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 <[email protected]>
   9  * Copyright (C) 2006 Rafael J. Wysocki <[email protected]>
  10  * Copyright (C) 2010-2012 Bojan Smojver <[email protected]>
  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/vmalloc.h>
  27 #include <linux/cpumask.h>
  28 #include <linux/atomic.h>
  29 #include <linux/kthread.h>
  30 #include <linux/crc32.h>
  31 #include <linux/ktime.h>
  32
  33 #include "power.h"
  34
  35 #define HIBERNATE_SIG   "S1SUSPEND"
  36
  37 u32 swsusp_hardware_signature;
  38
  39 /*
  40  * When reading an {un,}compressed image, we may restore pages in place,
  41  * in which case some architectures need these pages cleaning before they
  42  * can be executed. We don't know which pages these may be, so clean the lot.
  43  */
  44 static bool clean_pages_on_read;
  45 static bool clean_pages_on_decompress;
  46
  47 /*
  48  *      The swap map is a data structure used for keeping track of each page
  49  *      written to a swap partition.  It consists of many swap_map_page
  50  *      structures that contain each an array of MAP_PAGE_ENTRIES swap entries.
  51  *      These structures are stored on the swap and linked together with the
  52  *      help of the .next_swap member.
  53  *
  54  *      The swap map is created during suspend.  The swap map pages are
  55  *      allocated and populated one at a time, so we only need one memory
  56  *      page to set up the entire structure.
  57  *
  58  *      During resume we pick up all swap_map_page structures into a list.
  59  */
  60
  61 #define MAP_PAGE_ENTRIES        (PAGE_SIZE / sizeof(sector_t) - 1)
  62
  63 /*
  64  * Number of free pages that are not high.
  65  */
  66 static inline unsigned long low_free_pages(void)
  67 {
  68         return nr_free_pages() - nr_free_highpages();
  69 }
  70
  71 /*
  72  * Number of pages required to be kept free while writing the image. Always
  73  * half of all available low pages before the writing starts.
  74  */
  75 static inline unsigned long reqd_free_pages(void)
  76 {
  77         return low_free_pages() / 2;
  78 }
  79
  80 struct swap_map_page {
  81         sector_t entries[MAP_PAGE_ENTRIES];
  82         sector_t next_swap;
  83 };
  84
  85 struct swap_map_page_list {
  86         struct swap_map_page *map;
  87         struct swap_map_page_list *next;
  88 };
  89
  90 /*
  91  *      The swap_map_handle structure is used for handling swap in
  92  *      a file-alike way
  93  */
  94
  95 struct swap_map_handle {
  96         struct swap_map_page *cur;
  97         struct swap_map_page_list *maps;
  98         sector_t cur_swap;
  99         sector_t first_sector;
 100         unsigned int k;
 101         unsigned long reqd_free_pages;
 102         u32 crc32;
 103 };
 104
 105 struct swsusp_header {
 106         char reserved[PAGE_SIZE - 20 - sizeof(sector_t) - sizeof(int) -
 107                       sizeof(u32) - sizeof(u32)];
 108         u32     hw_sig;
 109         u32     crc32;
 110         sector_t image;
 111         unsigned int flags;     /* Flags to pass to the "boot" kernel */
 112         char    orig_sig[10];
 113         char    sig[10];
 114 } __packed;
 115
 116 static struct swsusp_header *swsusp_header;
 117
 118 /*
 119  *      The following functions are used for tracing the allocated
 120  *      swap pages, so that they can be freed in case of an error.
 121  */
 122
 123 struct swsusp_extent {
 124         struct rb_node node;
 125         unsigned long start;
 126         unsigned long end;
 127 };
 128
 129 static struct rb_root swsusp_extents = RB_ROOT;
 130
 131 static int swsusp_extents_insert(unsigned long swap_offset)
 132 {
 133         struct rb_node **new = &(swsusp_extents.rb_node);
 134         struct rb_node *parent = NULL;
 135         struct swsusp_extent *ext;
 136
 137         /* Figure out where to put the new node */
 138         while (*new) {
 139                 ext = rb_entry(*new, struct swsusp_extent, node);
 140                 parent = *new;
 141                 if (swap_offset < ext->start) {
 142                         /* Try to merge */
 143                         if (swap_offset == ext->start - 1) {
 144                                 ext->start--;
 145                                 return 0;
 146                         }
 147                         new = &((*new)->rb_left);
 148                 } else if (swap_offset > ext->end) {
 149                         /* Try to merge */
 150                         if (swap_offset == ext->end + 1) {
 151                                 ext->end++;
 152                                 return 0;
 153                         }
 154                         new = &((*new)->rb_right);
 155                 } else {
 156                         /* It already is in the tree */
 157                         return -EINVAL;
 158                 }
 159         }
 160         /* Add the new node and rebalance the tree. */
 161         ext = kzalloc(sizeof(struct swsusp_extent), GFP_KERNEL);
 162         if (!ext)
 163                 return -ENOMEM;
 164
 165         ext->start = swap_offset;
 166         ext->end = swap_offset;
 167         rb_link_node(&ext->node, parent, new);
 168         rb_insert_color(&ext->node, &swsusp_extents);
 169         return 0;
 170 }
 171
 172 /*
 173  *      alloc_swapdev_block - allocate a swap page and register that it has
 174  *      been allocated, so that it can be freed in case of an error.
