thr;                  /* thread */
1108         atomic_t ready;                           /* ready to start flag */
1109         atomic_t stop;                            /* ready to stop flag */
1110         int ret;                                  /* return code */
1111         wait_queue_head_t go;                     /* start decompression */
1112         wait_queue_head_t done;                   /* decompression done */
1113         size_t unc_len;                           /* uncompressed length */
1114         size_t cmp_len;                           /* compressed length */
1115         unsigned char unc[LZO_UNC_SIZE];          /* uncompressed buffer */
1116         unsigned char cmp[LZO_CMP_SIZE];          /* compressed buffer */
1117 };
1118
1119 /**
1120  * Deompression function that runs in its own thread.
1121  */
1122 static int lzo_decompress_threadfn(void *data)
1123 {
1124         struct dec_data *d = data;
1125
1126         while (1) {
1127                 wait_event(d->go, atomic_read(&d->ready) ||
1128                                   kthread_should_stop());
1129                 if (kthread_should_stop()) {
1130                         d->thr = NULL;
1131                         d->ret = -1;
1132                         atomic_set(&d->stop, 1);
1133                         wake_up(&d->done);
1134                         break;
1135                 }
1136                 atomic_set(&d->ready, 0);
1137
1138                 d->unc_len = LZO_UNC_SIZE;
1139                 d->ret = lzo1x_decompress_safe(d->cmp + LZO_HEADER, d->cmp_len,
1140                                                d->unc, &d->unc_len);
1141                 if (clean_pages_on_decompress)
1142                         flush_icache_range((unsigned long)d->unc,
1143                                            (unsigned long)d->unc + d->unc_len);
1144
1145                 atomic_set(&d->stop, 1);
1146                 wake_up(&d->done);
1147         }
1148         return 0;
1149 }
1150
1151 /**
1152  * load_image_lzo - Load compressed image data and decompress them with LZO.
1153  * @handle: Swap map handle to use for loading data.
1154  * @snapshot: Image to copy uncompressed data into.
1155  * @nr_to_read: Number of pages to load.
1156  */
1157 static int load_image_lzo(struct swap_map_handle *handle,
1158                           struct snapshot_handle *snapshot,
1159                           unsigned int nr_to_read)
1160 {
1161         unsigned int m;
1162         int ret = 0;
1163         int eof = 0;
1164         struct hib_bio_batch hb;
1165         ktime_t start;
1166         ktime_t stop;
1167         unsigned nr_pages;
1168         size_t off;
1169         unsigned i, thr, run_threads, nr_threads;
1170         unsigned ring = 0, pg = 0, ring_size = 0,
1171                  have = 0, want, need, asked = 0;
1172         unsigned long read_pages = 0;
1173         unsigned char **page = NULL;
1174         struct dec_data *data = NULL;
1175         struct crc_data *crc = NULL;
1176
1177         hib_init_batch(&hb);
1178
1179         /*
1180          * We'll limit the number of threads for decompression to limit memory
1181          * footprint.
1182          */
1183         nr_threads = num_online_cpus() - 1;
1184         nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
1185
1186         page = vmalloc(sizeof(*page) * LZO_MAX_RD_PAGES);
1187         if (!page) {
1188                 pr_err("Failed to allocate LZO page\n");
1189                 ret = -ENOMEM;
1190                 goto out_clean;
1191         }
1192
1193         data = vmalloc(sizeof(*data) * nr_threads);
1194         if (!data) {
1195                 pr_err("Failed to allocate LZO data\n");
1196                 ret = -ENOMEM;
1197                 goto out_clean;
1198         }
1199         for (thr = 0; thr < nr_threads; thr++)
1200                 memset(&data[thr], 0, offsetof(struct dec_data, go));
1201
1202         crc = kmalloc(sizeof(*crc), GFP_KERNEL);
1203         if (!crc) {
1204                 pr_err("Failed to allocate crc\n");
1205                 ret = -ENOMEM;
1206                 goto out_clean;
1207         }
1208         memset(crc, 0, offsetof(struct crc_data, go));
1209
1210         clean_pages_on_decompress = true;
1211
1212         /*
1213          * Start the decompression threads.
1214          */
1215         for (thr = 0; thr < nr_threads; thr++) {
1216                 init_waitqueue_head(&data[thr].go);
1217                 init_waitqueue_head(&data[thr].done);
1218
1219                 data[thr].thr = kthread_run(lzo_decompress_threadfn,
1220                                             &data[thr],
1221                                             "image_decompress/%u", thr);
1222                 if (IS_ERR(data[thr].thr)) {
1223                         data[thr].thr = NULL;
1224                         pr_err("Cannot start decompression threads\n");
1225                         ret = -ENOMEM;
1226                         goto out_clean;
1227                 }
1228         }
1229
1230         /*
1231          * Start the CRC32 thread.
