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 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         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(hib_resume_bdev, 1, opf, GFP_NOIO | __GFP_HIGH);
 280         bio->bi_iter.bi_sector = page_off * (PAGE_SIZE >> 9);
 281
 282         if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 283                 pr_err("Adding page to bio failed at %llu\n",
 284                        (unsigned long long)bio->bi_iter.bi_sector);
 285                 bio_put(bio);
 286                 return -EFAULT;
 287         }
 288
 289         if (hb) {
 290                 bio->bi_end_io = hib_end_io;
 291                 bio->bi_private = hb;
 292                 atomic_inc(&hb->count);
 293                 submit_bio(bio);
 294         } else {
 295                 error = submit_bio_wait(bio);
 296                 bio_put(bio);
 297         }
 298
 299         return error;
 300 }
 301
 302 static int hib_wait_io(struct hib_bio_batch *hb)
 303 {
 304         /*
 305          * We are relying on the behavior of blk_plug that a thread with
 306          * a plug will flush the plug list before sleeping.
 307          */
 308         wait_event(hb->wait, atomic_read(&hb->count) == 0);
 309         return blk_status_to_errno(hb->error);
 310 }
 311
 312 /*
 313  * Saving part
 314  */
 315 static int mark_swapfiles(struct swap_map_handle *handle, unsigned int flags)
 316 {
 317         int error;
 318
 319         hib_submit_io(REQ_OP_READ, swsusp_resume_block, swsusp_header, NULL);
 320         if (!memcmp("SWAP-SPACE",swsusp_header->sig, 10) ||
 321             !memcmp("SWAPSPACE2",swsusp_header->sig, 10)) {
 322                 memcpy(swsusp_header->orig_sig,swsusp_header->sig, 10);
 323                 memcpy(swsusp_header->sig, HIBERNATE_SIG, 10);
 324                 swsusp_header->image = handle->first_sector;
 325                 if (swsusp_hardware_signature) {
 326                         swsusp_header->hw_sig = swsusp_hardware_signature;
 327                         flags |= SF_HW_SIG;
 328                 }
 329                 swsusp_header->flags = flags;
 330                 if (flags & SF_CRC32_MODE)
 331                         swsusp_header->crc32 = handle->crc32;
 332                 error = hib_submit_io(REQ_OP_WRITE | REQ_SYNC,
 333                                       swsusp_resume_block, swsusp_header, NULL);
 334         } else {
 335                 pr_err("Swap header not found!\n");
 336                 error = -ENODEV;
 337         }
 338         return error;
 339 }
 340
 341 /**
 342  *      swsusp_swap_check - check if the resume device is a swap device
 343  *      and get its index (if so)
 344  *
 345  *      This is called before saving image
 346  */
 347 static int swsusp_swap_check(void)
 348 {
 349         int res;
 350
 351         if (swsusp_resume_device)
 352                 res = swap_type_of(swsusp_resume_device, swsusp_resume_block);
 353         else
 354                 res = find_first_swap(&swsusp_resume_device);
 355         if (res < 0)
 356                 return res;
 357         root_swap = res;
 358
 359         hib_resume_bdev = blkdev_get_by_dev(swsusp_resume_device, FMODE_WRITE,
 360                         NULL);
 361         if (IS_ERR(hib_resume_bdev))
 362                 return PTR_ERR(hib_resume_bdev);
 363
 364         res = set_blocksize(hib_resume_bdev, PAGE_SIZE);
 365         if (res < 0)
 366                 blkdev_put(hib_resume_bdev, FMODE_WRITE);
 367
 368         return res;
 369 }
 370
 371 /**
 372  *      write_page - Write one page to given swap location.
 373  *      @buf:           Address we're writing.
 374  *      @offset:        Offset of the swap page we're writing to.
 375  *      @hb:            bio completion batch
 376  */
 377
 378 static int write_page(void *buf, sector_t offset, struct hib_bio_batch *hb)
 379 {
 380         void *src;
 381         int ret;
 382
 383         if (!offset)
 384                 return -ENOSPC;
 385
 386         if (hb) {
 387                 src = (void *)__get_free_page(GFP_NOIO | __GFP_NOWARN |
 388                                               __GFP_NORETRY);
 389                 if (src) {
 390                         copy_page(src, buf);
 391                 } else {
 392                         ret = hib_wait_io(hb); /* Free pages */
 393                         if (ret)
 394                                 return ret;
 395                         src = (void *)__get_free_page(GFP_NOIO |
 396                                                       __GFP_NOWARN |
 397                                                       __GFP_NORETRY);
 398                         if (src) {
 399                                 copy_page(src, buf);
 400                         } else {
 401                                 WARN_ON_ONCE(1);
 402                                 hb = NULL;      /* Go synchronous */
 403                                 src = buf;
 404                         }
 405                 }
 406         } else {
 407                 src = buf;
 408         }
 409         return hib_submit_io(REQ_OP_WRITE | REQ_SYNC, offset, src, hb);
 410 }
 411
 412 static void release_swap_writer(struct swap_map_handle *handle)
 413 {
 414         if (handle->cur)
 415                 free_page((unsigned long)handle->cur);
 416         handle->cur = NULL;
 417 }
 418
 419 static int get_swap_writer(struct swap_map_handle *handle)
 420 {
 421         int ret;
 422
 423         ret = swsusp_swap_check();
 424         if (ret) {
 425                 if (ret != -ENOSPC)
 426                         pr_err("Cannot find swap device, try swapon -a\n");
 427                 return ret;
 428         }
 429         handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_KERNEL);
 430         if (!handle->cur) {
 431                 ret = -ENOMEM;
 432                 goto err_close;
 433         }
 434         handle->cur_swap = alloc_swapdev_block(root_swap);
 435         if (!handle->cur_swap) {
 436                 ret = -ENOSPC;
 437                 goto err_rel;
 438         }
 439         handle->k = 0;
 440         handle->reqd_free_pages = reqd_free_pages();
 441         handle->first_sector = handle->cur_swap;
 442         return 0;
 443 err_rel:
 444         release_swap_writer(handle);
 445 err_close:
 446         swsusp_close(FMODE_WRITE);
 447         return ret;
 448 }
 449
 450 static int swap_write_page(struct swap_map_handle *handle, void *buf,
 451                 struct hib_bio_batch *hb)
 452 {
 453         int error = 0;
 454         sector_t offset;
 455
 456         if (!handle->cur)
 457                 return -EINVAL;
 458         offset = alloc_swapdev_block(root_swap);
 459         error = write_page(buf, offset, hb);
 460         if (error)
 461                 return error;
 462         handle->cur->entries[handle->k++] = offset;
 463         if (handle->k >= MAP_PAGE_ENTRIES) {
 464                 offset = alloc_swapdev_block(root_swap);
 465                 if (!offset)
 466                         return -ENOSPC;
 467                 handle->cur->next_swap = offset;
 468                 error = write_page(handle->cur, handle->cur_swap, hb);
 469                 if (error)
 470                         goto out;
 471                 clear_page(handle->cur);
 472                 handle->cur_swap = offset;
 473                 handle->k = 0;
 474
 475                 if (hb && low_free_pages() <= handle->reqd_free_pages) {
 476                         error = hib_wait_io(hb);
 477                         if (error)
 478                                 goto out;
 479                         /*
 480                          * Recalculate the number of required free pages, to
 481                          * make sure we never take more than half.