 175  */
 176
 177 sector_t alloc_swapdev_block(int swap)
 178 {
 179         unsigned long offset;
 180
 181         offset = swp_offset(get_swap_page_of_type(swap));
 182         if (offset) {
 183                 if (swsusp_extents_insert(offset))
 184                         swap_free(swp_entry(swap, offset));
 185                 else
 186                         return swapdev_block(swap, offset);
 187         }
 188         return 0;
 189 }
 190
 191 /*
 192  *      free_all_swap_pages - free swap pages allocated for saving image data.
 193  *      It also frees the extents used to register which swap entries had been
 194  *      allocated.
 195  */
 196
 197 void free_all_swap_pages(int swap)
 198 {
 199         struct rb_node *node;
 200
 201         while ((node = swsusp_extents.rb_node)) {
 202                 struct swsusp_extent *ext;
 203                 unsigned long offset;
 204
 205                 ext = rb_entry(node, struct swsusp_extent, node);
 206                 rb_erase(node, &swsusp_extents);
 207                 for (offset = ext->start; offset <= ext->end; offset++)
 208                         swap_free(swp_entry(swap, offset));
 209
 210                 kfree(ext);
 211         }
 212 }
 213
 214 int swsusp_swap_in_use(void)
 215 {
 216         return (swsusp_extents.rb_node != NULL);
 217 }
 218
 219 /*
 220  * General things
 221  */
 222
 223 static unsigned short root_swap = 0xffff;
 224 static struct file *hib_resume_bdev_file;
 225
 226 struct hib_bio_batch {
 227         atomic_t                count;
 228         wait_queue_head_t       wait;
 229         blk_status_t            error;
 230         struct blk_plug         plug;
 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         blk_start_plug(&hb->plug);
 239 }
 240
 241 static void hib_finish_batch(struct hib_bio_batch *hb)
 242 {
 243         blk_finish_plug(&hb->plug);
 244 }
 245
 246 static void hib_end_io(struct bio *bio)
 247 {
 248         struct hib_bio_batch *hb = bio->bi_private;
 249         struct page *page = bio_first_page_all(bio);
 250
 251         if (bio->bi_status) {
 252                 pr_alert("Read-error on swap-device (%u:%u:%Lu)\n",
 253                          MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
 254                          (unsigned long long)bio->bi_iter.bi_sector);
 255         }
 256
 257         if (bio_data_dir(bio) == WRITE)
 258                 put_page(page);
 259         else if (clean_pages_on_read)
 260                 flush_icache_range((unsigned long)page_address(page),
 261                                    (unsigned long)page_address(page) + PAGE_SIZE);
 262
 263         if (bio->bi_status && !hb->error)
 264                 hb->error = bio->bi_status;
 265         if (atomic_dec_and_test(&hb->count))
 266                 wake_up(&hb->wait);
 267
 268         bio_put(bio);
 269 }
 270
 271 static int hib_submit_io(blk_opf_t opf, pgoff_t page_off, void *addr,
 272                          struct hib_bio_batch *hb)
 273 {
 274         struct page *page = virt_to_page(addr);
 275         struct bio *bio;
 276         int error = 0;
 277
 278         bio = bio_alloc(file_bdev(hib_resume_bdev_file), 1, opf,
 279                         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  * Hold the swsusp_header flag. This is used in software_resume() in
 343  * 'kernel/power/hibernate' to check if the image is compressed and query
 344  * for the compression algorithm support(if so).
 345  */
 346 unsigned int swsusp_header_flags;
 347
 348 /**
 349  *      swsusp_swap_check - check if the resume device is a swap device
 350  *      and get its index (if so)
 351  *
 352  *      This is called before saving image
 353  */
 354 static int swsusp_swap_check(void)
 355 {
 356         int res;
 357
 358         if (swsusp_resume_device)
 359                 res = swap_type_of(swsusp_resume_device, swsusp_resume_block);
 360         else
 361                 res = find_first_swap(&swsusp_resume_device);
 362         if (res < 0)
 363                 return res;
 364         root_swap = res;
 365
 366         hib_resume_bdev_file = bdev_file_open_by_dev(swsusp_resume_device,
 367                         BLK_OPEN_WRITE, NULL, NULL);
 368         if (IS_ERR(hib_resume_bdev_file))
 369                 return PTR_ERR(hib_resume_bdev_file);
 370
 371         return 0;
 372 }
 373
 374 /**
 375  *      write_page - Write one page to given swap location.