1232          */
1233         init_waitqueue_head(&crc->go);
1234         init_waitqueue_head(&crc->done);
1235
1236         handle->crc32 = 0;
1237         crc->crc32 = &handle->crc32;
1238         for (thr = 0; thr < nr_threads; thr++) {
1239                 crc->unc[thr] = data[thr].unc;
1240                 crc->unc_len[thr] = &data[thr].unc_len;
1241         }
1242
1243         crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
1244         if (IS_ERR(crc->thr)) {
1245                 crc->thr = NULL;
1246                 pr_err("Cannot start CRC32 thread\n");
1247                 ret = -ENOMEM;
1248                 goto out_clean;
1249         }
1250
1251         /*
1252          * Set the number of pages for read buffering.
1253          * This is complete guesswork, because we'll only know the real
1254          * picture once prepare_image() is called, which is much later on
1255          * during the image load phase. We'll assume the worst case and
1256          * say that none of the image pages are from high memory.
1257          */
1258         if (low_free_pages() > snapshot_get_image_size())
1259                 read_pages = (low_free_pages() - snapshot_get_image_size()) / 2;
1260         read_pages = clamp_val(read_pages, LZO_MIN_RD_PAGES, LZO_MAX_RD_PAGES);
1261
1262         for (i = 0; i < read_pages; i++) {
1263                 page[i] = (void *)__get_free_page(i < LZO_CMP_PAGES ?
1264                                                   __GFP_RECLAIM | __GFP_HIGH :
1265                                                   __GFP_RECLAIM | __GFP_NOWARN |
1266                                                   __GFP_NORETRY);
1267
1268                 if (!page[i]) {
1269                         if (i < LZO_CMP_PAGES) {
1270                                 ring_size = i;
1271                                 pr_err("Failed to allocate LZO pages\n");
1272                                 ret = -ENOMEM;
1273                                 goto out_clean;
1274                         } else {
1275                                 break;
1276                         }
1277                 }
1278         }
1279         want = ring_size = i;
1280
1281         pr_info("Using %u thread(s) for decompression\n", nr_threads);
1282         pr_info("Loading and decompressing image data (%u pages)...\n",
1283                 nr_to_read);
1284         m = nr_to_read / 10;
1285         if (!m)
1286                 m = 1;
1287         nr_pages = 0;
1288         start = ktime_get();
1289
1290         ret = snapshot_write_next(snapshot);
1291         if (ret <= 0)
1292                 goto out_finish;
1293
1294         for(;;) {
1295                 for (i = 0; !eof && i < want; i++) {
1296                         ret = swap_read_page(handle, page[ring], &hb);
1297                         if (ret) {
1298                                 /*
1299                                  * On real read error, finish. On end of data,
1300                                  * set EOF flag and just exit the read loop.
1301                                  */
1302                                 if (handle->cur &&
1303                                     handle->cur->entries[handle->k]) {
1304                                         goto out_finish;
1305                                 } else {
1306                                         eof = 1;
1307                                         break;
1308                                 }
1309                         }
1310                         if (++ring >= ring_size)
1311                                 ring = 0;
1312                 }
1313                 asked += i;
1314                 want -= i;
1315
1316                 /*
1317                  * We are out of data, wait for some more.
1318                  */
1319                 if (!have) {
1320                         if (!asked)
1321                                 break;
1322
1323                         ret = hib_wait_io(&hb);
1324                         if (ret)
1325                                 goto out_finish;
1326                         have += asked;
1327                         asked = 0;
1328                         if (eof)
1329                                 eof = 2;
1330                 }
1331
1332                 if (crc->run_threads) {
1333                         wait_event(crc->done, atomic_read(&crc->stop));
1334                         atomic_set(&crc->stop, 0);
1335                         crc->run_threads = 0;
1336                 }
1337
1338                 for (thr = 0; have && thr < nr_threads; thr++) {
1339                         data[thr].cmp_len = *(size_t *)page[pg];
1340                         if (unlikely(!data[thr].cmp_len ||
1341                                      data[thr].cmp_len >
1342                                      lzo1x_worst_compress(LZO_UNC_SIZE))) {
1343                                 pr_err("Invalid LZO compressed length\n");
1344                                 ret = -1;
1345                                 goto out_finish;
1346                         }
1347
1348                         need = DIV_ROUND_UP(data[thr].cmp_len + LZO_HEADER,
1349                                             PAGE_SIZE);
1350                         if (need > have) {
1351                                 if (eof > 1) {
1352                                         ret = -1;
1353                                         goto out_finish;
1354                                 }
1355                                 break;
1356                         }
1357
1358                         for (off = 0;
1359                              off < LZO_HEADER + data[thr].cmp_len;
1360                              off += PAGE_SIZE) {
1361                                 memcpy(data[thr].cmp + off,
1362                                        page[pg], PAGE_SIZE);
1363                                 have--;
1364                                 want++;
1365                                 if (++pg >= ring_size)
1366                                         pg = 0;
1367                         }
1368
1369                         atomic_set(&data[thr].ready, 1);
1370                         wake_up(&data[thr].go);
1371                 }
1372
1373                 /*
1374                  * Wait for more data while we are decompressing.