 482                          */
 483                         handle->reqd_free_pages = reqd_free_pages();
 484                 }
 485         }
 486  out:
 487         return error;
 488 }
 489
 490 static int flush_swap_writer(struct swap_map_handle *handle)
 491 {
 492         if (handle->cur && handle->cur_swap)
 493                 return write_page(handle->cur, handle->cur_swap, NULL);
 494         else
 495                 return -EINVAL;
 496 }
 497
 498 static int swap_writer_finish(struct swap_map_handle *handle,
 499                 unsigned int flags, int error)
 500 {
 501         if (!error) {
 502                 pr_info("S");
 503                 error = mark_swapfiles(handle, flags);
 504                 pr_cont("|\n");
 505                 flush_swap_writer(handle);
 506         }
 507
 508         if (error)
 509                 free_all_swap_pages(root_swap);
 510         release_swap_writer(handle);
 511         swsusp_close(FMODE_WRITE);
 512
 513         return error;
 514 }
 515
 516 /* We need to remember how much compressed data we need to read. */
 517 #define LZO_HEADER      sizeof(size_t)
 518
 519 /* Number of pages/bytes we'll compress at one time. */
 520 #define LZO_UNC_PAGES   32
 521 #define LZO_UNC_SIZE    (LZO_UNC_PAGES * PAGE_SIZE)
 522
 523 /* Number of pages/bytes we need for compressed data (worst case). */
 524 #define LZO_CMP_PAGES   DIV_ROUND_UP(lzo1x_worst_compress(LZO_UNC_SIZE) + \
 525                                      LZO_HEADER, PAGE_SIZE)
 526 #define LZO_CMP_SIZE    (LZO_CMP_PAGES * PAGE_SIZE)
 527
 528 /* Maximum number of threads for compression/decompression. */
 529 #define LZO_THREADS     3
 530
 531 /* Minimum/maximum number of pages for read buffering. */
 532 #define LZO_MIN_RD_PAGES        1024
 533 #define LZO_MAX_RD_PAGES        8192
 534
 535
 536 /**
 537  *      save_image - save the suspend image data
 538  */
 539
 540 static int save_image(struct swap_map_handle *handle,
 541                       struct snapshot_handle *snapshot,
 542                       unsigned int nr_to_write)
 543 {
 544         unsigned int m;
 545         int ret;
 546         int nr_pages;
 547         int err2;
 548         struct hib_bio_batch hb;
 549         ktime_t start;
 550         ktime_t stop;
 551
 552         hib_init_batch(&hb);
 553
 554         pr_info("Saving image data pages (%u pages)...\n",
 555                 nr_to_write);
 556         m = nr_to_write / 10;
 557         if (!m)
 558                 m = 1;
 559         nr_pages = 0;
 560         start = ktime_get();
 561         while (1) {
 562                 ret = snapshot_read_next(snapshot);
 563                 if (ret <= 0)
 564                         break;
 565                 ret = swap_write_page(handle, data_of(*snapshot), &hb);
 566                 if (ret)
 567                         break;
 568                 if (!(nr_pages % m))
 569                         pr_info("Image saving progress: %3d%%\n",
 570                                 nr_pages / m * 10);
 571                 nr_pages++;
 572         }
 573         err2 = hib_wait_io(&hb);
 574         hib_finish_batch(&hb);
 575         stop = ktime_get();
 576         if (!ret)
 577                 ret = err2;
 578         if (!ret)
 579                 pr_info("Image saving done\n");
 580         swsusp_show_speed(start, stop, nr_to_write, "Wrote");
 581         return ret;
 582 }
 583
 584 /**
 585  * Structure used for CRC32.
 586  */
 587 struct crc_data {
 588         struct task_struct *thr;                  /* thread */
 589         atomic_t ready;                           /* ready to start flag */
 590         atomic_t stop;                            /* ready to stop flag */
 591         unsigned run_threads;                     /* nr current threads */
 592         wait_queue_head_t go;                     /* start crc update */
 593         wait_queue_head_t done;                   /* crc update done */
 594         u32 *crc32;                               /* points to handle's crc32 */
 595         size_t *unc_len[LZO_THREADS];             /* uncompressed lengths */
 596         unsigned char *unc[LZO_THREADS];          /* uncompressed data */
 597 };
 598
 599 /**
 600  * CRC32 update function that runs in its own thread.