 376  *      @buf:           Address we're writing.
 377  *      @offset:        Offset of the swap page we're writing to.
 378  *      @hb:            bio completion batch
 379  */
 380
 381 static int write_page(void *buf, sector_t offset, struct hib_bio_batch *hb)
 382 {
 383         void *src;
 384         int ret;
 385
 386         if (!offset)
 387                 return -ENOSPC;
 388
 389         if (hb) {
 390                 src = (void *)__get_free_page(GFP_NOIO | __GFP_NOWARN |
 391                                               __GFP_NORETRY);
 392                 if (src) {
 393                         copy_page(src, buf);
 394                 } else {
 395                         ret = hib_wait_io(hb); /* Free pages */
 396                         if (ret)
 397                                 return ret;
 398                         src = (void *)__get_free_page(GFP_NOIO |
 399                                                       __GFP_NOWARN |
 400                                                       __GFP_NORETRY);
 401                         if (src) {
 402                                 copy_page(src, buf);
 403                         } else {
 404                                 WARN_ON_ONCE(1);
 405                                 hb = NULL;      /* Go synchronous */
 406                                 src = buf;
 407                         }
 408                 }
 409         } else {
 410                 src = buf;
 411         }
 412         return hib_submit_io(REQ_OP_WRITE | REQ_SYNC, offset, src, hb);
 413 }
 414
 415 static void release_swap_writer(struct swap_map_handle *handle)
 416 {
 417         if (handle->cur)
 418                 free_page((unsigned long)handle->cur);
 419         handle->cur = NULL;
 420 }
 421
 422 static int get_swap_writer(struct swap_map_handle *handle)
 423 {
 424         int ret;
 425
 426         ret = swsusp_swap_check();
 427         if (ret) {
 428                 if (ret != -ENOSPC)
 429                         pr_err("Cannot find swap device, try swapon -a\n");
 430                 return ret;
 431         }
 432         handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_KERNEL);
 433         if (!handle->cur) {
 434                 ret = -ENOMEM;
 435                 goto err_close;
 436         }
 437         handle->cur_swap = alloc_swapdev_block(root_swap);
 438         if (!handle->cur_swap) {
 439                 ret = -ENOSPC;
 440                 goto err_rel;
 441         }
 442         handle->k = 0;
 443         handle->reqd_free_pages = reqd_free_pages();
 444         handle->first_sector = handle->cur_swap;
 445         return 0;
 446 err_rel:
 447         release_swap_writer(handle);
 448 err_close:
 449         swsusp_close();
 450         return ret;
 451 }
 452
 453 static int swap_write_page(struct swap_map_handle *handle, void *buf,
 454                 struct hib_bio_batch *hb)
 455 {
 456         int error;
 457         sector_t offset;
 458
 459         if (!handle->cur)
 460                 return -EINVAL;
 461         offset = alloc_swapdev_block(root_swap);
 462         error = write_page(buf, offset, hb);
 463         if (error)
 464                 return error;
 465         handle->cur->entries[handle->k++] = offset;
 466         if (handle->k >= MAP_PAGE_ENTRIES) {
 467                 offset = alloc_swapdev_block(root_swap);
 468                 if (!offset)
 469                         return -ENOSPC;
 470                 handle->cur->next_swap = offset;
 471                 error = write_page(handle->cur, handle->cur_swap, hb);
 472                 if (error)
 473                         goto out;
 474                 clear_page(handle->cur);
 475                 handle->cur_swap = offset;
 476                 handle->k = 0;
 477
 478                 if (hb && low_free_pages() <= handle->reqd_free_pages) {
 479                         error = hib_wait_io(hb);
 480                         if (error)
 481                                 goto out;
 482                         /*
 483                          * Recalculate the number of required free pages, to
 484                          * make sure we never take more than half.