1375                  */
1376                 if (have < LZO_CMP_PAGES && asked) {
1377                         ret = hib_wait_io(&hb);
1378                         if (ret)
1379                                 goto out_finish;
1380                         have += asked;
1381                         asked = 0;
1382                         if (eof)
1383                                 eof = 2;
1384                 }
1385
1386                 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
1387                         wait_event(data[thr].done,
1388                                    atomic_read(&data[thr].stop));
1389                         atomic_set(&data[thr].stop, 0);
1390
1391                         ret = data[thr].ret;
1392
1393                         if (ret < 0) {
1394                                 pr_err("LZO decompression failed\n");
1395                                 goto out_finish;
1396                         }
1397
1398                         if (unlikely(!data[thr].unc_len ||
1399                                      data[thr].unc_len > LZO_UNC_SIZE ||
1400                                      data[thr].unc_len & (PAGE_SIZE - 1))) {
1401                                 pr_err("Invalid LZO uncompressed length\n");
1402                                 ret = -1;
1403                                 goto out_finish;
1404                         }
1405
1406                         for (off = 0;
1407                              off < data[thr].unc_len; off += PAGE_SIZE) {
1408                                 memcpy(data_of(*snapshot),
1409                                        data[thr].unc + off, PAGE_SIZE);
1410
1411                                 if (!(nr_pages % m))
1412                                         pr_info("Image loading progress: %3d%%\n",
1413                                                 nr_pages / m * 10);
1414                                 nr_pages++;
1415
1416                                 ret = snapshot_write_next(snapshot);
1417                                 if (ret <= 0) {
1418                                         crc->run_threads = thr + 1;
1419                                         atomic_set(&crc->ready, 1);
1420                                         wake_up(&crc->go);
1421                                         goto out_finish;
1422                                 }
1423                         }
1424                 }
1425
1426                 crc->run_threads = thr;
1427                 atomic_set(&crc->ready, 1);
1428                 wake_up(&crc->go);
1429         }
1430
1431 out_finish:
1432         if (crc->run_threads) {
1433                 wait_event(crc->done, atomic_read(&crc->stop));
1434                 atomic_set(&crc->stop, 0);
1435         }
1436         stop = ktime_get();
1437         if (!ret) {
1438                 pr_info("Image loading done\n");
1439                 snapshot_write_finalize(snapshot);
1440                 if (!snapshot_image_loaded(snapshot))
1441                         ret = -ENODATA;
1442                 if (!ret) {
1443                         if (swsusp_header->flags & SF_CRC32_MODE) {
1444                                 if(handle->crc32 != swsusp_header->crc32) {
1445                                         pr_err("Invalid image CRC32!\n");
1446                                         ret = -ENODATA;
1447                                 }
1448                         }
1449                 }
1450         }
1451         swsusp_show_speed(start, stop, nr_to_read, "Read");
1452 out_clean:
1453         for (i = 0; i < ring_size; i++)
1454                 free_page((unsigned long)page[i]);
1455         if (crc) {
1456                 if (crc->thr)
1457                         kthread_stop(crc->thr);
1458                 kfree(crc);
1459         }
1460         if (data) {
1461                 for (thr = 0; thr < nr_threads; thr++)
1462                         if (data[thr].thr)
1463                                 kthread_stop(data[thr].thr);
1464                 vfree(data);
1465         }
1466         vfree(page);
1467
1468         return ret;
1469 }
1470
1471 /**
1472  *      swsusp_read - read the hibernation image.