 601  */
 602 static int crc32_threadfn(void *data)
 603 {
 604         struct crc_data *d = data;
 605         unsigned i;
 606
 607         while (1) {
 608                 wait_event(d->go, atomic_read(&d->ready) ||
 609                                   kthread_should_stop());
 610                 if (kthread_should_stop()) {
 611                         d->thr = NULL;
 612                         atomic_set(&d->stop, 1);
 613                         wake_up(&d->done);
 614                         break;
 615                 }
 616                 atomic_set(&d->ready, 0);
 617
 618                 for (i = 0; i < d->run_threads; i++)
 619                         *d->crc32 = crc32_le(*d->crc32,
 620                                              d->unc[i], *d->unc_len[i]);
 621                 atomic_set(&d->stop, 1);
 622                 wake_up(&d->done);
 623         }
 624         return 0;
 625 }
 626 /**
 627  * Structure used for LZO data compression.
 628  */
 629 struct cmp_data {
 630         struct task_struct *thr;                  /* thread */
 631         atomic_t ready;                           /* ready to start flag */
 632         atomic_t stop;                            /* ready to stop flag */
 633         int ret;                                  /* return code */
 634         wait_queue_head_t go;                     /* start compression */
 635         wait_queue_head_t done;                   /* compression done */
 636         size_t unc_len;                           /* uncompressed length */
 637         size_t cmp_len;                           /* compressed length */
 638         unsigned char unc[LZO_UNC_SIZE];          /* uncompressed buffer */
 639         unsigned char cmp[LZO_CMP_SIZE];          /* compressed buffer */
 640         unsigned char wrk[LZO1X_1_MEM_COMPRESS];  /* compression workspace */
 641 };
 642
 643 /**
 644  * Compression function that runs in its own thread.
 645  */
 646 static int lzo_compress_threadfn(void *data)
 647 {
 648         struct cmp_data *d = data;
 649
 650         while (1) {
 651                 wait_event(d->go, atomic_read(&d->ready) ||
 652                                   kthread_should_stop());
 653                 if (kthread_should_stop()) {
 654                         d->thr = NULL;
 655                         d->ret = -1;
 656                         atomic_set(&d->stop, 1);
 657                         wake_up(&d->done);
 658                         break;
 659                 }
 660                 atomic_set(&d->ready, 0);
 661
 662                 d->ret = lzo1x_1_compress(d->unc, d->unc_len,
 663                                           d->cmp + LZO_HEADER, &d->cmp_len,
 664                                           d->wrk);
 665                 atomic_set(&d->stop, 1);
 666                 wake_up(&d->done);
 667         }
 668         return 0;
 669 }
 670
 671 /**
 672  * save_image_lzo - Save the suspend image data compressed with LZO.
 673  * @handle: Swap map handle to use for saving the image.
 674  * @snapshot: Image to read data from.
 675  * @nr_to_write: Number of pages to save.
 676  */
 677 static int save_image_lzo(struct swap_map_handle *handle,
 678                           struct snapshot_handle *snapshot,
 679                           unsigned int nr_to_write)
 680 {
 681         unsigned int m;
 682         int ret = 0;
 683         int nr_pages;
 684         int err2;
 685         struct hib_bio_batch hb;
 686         ktime_t start;
 687         ktime_t stop;
 688         size_t off;
 689         unsigned thr, run_threads, nr_threads;
 690         unsigned char *page = NULL;
 691         struct cmp_data *data = NULL;
 692         struct crc_data *crc = NULL;
 693
 694         hib_init_batch(&hb);
 695
 696         /*
 697          * We'll limit the number of threads for compression to limit memory
 698          * footprint.
 699          */
 700         nr_threads = num_online_cpus() - 1;
 701         nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
 702
 703         page = (void *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 704         if (!page) {
 705                 pr_err("Failed to allocate LZO page\n");
 706                 ret = -ENOMEM;
 707                 goto out_clean;
 708         }
 709
 710         data = vzalloc(array_size(nr_threads, sizeof(*data)));
 711         if (!data) {
 712                 pr_err("Failed to allocate LZO data\n");
 713                 ret = -ENOMEM;
 714                 goto out_clean;
 715         }
 716
 717         crc = kzalloc(sizeof(*crc), GFP_KERNEL);
 718         if (!crc) {
 719                 pr_err("Failed to allocate crc\n");
 720                 ret = -ENOMEM;
 721                 goto out_clean;
 722         }
 723
 724         /*
 725          * Start the compression threads.
 726          */
 727         for (thr = 0; thr < nr_threads; thr++) {
 728                 init_waitqueue_head(&data[thr].go);
 729                 init_waitqueue_head(&data[thr].done);
 730
 731                 data[thr].thr = kthread_run(lzo_compress_threadfn,
 732                                             &data[thr],
 733                                             "image_compress/%u", thr);
 734                 if (IS_ERR(data[thr].thr)) {
 735                         data[thr].thr = NULL;
 736                         pr_err("Cannot start compression threads\n");
 737                         ret = -ENOMEM;
 738                         goto out_clean;
 739                 }
 740         }
 741
 742         /*
 743          * Start the CRC32 thread.
 744          */
 745         init_waitqueue_head(&crc->go);
 746         init_waitqueue_head(&crc->done);
 747
 748         handle->crc32 = 0;
 749         crc->crc32 = &handle->crc32;
 750         for (thr = 0; thr < nr_threads; thr++) {
 751                 crc->unc[thr] = data[thr].unc;
 752                 crc->unc_len[thr] = &data[thr].unc_len;
 753         }
 754
 755         crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
 756         if (IS_ERR(crc->thr)) {
 757                 crc->thr = NULL;
 758                 pr_err("Cannot start CRC32 thread\n");
 759                 ret = -ENOMEM;
 760                 goto out_clean;
 761         }
 762
 763         /*
 764          * Adjust the number of required free pages after all allocations have
 765          * been done. We don't want to run out of pages when writing.