 485                          */
 486                         handle->reqd_free_pages = reqd_free_pages();
 487                 }
 488         }
 489  out:
 490         return error;
 491 }
 492
 493 static int flush_swap_writer(struct swap_map_handle *handle)
 494 {
 495         if (handle->cur && handle->cur_swap)
 496                 return write_page(handle->cur, handle->cur_swap, NULL);
 497         else
 498                 return -EINVAL;
 499 }
 500
 501 static int swap_writer_finish(struct swap_map_handle *handle,
 502                 unsigned int flags, int error)
 503 {
 504         if (!error) {
 505                 pr_info("S");
 506                 error = mark_swapfiles(handle, flags);
 507                 pr_cont("|\n");
 508                 flush_swap_writer(handle);
 509         }
 510
 511         if (error)
 512                 free_all_swap_pages(root_swap);
 513         release_swap_writer(handle);
 514         swsusp_close();
 515
 516         return error;
 517 }
 518
 519 /*
 520  * Bytes we need for compressed data in worst case. We assume(limitation)
 521  * this is the worst of all the compression algorithms.
 522  */
 523 #define bytes_worst_compress(x) ((x) + ((x) / 16) + 64 + 3 + 2)
 524
 525 /* We need to remember how much compressed data we need to read. */
 526 #define CMP_HEADER      sizeof(size_t)
 527
 528 /* Number of pages/bytes we'll compress at one time. */
 529 #define UNC_PAGES       32
 530 #define UNC_SIZE        (UNC_PAGES * PAGE_SIZE)
 531
 532 /* Number of pages we need for compressed data (worst case). */
 533 #define CMP_PAGES       DIV_ROUND_UP(bytes_worst_compress(UNC_SIZE) + \
 534                                 CMP_HEADER, PAGE_SIZE)
 535 #define CMP_SIZE        (CMP_PAGES * PAGE_SIZE)
 536
 537 /* Maximum number of threads for compression/decompression. */
 538 #define CMP_THREADS     3
 539
 540 /* Minimum/maximum number of pages for read buffering. */
 541 #define CMP_MIN_RD_PAGES        1024
 542 #define CMP_MAX_RD_PAGES        8192
 543
 544 /**
 545  *      save_image - save the suspend image data
 546  */
 547
 548 static int save_image(struct swap_map_handle *handle,
 549                       struct snapshot_handle *snapshot,
 550                       unsigned int nr_to_write)
 551 {
 552         unsigned int m;
 553         int ret;
 554         int nr_pages;
 555         int err2;
 556         struct hib_bio_batch hb;
 557         ktime_t start;
 558         ktime_t stop;
 559
 560         hib_init_batch(&hb);
 561
 562         pr_info("Saving image data pages (%u pages)...\n",
 563                 nr_to_write);
 564         m = nr_to_write / 10;
 565         if (!m)
 566                 m = 1;
 567         nr_pages = 0;
 568         start = ktime_get();
 569         while (1) {
 570                 ret = snapshot_read_next(snapshot);
 571                 if (ret <= 0)
 572                         break;
 573                 ret = swap_write_page(handle, data_of(*snapshot), &hb);
 574                 if (ret)
 575                         break;
 576                 if (!(nr_pages % m))
 577                         pr_info("Image saving progress: %3d%%\n",
 578                                 nr_pages / m * 10);
 579                 nr_pages++;
 580         }
 581         err2 = hib_wait_io(&hb);
 582         hib_finish_batch(&hb);
 583         stop = ktime_get();
 584         if (!ret)
 585                 ret = err2;
 586         if (!ret)
 587                 pr_info("Image saving done\n");
 588         swsusp_show_speed(start, stop, nr_to_write, "Wrote");
 589         return ret;
 590 }
 591
 592 /*
 593  * Structure used for CRC32.
 594  */
 595 struct crc_data {
 596         struct task_struct *thr;                  /* thread */
 597         atomic_t ready;                           /* ready to start flag */
 598         atomic_t stop;                            /* ready to stop flag */
 599         unsigned run_threads;                     /* nr current threads */
 600         wait_queue_head_t go;                     /* start crc update */
 601         wait_queue_head_t done;                   /* crc update done */
 602         u32 *crc32;                               /* points to handle's crc32 */
 603         size_t *unc_len[CMP_THREADS];             /* uncompressed lengths */
 604         unsigned char *unc[CMP_THREADS];          /* uncompressed data */
 605 };
 606
 607 /*
 608  * CRC32 update function that runs in its own thread.
 609  */
 610 static int crc32_threadfn(void *data)
 611 {
 612         struct crc_data *d = data;
 613         unsigned i;
 614
 615         while (1) {
 616                 wait_event(d->go, atomic_read_acquire(&d->ready) ||
 617                                   kthread_should_stop());
 618                 if (kthread_should_stop()) {
 619                         d->thr = NULL;
 620                         atomic_set_release(&d->stop, 1);
 621                         wake_up(&d->done);
 622                         break;
 623                 }
 624                 atomic_set(&d->ready, 0);
 625
 626                 for (i = 0; i < d->run_threads; i++)
 627                         *d->crc32 = crc32_le(*d->crc32,
 628                                              d->unc[i], *d->unc_len[i]);
 629                 atomic_set_release(&d->stop, 1);
 630                 wake_up(&d->done);
 631         }
 632         return 0;
 633 }
 634 /*
 635  * Structure used for data compression.