1473  *      @flags_p: flags passed by the "frozen" kernel in the image header should
1474  *                be written into this memory location
1475  */
1476
1477 int swsusp_read(unsigned int *flags_p)
1478 {
1479         int error;
1480         struct swap_map_handle handle;
1481         struct snapshot_handle snapshot;
1482         struct swsusp_info *header;
1483
1484         memset(&snapshot, 0, sizeof(struct snapshot_handle));
1485         error = snapshot_write_next(&snapshot);
1486         if (error < PAGE_SIZE)
1487                 return error < 0 ? error : -EFAULT;
1488         header = (struct swsusp_info *)data_of(snapshot);
1489         error = get_swap_reader(&handle, flags_p);
1490         if (error)
1491                 goto end;
1492         if (!error)
1493                 error = swap_read_page(&handle, header, NULL);
1494         if (!error) {
1495                 error = (*flags_p & SF_NOCOMPRESS_MODE) ?
1496                         load_image(&handle, &snapshot, header->pages - 1) :
1497                         load_image_lzo(&handle, &snapshot, header->pages - 1);
1498         }
1499         swap_reader_finish(&handle);
1500 end:
1501         if (!error)
1502                 pr_debug("Image successfully loaded\n");
1503         else
1504                 pr_debug("Error %d resuming\n", error);
1505         return error;
1506 }
1507
1508 /**
1509  *      swsusp_check - Check for swsusp signature in the resume device
1510  */
1511
1512 int swsusp_check(void)
1513 {
1514         int error;
1515
1516         hib_resume_bdev = blkdev_get_by_dev(swsusp_resume_device,
1517                                             FMODE_READ, NULL);
1518         if (!IS_ERR(hib_resume_bdev)) {
1519                 set_blocksize(hib_resume_bdev, PAGE_SIZE);
1520                 clear_page(swsusp_header);
1521                 error = hib_submit_io(REQ_OP_READ, 0,
1522                                         swsusp_resume_block,
1523                                         swsusp_header, NULL);
1524                 if (error)
1525                         goto put;
1526
1527                 if (!memcmp(HIBERNATE_SIG, swsusp_header->sig, 10)) {
1528                         memcpy(swsusp_header->sig, swsusp_header->orig_sig, 10);
1529                         /* Reset swap signature now */
1530                         error = hib_submit_io(REQ_OP_WRITE, REQ_SYNC,
1531                                                 swsusp_resume_block,
1532                                                 swsusp_header, NULL);
1533                 } else {
1534                         error = -EINVAL;
1535                 }
1536
1537 put:
1538                 if (error)
1539                         blkdev_put(hib_resume_bdev, FMODE_READ);
1540                 else
1541                         pr_debug("Image signature found, resuming\n");
1542         } else {
1543                 error = PTR_ERR(hib_resume_bdev);
1544         }
1545
1546         if (error)
1547                 pr_debug("Image not found (code %d)\n", error);
1548
1549         return error;
1550 }
1551
1552 /**
1553  *      swsusp_close - close swap device.
1554  */
1555
1556 void swsusp_close(fmode_t mode)
1557 {
1558         if (IS_ERR(hib_resume_bdev)) {
1559                 pr_debug("Image device not initialised\n");
1560                 return;
1561         }
1562
1563         blkdev_put(hib_resume_bdev, mode);
1564 }
1565
1566 /**
1567  *      swsusp_unmark - Unmark swsusp signature in the resume device
1568  */
1569
1570 #ifdef CONFIG_SUSPEND
1571 int swsusp_unmark(void)
1572 {
1573         int error;
1574
1575         hib_submit_io(REQ_OP_READ, 0, swsusp_resume_block,
1576                       swsusp_header, NULL);
1577         if (!memcmp(HIBERNATE_SIG,swsusp_header->sig, 10)) {
1578                 memcpy(swsusp_header->sig,swsusp_header->orig_sig, 10);
1579                 error = hib_submit_io(REQ_OP_WRITE, REQ_SYNC,
1580                                         swsusp_resume_block,
1581                                         swsusp_header, NULL);
1582         } else {
1583                 pr_err("Cannot find swsusp signature!\n");
1584                 error = -ENODEV;
1585         }
1586
1587         /*
1588          * We just returned from suspend, we don't need the image any more.
1589          */
1590         free_all_swap_pages(root_swap);
1591
1592         return error;
1593 }
1594 #endif
1595
1596 static int swsusp_header_init(void)
1597 {
1598         swsusp_header = (struct swsusp_header*) __get_free_page(GFP_KERNEL);
1599         if (!swsusp_header)
1600                 panic("Could not allocate memory for swsusp_header\n");
1601         return 0;
1602 }
1603
1604 core_initcall(swsusp_header_init);