 766          */
 767         handle->reqd_free_pages = reqd_free_pages();
 768
 769         pr_info("Using %u thread(s) for compression\n", nr_threads);
 770         pr_info("Compressing and saving image data (%u pages)...\n",
 771                 nr_to_write);
 772         m = nr_to_write / 10;
 773         if (!m)
 774                 m = 1;
 775         nr_pages = 0;
 776         start = ktime_get();
 777         for (;;) {
 778                 for (thr = 0; thr < nr_threads; thr++) {
 779                         for (off = 0; off < LZO_UNC_SIZE; off += PAGE_SIZE) {
 780                                 ret = snapshot_read_next(snapshot);
 781                                 if (ret < 0)
 782                                         goto out_finish;
 783
 784                                 if (!ret)
 785                                         break;
 786
 787                                 memcpy(data[thr].unc + off,
 788                                        data_of(*snapshot), PAGE_SIZE);
 789
 790                                 if (!(nr_pages % m))
 791                                         pr_info("Image saving progress: %3d%%\n",
 792                                                 nr_pages / m * 10);
 793                                 nr_pages++;
 794                         }
 795                         if (!off)
 796                                 break;
 797
 798                         data[thr].unc_len = off;
 799
 800                         atomic_set(&data[thr].ready, 1);
 801                         wake_up(&data[thr].go);
 802                 }
 803
 804                 if (!thr)
 805                         break;
 806
 807                 crc->run_threads = thr;
 808                 atomic_set(&crc->ready, 1);
 809                 wake_up(&crc->go);
 810
 811                 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
 812                         wait_event(data[thr].done,
 813                                    atomic_read(&data[thr].stop));
 814                         atomic_set(&data[thr].stop, 0);
 815
 816                         ret = data[thr].ret;
 817
 818                         if (ret < 0) {
 819                                 pr_err("LZO compression failed\n");
 820                                 goto out_finish;
 821                         }
 822
 823                         if (unlikely(!data[thr].cmp_len ||
 824                                      data[thr].cmp_len >
 825                                      lzo1x_worst_compress(data[thr].unc_len))) {
 826                                 pr_err("Invalid LZO compressed length\n");
 827                                 ret = -1;
 828                                 goto out_finish;
 829                         }
 830
 831                         *(size_t *)data[thr].cmp = data[thr].cmp_len;
 832
 833                         /*
 834                          * Given we are writing one page at a time to disk, we
 835                          * copy that much from the buffer, although the last
 836                          * bit will likely be smaller than full page. This is
 837                          * OK - we saved the length of the compressed data, so
 838                          * any garbage at the end will be discarded when we
 839                          * read it.
 840                          */
 841                         for (off = 0;
 842                              off < LZO_HEADER + data[thr].cmp_len;
 843                              off += PAGE_SIZE) {
 844                                 memcpy(page, data[thr].cmp + off, PAGE_SIZE);
 845
 846                                 ret = swap_write_page(handle, page, &hb);
 847                                 if (ret)
 848                                         goto out_finish;
 849                         }
 850                 }
 851
 852                 wait_event(crc->done, atomic_read(&crc->stop));
 853                 atomic_set(&crc->stop, 0);
 854         }
 855
 856 out_finish:
 857         err2 = hib_wait_io(&hb);
 858         stop = ktime_get();
 859         if (!ret)
 860                 ret = err2;
 861         if (!ret)
 862                 pr_info("Image saving done\n");
 863         swsusp_show_speed(start, stop, nr_to_write, "Wrote");
 864 out_clean:
 865         hib_finish_batch(&hb);
 866         if (crc) {
 867                 if (crc->thr)
 868                         kthread_stop(crc->thr);
 869                 kfree(crc);
 870         }
 871         if (data) {
 872                 for (thr = 0; thr < nr_threads; thr++)
 873                         if (data[thr].thr)
 874                                 kthread_stop(data[thr].thr);
 875                 vfree(data);
 876         }
 877         if (page) free_page((unsigned long)page);
 878
 879         return ret;
 880 }
 881
 882 /**
 883  *      enough_swap - Make sure we have enough swap to save the image.
 884  *
 885  *      Returns TRUE or FALSE after checking the total amount of swap
 886  *      space available from the resume partition.
 887  */
 888
 889 static int enough_swap(unsigned int nr_pages)
 890 {
 891         unsigned int free_swap = count_swap_pages(root_swap, 1);
 892         unsigned int required;
 893
 894         pr_debug("Free swap pages: %u\n", free_swap);
 895
 896         required = PAGES_FOR_IO + nr_pages;
 897         return free_swap > required;
 898 }
 899
 900 /**
 901  *      swsusp_write - Write entire image and metadata.
 902  *      @flags: flags to pass to the "boot" kernel in the image header
 903  *
 904  *      It is important _NOT_ to umount filesystems at this point. We want
 905  *      them synced (in case something goes wrong) but we DO not want to mark
 906  *      filesystem clean: it is not. (And it does not matter, if we resume
 907  *      correctly, we'll mark system clean, anyway.)
 908  */
 909
 910 int swsusp_write(unsigned int flags)
 911 {
 912         struct swap_map_handle handle;
 913         struct snapshot_handle snapshot;
 914         struct swsusp_info *header;
 915         unsigned long pages;
 916         int error;
 917
 918         pages = snapshot_get_image_size();
 919         error = get_swap_writer(&handle);
 920         if (error) {
 921                 pr_err("Cannot get swap writer\n");
 922                 return error;
 923         }
 924         if (flags & SF_NOCOMPRESS_MODE) {
 925                 if (!enough_swap(pages)) {
 926                         pr_err("Not enough free swap\n");
 927                         error = -ENOSPC;
 928                         goto out_finish;
 929                 }
 930         }
 931         memset(&snapshot, 0, sizeof(struct snapshot_handle));
 932         error = snapshot_read_next(&snapshot);
 933         if (error < (int)PAGE_SIZE) {
 934                 if (error >= 0)
 935                         error = -EFAULT;
 936
 937                 goto out_finish;
 938         }
 939         header = (struct swsusp_info *)data_of(snapshot);
 940         error = swap_write_page(&handle, header, NULL);
 941         if (!error) {
 942                 error = (flags & SF_NOCOMPRESS_MODE) ?