 636  */
 637 struct cmp_data {
 638         struct task_struct *thr;                  /* thread */
 639         struct crypto_comp *cc;                   /* crypto compressor stream */
 640         atomic_t ready;                           /* ready to start flag */
 641         atomic_t stop;                            /* ready to stop flag */
 642         int ret;                                  /* return code */
 643         wait_queue_head_t go;                     /* start compression */
 644         wait_queue_head_t done;                   /* compression done */
 645         size_t unc_len;                           /* uncompressed length */
 646         size_t cmp_len;                           /* compressed length */
 647         unsigned char unc[UNC_SIZE];              /* uncompressed buffer */
 648         unsigned char cmp[CMP_SIZE];              /* compressed buffer */
 649 };
 650
 651 /* Indicates the image size after compression */
 652 static atomic_t compressed_size = ATOMIC_INIT(0);
 653
 654 /*
 655  * Compression function that runs in its own thread.
 656  */
 657 static int compress_threadfn(void *data)
 658 {
 659         struct cmp_data *d = data;
 660         unsigned int cmp_len = 0;
 661
 662         while (1) {
 663                 wait_event(d->go, atomic_read_acquire(&d->ready) ||
 664                                   kthread_should_stop());
 665                 if (kthread_should_stop()) {
 666                         d->thr = NULL;
 667                         d->ret = -1;
 668                         atomic_set_release(&d->stop, 1);
 669                         wake_up(&d->done);
 670                         break;
 671                 }
 672                 atomic_set(&d->ready, 0);
 673
 674                 cmp_len = CMP_SIZE - CMP_HEADER;
 675                 d->ret = crypto_comp_compress(d->cc, d->unc, d->unc_len,
 676                                               d->cmp + CMP_HEADER,
 677                                               &cmp_len);
 678                 d->cmp_len = cmp_len;
 679
 680                 atomic_set(&compressed_size, atomic_read(&compressed_size) + d->cmp_len);
 681                 atomic_set_release(&d->stop, 1);
 682                 wake_up(&d->done);
 683         }
 684         return 0;
 685 }
 686
 687 /**
 688  * save_compressed_image - Save the suspend image data after compression.
 689  * @handle: Swap map handle to use for saving the image.
 690  * @snapshot: Image to read data from.
 691  * @nr_to_write: Number of pages to save.
 692  */
 693 static int save_compressed_image(struct swap_map_handle *handle,
 694                                  struct snapshot_handle *snapshot,
 695                                  unsigned int nr_to_write)
 696 {
 697         unsigned int m;
 698         int ret = 0;
 699         int nr_pages;
 700         int err2;
 701         struct hib_bio_batch hb;
 702         ktime_t start;
 703         ktime_t stop;
 704         size_t off;
 705         unsigned thr, run_threads, nr_threads;
 706         unsigned char *page = NULL;
 707         struct cmp_data *data = NULL;
 708         struct crc_data *crc = NULL;
 709
 710         hib_init_batch(&hb);
 711
 712         atomic_set(&compressed_size, 0);
 713
 714         /*
 715          * We'll limit the number of threads for compression to limit memory
 716          * footprint.
 717          */
 718         nr_threads = num_online_cpus() - 1;
 719         nr_threads = clamp_val(nr_threads, 1, CMP_THREADS);
 720
 721         page = (void *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 722         if (!page) {
 723                 pr_err("Failed to allocate %s page\n", hib_comp_algo);
 724                 ret = -ENOMEM;
 725                 goto out_clean;
 726         }
 727
 728         data = vzalloc(array_size(nr_threads, sizeof(*data)));
 729         if (!data) {
 730                 pr_err("Failed to allocate %s data\n", hib_comp_algo);
 731                 ret = -ENOMEM;
 732                 goto out_clean;
 733         }
 734
 735         crc = kzalloc(sizeof(*crc), GFP_KERNEL);
 736         if (!crc) {
 737                 pr_err("Failed to allocate crc\n");
 738                 ret = -ENOMEM;
 739                 goto out_clean;
 740         }
 741
 742         /*
 743          * Start the compression threads.
 744          */
 745         for (thr = 0; thr < nr_threads; thr++) {
 746                 init_waitqueue_head(&data[thr].go);
 747                 init_waitqueue_head(&data[thr].done);
 748
 749                 data[thr].cc = crypto_alloc_comp(hib_comp_algo, 0, 0);
 750                 if (IS_ERR_OR_NULL(data[thr].cc)) {
 751                         pr_err("Could not allocate comp stream %ld\n", PTR_ERR(data[thr].cc));
 752                         ret = -EFAULT;
 753                         goto out_clean;
 754                 }
 755
 756                 data[thr].thr = kthread_run(compress_threadfn,
 757                                             &data[thr],
 758                                             "image_compress/%u", thr);
 759                 if (IS_ERR(data[thr].thr)) {
 760                         data[thr].thr = NULL;
 761                         pr_err("Cannot start compression threads\n");
 762                         ret = -ENOMEM;
 763                         goto out_clean;
 764                 }
 765         }
 766
 767         /*
 768          * Start the CRC32 thread.
 769          */
 770         init_waitqueue_head(&crc->go);
 771         init_waitqueue_head(&crc->done);
 772
 773         handle->crc32 = 0;
 774         crc->crc32 = &handle->crc32;
 775         for (thr = 0; thr < nr_threads; thr++) {
 776                 crc->unc[thr] = data[thr].unc;
 777                 crc->unc_len[thr] = &data[thr].unc_len;
 778         }
 779
 780         crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
 781         if (IS_ERR(crc->thr)) {
 782                 crc->thr = NULL;
 783                 pr_err("Cannot start CRC32 thread\n");
 784                 ret = -ENOMEM;
 785                 goto out_clean;
 786         }
 787
 788         /*
 789          * Adjust the number of required free pages after all allocations have
 790          * been done. We don't want to run out of pages when writing.
 791          */
 792         handle->reqd_free_pages = reqd_free_pages();
 793
 794         pr_info("Using %u thread(s) for %s compression\n", nr_threads, hib_comp_algo);
 795         pr_info("Compressing and saving image data (%u pages)...\n",
 796                 nr_to_write);
 797         m = nr_to_write / 10;
 798         if (!m)
 799                 m = 1;
 800         nr_pages = 0;
 801         start = ktime_get();
 802         for (;;) {
 803                 for (thr = 0; thr < nr_threads; thr++) {
 804                         for (off = 0; off < UNC_SIZE; off += PAGE_SIZE) {
 805                                 ret = snapshot_read_next(snapshot);
 806                                 if (ret < 0)
 807                                         goto out_finish;
 808
 809                                 if (!ret)
 810                                         break;
 811
 812                                 memcpy(data[thr].unc + off,
 813                                        data_of(*snapshot), PAGE_SIZE);
 814
 815                                 if (!(nr_pages % m))
 816                                         pr_info("Image saving progress: %3d%%\n",
 817                                                 nr_pages / m * 10);
 818                                 nr_pages++;
 819                         }
 820                         if (!off)
 821                                 break;
 822
 823                         data[thr].unc_len = off;
 824
 825                         atomic_set_release(&data[thr].ready, 1);
 826                         wake_up(&data[thr].go);
 827                 }
 828
 829                 if (!thr)
 830                         break;
 831
 832                 crc->run_threads = thr;
 833                 atomic_set_release(&crc->ready, 1);
 834                 wake_up(&crc->go);
 835
 836                 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
 837                         wait_event(data[thr].done,
 838                                 atomic_read_acquire(&data[thr].stop));
 839                         atomic_set(&data[thr].stop, 0);
 840
 841                         ret = data[thr].ret;
 842
 843                         if (ret < 0) {
 844                                 pr_err("%s compression failed\n", hib_comp_algo);
 845                                 goto out_finish;
 846                         }
 847
 848                         if (unlikely(!data[thr].cmp_len ||
 849                                      data[thr].cmp_len >
 850                                      bytes_worst_compress(data[thr].unc_len))) {
 851                                 pr_err("Invalid %s compressed length\n", hib_comp_algo);
 852                                 ret = -1;
 853                                 goto out_finish;
 854                         }
 855
 856                         *(size_t *)data[thr].cmp = data[thr].cmp_len;
 857
 858                         /*
 859                          * Given we are writing one page at a time to disk, we
 860                          * copy that much from the buffer, although the last
 861                          * bit will likely be smaller than full page. This is
 862                          * OK - we saved the length of the compressed data, so
 863                          * any garbage at the end will be discarded when we
 864                          * read it.