 943                         save_image(&handle, &snapshot, pages - 1) :
 944                         save_image_lzo(&handle, &snapshot, pages - 1);
 945         }
 946 out_finish:
 947         error = swap_writer_finish(&handle, flags, error);
 948         return error;
 949 }
 950
 951 /**
 952  *      The following functions allow us to read data using a swap map
 953  *      in a file-alike way
 954  */
 955
 956 static void release_swap_reader(struct swap_map_handle *handle)
 957 {
 958         struct swap_map_page_list *tmp;
 959
 960         while (handle->maps) {
 961                 if (handle->maps->map)
 962                         free_page((unsigned long)handle->maps->map);
 963                 tmp = handle->maps;
 964                 handle->maps = handle->maps->next;
 965                 kfree(tmp);
 966         }
 967         handle->cur = NULL;
 968 }
 969
 970 static int get_swap_reader(struct swap_map_handle *handle,
 971                 unsigned int *flags_p)
 972 {
 973         int error;
 974         struct swap_map_page_list *tmp, *last;
 975         sector_t offset;
 976
 977         *flags_p = swsusp_header->flags;
 978
 979         if (!swsusp_header->image) /* how can this happen? */
 980                 return -EINVAL;
 981
 982         handle->cur = NULL;
 983         last = handle->maps = NULL;
 984         offset = swsusp_header->image;
 985         while (offset) {
 986                 tmp = kzalloc(sizeof(*handle->maps), GFP_KERNEL);
 987                 if (!tmp) {
 988                         release_swap_reader(handle);
 989                         return -ENOMEM;
 990                 }
 991                 if (!handle->maps)
 992                         handle->maps = tmp;
 993                 if (last)
 994                         last->next = tmp;
 995                 last = tmp;
 996
 997                 tmp->map = (struct swap_map_page *)
 998                            __get_free_page(GFP_NOIO | __GFP_HIGH);
 999                 if (!tmp->map) {
1000                         release_swap_reader(handle);
1001                         return -ENOMEM;
1002                 }
1003
1004                 error = hib_submit_io(REQ_OP_READ, offset, tmp->map, NULL);
1005                 if (error) {
1006                         release_swap_reader(handle);
1007                         return error;
1008                 }
1009                 offset = tmp->map->next_swap;
1010         }
1011         handle->k = 0;
1012         handle->cur = handle->maps->map;
1013         return 0;
1014 }
1015
1016 static int swap_read_page(struct swap_map_handle *handle, void *buf,
1017                 struct hib_bio_batch *hb)
1018 {
1019         sector_t offset;
1020         int error;
1021         struct swap_map_page_list *tmp;
1022
1023         if (!handle->cur)
1024                 return -EINVAL;
1025         offset = handle->cur->entries[handle->k];
1026         if (!offset)
1027                 return -EFAULT;
1028         error = hib_submit_io(REQ_OP_READ, offset, buf, hb);
1029         if (error)
1030                 return error;
1031         if (++handle->k >= MAP_PAGE_ENTRIES) {
1032                 handle->k = 0;
1033                 free_page((unsigned long)handle->maps->map);
1034                 tmp = handle->maps;
1035                 handle->maps = handle->maps->next;
1036                 kfree(tmp);
1037                 if (!handle->maps)
1038                         release_swap_reader(handle);
1039                 else
1040                         handle->cur = handle->maps->map;
1041         }
1042         return error;
1043 }
1044
1045 static int swap_reader_finish(struct swap_map_handle *handle)
1046 {
1047         release_swap_reader(handle);
1048
1049         return 0;
1050 }
1051
1052 /**
1053  *      load_image - load the image using the swap map handle
1054  *      @handle and the snapshot handle @snapshot
1055  *      (assume there are @nr_pages pages to load)
1056  */
1057
1058 static int load_image(struct swap_map_handle *handle,
1059                       struct snapshot_handle *snapshot,
1060                       unsigned int nr_to_read)
1061 {
1062         unsigned int m;
1063         int ret = 0;
1064         ktime_t start;
1065         ktime_t stop;
1066         struct hib_bio_batch hb;
1067         int err2;
1068         unsigned nr_pages;
1069
1070         hib_init_batch(&hb);
1071
1072         clean_pages_on_read = true;
1073         pr_info("Loading image data pages (%u pages)...\n", nr_to_read);
1074         m = nr_to_read / 10;
1075         if (!m)
1076                 m = 1;
1077         nr_pages = 0;
1078         start = ktime_get();
1079         for ( ; ; ) {
1080                 ret = snapshot_write_next(snapshot);
1081                 if (ret <= 0)
1082                         break;
1083                 ret = swap_read_page(handle, data_of(*snapshot), &hb);
1084                 if (ret)
1085                         break;
1086                 if (snapshot->sync_read)
1087                         ret = hib_wait_io(&hb);
1088                 if (ret)
1089                         break;
1090                 if (!(nr_pages % m))
1091                         pr_info("Image loading progress: %3d%%\n",
1092                                 nr_pages / m * 10);
1093                 nr_pages++;
1094         }
1095         err2 = hib_wait_io(&hb);
1096         hib_finish_batch(&hb);
1097         stop = ktime_get();
1098         if (!ret)
1099                 ret = err2;
1100         if (!ret) {
1101                 pr_info("Image loading done\n");
1102                 snapshot_write_finalize(snapshot);
1103                 if (!snapshot_image_loaded(snapshot))
1104                         ret = -ENODATA;
1105         }
1106         swsusp_show_speed(start, stop, nr_to_read, "Read");
1107         return ret;
1108 }
1109
1110 /**
1111  * Structure used for LZO data decompression.
1112  */
1113 struct dec_data {
1114         struct task_struct *thr;                  /* thread */
1115         atomic_t ready;                           /* ready to start flag */
1116         atomic_t stop;                            /* ready to stop flag */
1117         int ret;                                  /* return code */
1118         wait_queue_head_t go;                     /* start decompression */
1119         wait_queue_head_t done;                   /* decompression done */
1120         size_t unc_len;                           /* uncompressed length */
1121         size_t cmp_len;                           /* compressed length */
1122         unsigned char unc[LZO_UNC_SIZE];          /* uncompressed buffer */
1123         unsigned char cmp[LZO_CMP_SIZE];          /* compressed buffer */
1124 };
1125
1126 /**
1127  * Decompression function that runs in its own thread.
1128  */
1129 static int lzo_decompress_threadfn(void *data)
1130 {
1131         struct dec_data *d = data;
1132
1133         while (1) {
1134                 wait_event(d->go, atomic_read(&d->ready) ||
1135                                   kthread_should_stop());
1136                 if (kthread_should_stop()) {
1137                         d->thr = NULL;
1138                         d->ret = -1;
1139                         atomic_set(&d->stop, 1);
1140                         wake_up(&d->done);
1141                         break;
1142                 }
1143                 atomic_set(&d->ready, 0);
1144
1145                 d->unc_len = LZO_UNC_SIZE;
1146                 d->ret = lzo1x_decompress_safe(d->cmp + LZO_HEADER, d->cmp_len,
1147                                                d->unc, &d->unc_len);
1148                 if (clean_pages_on_decompress)
1149                         flush_icache_range((unsigned long)d->unc,
1150                                            (unsigned long)d->unc + d->unc_len);
1151
1152                 atomic_set(&d->stop, 1);
1153                 wake_up(&d->done);
1154         }
1155         return 0;
1156 }
1157
1158 /**
1159  * load_image_lzo - Load compressed image data and decompress them with LZO.
1160  * @handle: Swap map handle to use for loading data.
1161  * @snapshot: Image to copy uncompressed data into.
1162  * @nr_to_read: Number of pages to load.
1163  */
1164 static int load_image_lzo(struct swap_map_handle *handle,
1165                           struct snapshot_handle *snapshot,
1166                           unsigned int nr_to_read)
1167 {
1168         unsigned int m;
1169         int ret = 0;
1170         int eof = 0;
1171         struct hib_bio_batch hb;
1172         ktime_t start;
1173         ktime_t stop;
1174         unsigned nr_pages;
1175         size_t off;
1176         unsigned i, thr, run_threads, nr_threads;
1177         unsigned ring = 0, pg = 0, ring_size = 0,
1178                  have = 0, want, need, asked = 0;
1179         unsigned long read_pages = 0;
1180         unsigned char **page = NULL;
1181         struct dec_data *data = NULL;
1182         struct crc_data *crc = NULL;
1183
1184         hib_init_batch(&hb);
1185
1186         /*
1187          * We'll limit the number of threads for decompression to limit memory
1188          * footprint.
1189          */
1190         nr_threads = num_online_cpus() - 1;
1191         nr_threads = clamp_val(nr_threads, 1, LZO_THREADS);
1192
1193         page = vmalloc(array_size(LZO_MAX_RD_PAGES, sizeof(*page)));
1194         if (!page) {
1195                 pr_err("Failed to allocate LZO page\n");
1196                 ret = -ENOMEM;
1197                 goto out_clean;
1198         }
1199
1200         data = vzalloc(array_size(nr_threads, sizeof(*data)));
1201         if (!data) {
1202                 pr_err("Failed to allocate LZO data\n");
1203                 ret = -ENOMEM;
1204                 goto out_clean;
1205         }
1206
1207         crc = kzalloc(sizeof(*crc), GFP_KERNEL);
1208         if (!crc) {
1209                 pr_err("Failed to allocate crc\n");
1210                 ret = -ENOMEM;
1211                 goto out_clean;
1212         }
1213
1214         clean_pages_on_decompress = true;
1215
1216         /*
1217          * Start the decompression threads.
1218          */
1219         for (thr = 0; thr < nr_threads; thr++) {
1220                 init_waitqueue_head(&data[thr].go);
1221                 init_waitqueue_head(&data[thr].done);
1222
1223                 data[thr].thr = kthread_run(lzo_decompress_threadfn,
1224                                             &data[thr],
1225                                             "image_decompress/%u", thr);
1226                 if (IS_ERR(data[thr].thr)) {
1227                         data[thr].thr = NULL;
1228                         pr_err("Cannot start decompression threads\n");
1229                         ret = -ENOMEM;
1230                         goto out_clean;
1231                 }
1232         }
1233
1234         /*
1235          * Start the CRC32 thread.
1236          */
1237         init_waitqueue_head(&crc->go);
1238         init_waitqueue_head(&crc->done);
1239
1240         handle->crc32 = 0;
1241         crc->crc32 = &handle->crc32;
1242         for (thr = 0; thr < nr_threads; thr++) {
1243                 crc->unc[thr] = data[thr].unc;
1244                 crc->unc_len[thr] = &data[thr].unc_len;
1245         }
1246
1247         crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
1248         if (IS_ERR(crc->thr)) {
1249                 crc->thr = NULL;
1250                 pr_err("Cannot start CRC32 thread\n");
1251                 ret = -ENOMEM;
1252                 goto out_clean;
1253         }
1254
1255         /*
1256          * Set the number of pages for read buffering.
1257          * This is complete guesswork, because we'll only know the real
1258          * picture once prepare_image() is called, which is much later on
1259          * during the image load phase. We'll assume the worst case and
1260          * say that none of the image pages are from high memory.
1261          */
1262         if (low_free_pages() > snapshot_get_image_size())
1263                 read_pages = (low_free_pages() - snapshot_get_image_size()) / 2;
1264         read_pages = clamp_val(read_pages, LZO_MIN_RD_PAGES, LZO_MAX_RD_PAGES);
1265
1266         for (i = 0; i < read_pages; i++) {
1267                 page[i] = (void *)__get_free_page(i < LZO_CMP_PAGES ?