 865                          */
 866                         for (off = 0;
 867                              off < CMP_HEADER + data[thr].cmp_len;
 868                              off += PAGE_SIZE) {
 869                                 memcpy(page, data[thr].cmp + off, PAGE_SIZE);
 870
 871                                 ret = swap_write_page(handle, page, &hb);
 872                                 if (ret)
 873                                         goto out_finish;
 874                         }
 875                 }
 876
 877                 wait_event(crc->done, atomic_read_acquire(&crc->stop));
 878                 atomic_set(&crc->stop, 0);
 879         }
 880
 881 out_finish:
 882         err2 = hib_wait_io(&hb);
 883         stop = ktime_get();
 884         if (!ret)
 885                 ret = err2;
 886         if (!ret)
 887                 pr_info("Image saving done\n");
 888         swsusp_show_speed(start, stop, nr_to_write, "Wrote");
 889         pr_info("Image size after compression: %d kbytes\n",
 890                 (atomic_read(&compressed_size) / 1024));
 891
 892 out_clean:
 893         hib_finish_batch(&hb);
 894         if (crc) {
 895                 if (crc->thr)
 896                         kthread_stop(crc->thr);
 897                 kfree(crc);
 898         }
 899         if (data) {
 900                 for (thr = 0; thr < nr_threads; thr++) {
 901                         if (data[thr].thr)
 902                                 kthread_stop(data[thr].thr);
 903                         if (data[thr].cc)
 904                                 crypto_free_comp(data[thr].cc);
 905                 }
 906                 vfree(data);
 907         }
 908         if (page) free_page((unsigned long)page);
 909
 910         return ret;
 911 }
 912
 913 /**
 914  *      enough_swap - Make sure we have enough swap to save the image.
 915  *
 916  *      Returns TRUE or FALSE after checking the total amount of swap
 917  *      space available from the resume partition.
 918  */
 919
 920 static int enough_swap(unsigned int nr_pages)
 921 {
 922         unsigned int free_swap = count_swap_pages(root_swap, 1);
 923         unsigned int required;
 924
 925         pr_debug("Free swap pages: %u\n", free_swap);
 926
 927         required = PAGES_FOR_IO + nr_pages;
 928         return free_swap > required;
 929 }
 930
 931 /**
 932  *      swsusp_write - Write entire image and metadata.
 933  *      @flags: flags to pass to the "boot" kernel in the image header
 934  *
 935  *      It is important _NOT_ to umount filesystems at this point. We want
 936  *      them synced (in case something goes wrong) but we DO not want to mark
 937  *      filesystem clean: it is not. (And it does not matter, if we resume
 938  *      correctly, we'll mark system clean, anyway.)
 939  */
 940
 941 int swsusp_write(unsigned int flags)
 942 {
 943         struct swap_map_handle handle;
 944         struct snapshot_handle snapshot;
 945         struct swsusp_info *header;
 946         unsigned long pages;
 947         int error;
 948
 949         pages = snapshot_get_image_size();
 950         error = get_swap_writer(&handle);
 951         if (error) {
 952                 pr_err("Cannot get swap writer\n");
 953                 return error;
 954         }
 955         if (flags & SF_NOCOMPRESS_MODE) {
 956                 if (!enough_swap(pages)) {
 957                         pr_err("Not enough free swap\n");
 958                         error = -ENOSPC;
 959                         goto out_finish;
 960                 }
 961         }
 962         memset(&snapshot, 0, sizeof(struct snapshot_handle));
 963         error = snapshot_read_next(&snapshot);
 964         if (error < (int)PAGE_SIZE) {
 965                 if (error >= 0)
 966                         error = -EFAULT;
 967
 968                 goto out_finish;
 969         }
 970         header = (struct swsusp_info *)data_of(snapshot);
 971         error = swap_write_page(&handle, header, NULL);
 972         if (!error) {
 973                 error = (flags & SF_NOCOMPRESS_MODE) ?
 974                         save_image(&handle, &snapshot, pages - 1) :
 975                         save_compressed_image(&handle, &snapshot, pages - 1);
 976         }
 977 out_finish:
 978         error = swap_writer_finish(&handle, flags, error);
 979         return error;
 980 }
 981
 982 /*
 983  *      The following functions allow us to read data using a swap map
 984  *      in a file-like way.