1268                                                   GFP_NOIO | __GFP_HIGH :
1269                                                   GFP_NOIO | __GFP_NOWARN |
1270                                                   __GFP_NORETRY);
1271
1272                 if (!page[i]) {
1273                         if (i < LZO_CMP_PAGES) {
1274                                 ring_size = i;
1275                                 pr_err("Failed to allocate LZO pages\n");
1276                                 ret = -ENOMEM;
1277                                 goto out_clean;
1278                         } else {
1279                                 break;
1280                         }
1281                 }
1282         }
1283         want = ring_size = i;
1284
1285         pr_info("Using %u thread(s) for decompression\n", nr_threads);
1286         pr_info("Loading and decompressing image data (%u pages)...\n",
1287                 nr_to_read);
1288         m = nr_to_read / 10;
1289         if (!m)
1290                 m = 1;
1291         nr_pages = 0;
1292         start = ktime_get();
1293
1294         ret = snapshot_write_next(snapshot);
1295         if (ret <= 0)
1296                 goto out_finish;
1297
1298         for(;;) {
1299                 for (i = 0; !eof && i < want; i++) {
1300                         ret = swap_read_page(handle, page[ring], &hb);
1301                         if (ret) {
1302                                 /*
1303                                  * On real read error, finish. On end of data,
1304                                  * set EOF flag and just exit the read loop.
1305                                  */
1306                                 if (handle->cur &&
1307                                     handle->cur->entries[handle->k]) {
1308                                         goto out_finish;
1309                                 } else {
1310                                         eof = 1;
1311                                         break;
1312                                 }
1313                         }
1314                         if (++ring >= ring_size)
1315                                 ring = 0;
1316                 }
1317                 asked += i;
1318                 want -= i;
1319
1320                 /*
1321                  * We are out of data, wait for some more.
1322                  */
1323                 if (!have) {
1324                         if (!asked)
1325                                 break;
1326
1327                         ret = hib_wait_io(&hb);
1328                         if (ret)
1329                                 goto out_finish;
1330                         have += asked;
1331                         asked = 0;
1332                         if (eof)
1333                                 eof = 2;
1334                 }
1335
1336                 if (crc->run_threads) {
1337                         wait_event(crc->done, atomic_read(&crc->stop));
1338                         atomic_set(&crc->stop, 0);
1339                         crc->run_threads = 0;
1340                 }
1341
1342                 for (thr = 0; have && thr < nr_threads; thr++) {
1343                         data[thr].cmp_len = *(size_t *)page[pg];
1344                         if (unlikely(!data[thr].cmp_len ||
1345                                      data[thr].cmp_len >
1346                                      lzo1x_worst_compress(LZO_UNC_SIZE))) {
1347                                 pr_err("Invalid LZO compressed length\n");
1348                                 ret = -1;
1349                                 goto out_finish;
1350                         }
1351
1352                         need = DIV_ROUND_UP(data[thr].cmp_len + LZO_HEADER,
1353                                             PAGE_SIZE);
1354                         if (need > have) {
1355                                 if (eof > 1) {
1356                                         ret = -1;
1357                                         goto out_finish;
1358                                 }
1359                                 break;
1360                         }
1361
1362                         for (off = 0;
1363                              off < LZO_HEADER + data[thr].cmp_len;
1364                              off += PAGE_SIZE) {
1365                                 memcpy(data[thr].cmp + off,
1366                                        page[pg], PAGE_SIZE);
1367                                 have--;
1368                                 want++;
1369                                 if (++pg >= ring_size)
1370                                         pg = 0;
1371                         }
1372
1373                         atomic_set(&data[thr].ready, 1);
1374                         wake_up(&data[thr].go);
1375                 }
1376
1377                 /*
1378                  * Wait for more data while we are decompressing.
1379                  */
1380                 if (have < LZO_CMP_PAGES && asked) {
1381                         ret = hib_wait_io(&hb);
1382                         if (ret)
1383                                 goto out_finish;
1384                         have += asked;
1385                         asked = 0;
1386                         if (eof)
1387                                 eof = 2;
1388                 }
1389
1390                 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
1391                         wait_event(data[thr].done,
1392                                    atomic_read(&data[thr].stop));
1393                         atomic_set(&data[thr].stop, 0);
1394
1395                         ret = data[thr].ret;
1396
1397                         if (ret < 0) {
1398                                 pr_err("LZO decompression failed\n");
1399                                 goto out_finish;
1400                         }
1401
1402                         if (unlikely(!data[thr].unc_len ||
1403                                      data[thr].unc_len > LZO_UNC_SIZE ||
1404                                      data[thr].unc_len & (PAGE_SIZE - 1))) {
1405                                 pr_err("Invalid LZO uncompressed length\n");
1406                                 ret = -1;
1407                                 goto out_finish;
1408                         }
1409
1410                         for (off = 0;
1411                              off < data[thr].unc_len; off += PAGE_SIZE) {
1412                                 memcpy(data_of(*snapshot),
1413                                        data[thr].unc + off, PAGE_SIZE);
1414
1415                                 if (!(nr_pages % m))
1416                                         pr_info("Image loading progress: %3d%%\n",
1417                                                 nr_pages / m * 10);
1418                                 nr_pages++;
1419
1420                                 ret = snapshot_write_next(snapshot);
1421                                 if (ret <= 0) {
1422                                         crc->run_threads = thr + 1;
1423                                         atomic_set(&crc->ready, 1);
1424                                         wake_up(&crc->go);
1425                                         goto out_finish;
1426                                 }
1427                         }
1428                 }
1429
1430                 crc->run_threads = thr;
1431                 atomic_set(&crc->ready, 1);
1432                 wake_up(&crc->go);
1433         }
1434
1435 out_finish:
1436         if (crc->run_threads) {
1437                 wait_event(crc->done, atomic_read(&crc->stop));
1438                 atomic_set(&crc->stop, 0);
1439         }
1440         stop = ktime_get();
1441         if (!ret) {
1442                 pr_info("Image loading done\n");
1443                 snapshot_write_finalize(snapshot);
1444                 if (!snapshot_image_loaded(snapshot))
1445                         ret = -ENODATA;
1446                 if (!ret) {
1447                         if (swsusp_header->flags & SF_CRC32_MODE) {
1448                                 if(handle->crc32 != swsusp_header->crc32) {
1449                                         pr_err("Invalid image CRC32!\n");
1450                                         ret = -ENODATA;
1451                                 }
1452                         }
1453                 }
1454         }
1455         swsusp_show_speed(start, stop, nr_to_read, "Read");
1456 out_clean:
1457         hib_finish_batch(&hb);
1458         for (i = 0; i < ring_size; i++)
1459                 free_page((unsigned long)page[i]);
1460         if (crc) {
1461                 if (crc->thr)
1462                         kthread_stop(crc->thr);
1463                 kfree(crc);
1464         }
1465         if (data) {
1466                 for (thr = 0; thr < nr_threads; thr++)
1467                         if (data[thr].thr)
1468                                 kthread_stop(data[thr].thr);
1469                 vfree(data);
1470         }
1471         vfree(page);
1472
1473         return ret;
1474 }
1475
1476 /**
1477  *      swsusp_read - read the hibernation image.
1478  *      @flags_p: flags passed by the "frozen" kernel in the image header should
1479  *                be written into this memory location
1480  */
1481
1482 int swsusp_read(unsigned int *flags_p)
1483 {
1484         int error;
1485         struct swap_map_handle handle;
1486         struct snapshot_handle snapshot;
1487         struct swsusp_info *header;
1488
1489         memset(&snapshot, 0, sizeof(struct snapshot_handle));
1490         error = snapshot_write_next(&snapshot);
1491         if (error < (int)PAGE_SIZE)
1492                 return error < 0 ? error : -EFAULT;
1493         header = (struct swsusp_info *)data_of(snapshot);
1494         error = get_swap_reader(&handle, flags_p);
1495         if (error)
1496                 goto end;
1497         if (!error)
1498                 error = swap_read_page(&handle, header, NULL);
1499         if (!error) {
1500                 error = (*flags_p & SF_NOCOMPRESS_MODE) ?
1501                         load_image(&handle, &snapshot, header->pages - 1) :
1502                         load_image_lzo(&handle, &snapshot, header->pages - 1);
1503         }
1504         swap_reader_finish(&handle);
1505 end:
1506         if (!error)
1507                 pr_debug("Image successfully loaded\n");
1508         else
1509                 pr_debug("Error %d resuming\n", error);
1510         return error;
1511 }
1512
1513 /**
1514  *      swsusp_check - Check for swsusp signature in the resume device
1515  */
1516
1517 int swsusp_check(void)
1518 {
1519         int error;
1520         void *holder;
1521
1522         hib_resume_bdev = blkdev_get_by_dev(swsusp_resume_device,
1523                                             FMODE_READ | FMODE_EXCL, &holder);
1524         if (!IS_ERR(hib_resume_bdev)) {
1525                 set_blocksize(hib_resume_bdev, PAGE_SIZE);
1526                 clear_page(swsusp_header);
1527                 error = hib_submit_io(REQ_OP_READ, swsusp_resume_block,
1528                                         swsusp_header, NULL);
1529                 if (error)
1530                         goto put;
1531
1532                 if (!memcmp(HIBERNATE_SIG, swsusp_header->sig, 10)) {
1533                         memcpy(swsusp_header->sig, swsusp_header->orig_sig, 10);
1534                         /* Reset swap signature now */
1535                         error = hib_submit_io(REQ_OP_WRITE | REQ_SYNC,
1536                                                 swsusp_resume_block,
1537                                                 swsusp_header, NULL);
1538                 } else {
1539                         error = -EINVAL;
1540                 }
1541                 if (!error && swsusp_header->flags & SF_HW_SIG &&
1542                     swsusp_header->hw_sig != swsusp_hardware_signature) {
1543                         pr_info("Suspend image hardware signature mismatch (%08x now %08x); aborting resume.\n",
1544                                 swsusp_header->hw_sig, swsusp_hardware_signature);
1545                         error = -EINVAL;
1546                 }
1547
1548 put:
1549                 if (error)
1550                         blkdev_put(hib_resume_bdev, FMODE_READ | FMODE_EXCL);
1551                 else
1552                         pr_debug("Image signature found, resuming\n");
1553         } else {
1554                 error = PTR_ERR(hib_resume_bdev);
1555         }
1556
1557         if (error)
1558                 pr_debug("Image not found (code %d)\n", error);
1559
1560         return error;
1561 }
1562
1563 /**
1564  *      swsusp_close - close swap device.
1565  */
1566
1567 void swsusp_close(fmode_t mode)
1568 {
1569         if (IS_ERR(hib_resume_bdev)) {
1570                 pr_debug("Image device not initialised\n");
1571                 return;
1572         }
1573
1574         blkdev_put(hib_resume_bdev, mode);
1575 }
1576
1577 /**
1578  *      swsusp_unmark - Unmark swsusp signature in the resume device
1579  */
1580
1581 #ifdef CONFIG_SUSPEND
1582 int swsusp_unmark(void)
1583 {
1584         int error;
1585
1586         hib_submit_io(REQ_OP_READ, swsusp_resume_block,
1587                         swsusp_header, NULL);
1588         if (!memcmp(HIBERNATE_SIG,swsusp_header->sig, 10)) {
1589                 memcpy(swsusp_header->sig,swsusp_header->orig_sig, 10);
1590                 error = hib_submit_io(REQ_OP_WRITE | REQ_SYNC,
1591                                         swsusp_resume_block,
1592                                         swsusp_header, NULL);
1593         } else {
1594                 pr_err("Cannot find swsusp signature!\n");
1595                 error = -ENODEV;
1596         }
1597
1598         /*
1599          * We just returned from suspend, we don't need the image any more.
1600          */
1601         free_all_swap_pages(root_swap);
1602
1603         return error;
1604 }
1605 #endif
1606
1607 static int __init swsusp_header_init(void)
1608 {
1609         swsusp_header = (struct swsusp_header*) __get_free_page(GFP_KERNEL);
1610         if (!swsusp_header)
1611                 panic("Could not allocate memory for swsusp_header\n");
1612         return 0;
1613 }
1614
1615 core_initcall(swsusp_header_init);