 985  */
 986
 987 static void release_swap_reader(struct swap_map_handle *handle)
 988 {
 989         struct swap_map_page_list *tmp;
 990
 991         while (handle->maps) {
 992                 if (handle->maps->map)
 993                         free_page((unsigned long)handle->maps->map);
 994                 tmp = handle->maps;
 995                 handle->maps = handle->maps->next;
 996                 kfree(tmp);
 997         }
 998         handle->cur = NULL;
 999 }
1000
1001 static int get_swap_reader(struct swap_map_handle *handle,
1002                 unsigned int *flags_p)
1003 {
1004         int error;
1005         struct swap_map_page_list *tmp, *last;
1006         sector_t offset;
1007
1008         *flags_p = swsusp_header->flags;
1009
1010         if (!swsusp_header->image) /* how can this happen? */
1011                 return -EINVAL;
1012
1013         handle->cur = NULL;
1014         last = handle->maps = NULL;
1015         offset = swsusp_header->image;
1016         while (offset) {
1017                 tmp = kzalloc(sizeof(*handle->maps), GFP_KERNEL);
1018                 if (!tmp) {
1019                         release_swap_reader(handle);
1020                         return -ENOMEM;
1021                 }
1022                 if (!handle->maps)
1023                         handle->maps = tmp;
1024                 if (last)
1025                         last->next = tmp;
1026                 last = tmp;
1027
1028                 tmp->map = (struct swap_map_page *)
1029                            __get_free_page(GFP_NOIO | __GFP_HIGH);
1030                 if (!tmp->map) {
1031                         release_swap_reader(handle);
1032                         return -ENOMEM;
1033                 }
1034
1035                 error = hib_submit_io(REQ_OP_READ, offset, tmp->map, NULL);
1036                 if (error) {
1037                         release_swap_reader(handle);
1038                         return error;
1039                 }
1040                 offset = tmp->map->next_swap;
1041         }
1042         handle->k = 0;
1043         handle->cur = handle->maps->map;
1044         return 0;
1045 }
1046
1047 static int swap_read_page(struct swap_map_handle *handle, void *buf,
1048                 struct hib_bio_batch *hb)
1049 {
1050         sector_t offset;
1051         int error;
1052         struct swap_map_page_list *tmp;
1053
1054         if (!handle->cur)
1055                 return -EINVAL;
1056         offset = handle->cur->entries[handle->k];
1057         if (!offset)
1058                 return -EFAULT;
1059         error = hib_submit_io(REQ_OP_READ, offset, buf, hb);
1060         if (error)
1061                 return error;
1062         if (++handle->k >= MAP_PAGE_ENTRIES) {
1063                 handle->k = 0;
1064                 free_page((unsigned long)handle->maps->map);
1065                 tmp = handle->maps;
1066                 handle->maps = handle->maps->next;
1067                 kfree(tmp);
1068                 if (!handle->maps)
1069                         release_swap_reader(handle);
1070                 else
1071                         handle->cur = handle->maps->map;
1072         }
1073         return error;
1074 }
1075
1076 static int swap_reader_finish(struct swap_map_handle *handle)
1077 {
1078         release_swap_reader(handle);
1079
1080         return 0;
1081 }
1082
1083 /**
1084  *      load_image - load the image using the swap map handle
1085  *      @handle and the snapshot handle @snapshot
1086  *      (assume there are @nr_pages pages to load)
1087  */
1088
1089 static int load_image(struct swap_map_handle *handle,
1090                       struct snapshot_handle *snapshot,
1091                       unsigned int nr_to_read)
1092 {
1093         unsigned int m;
1094         int ret = 0;
1095         ktime_t start;
1096         ktime_t stop;
1097         struct hib_bio_batch hb;
1098         int err2;
1099         unsigned nr_pages;
1100
1101         hib_init_batch(&hb);
1102
1103         clean_pages_on_read = true;
1104         pr_info("Loading image data pages (%u pages)...\n", nr_to_read);
1105         m = nr_to_read / 10;
1106         if (!m)
1107                 m = 1;
1108         nr_pages = 0;
1109         start = ktime_get();
1110         for ( ; ; ) {
1111                 ret = snapshot_write_next(snapshot);
1112                 if (ret <= 0)
1113                         break;
1114                 ret = swap_read_page(handle, data_of(*snapshot), &hb);
1115                 if (ret)
1116                         break;
1117                 if (snapshot->sync_read)
1118                         ret = hib_wait_io(&hb);
1119                 if (ret)
1120                         break;
1121                 if (!(nr_pages % m))
1122                         pr_info("Image loading progress: %3d%%\n",
1123                                 nr_pages / m * 10);
1124                 nr_pages++;
1125         }
1126         err2 = hib_wait_io(&hb);
1127         hib_finish_batch(&hb);
1128         stop = ktime_get();
1129         if (!ret)
1130                 ret = err2;
1131         if (!ret) {
1132                 pr_info("Image loading done\n");
1133                 ret = snapshot_write_finalize(snapshot);
1134                 if (!ret && !snapshot_image_loaded(snapshot))
1135                         ret = -ENODATA;
1136         }
1137         swsusp_show_speed(start, stop, nr_to_read, "Read");
1138         return ret;
1139 }
1140
1141 /*
1142  * Structure used for data decompression.
1143  */
1144 struct dec_data {
1145         struct task_struct *thr;                  /* thread */
1146         struct crypto_comp *cc;                   /* crypto compressor stream */
1147         atomic_t ready;                           /* ready to start flag */
1148         atomic_t stop;                            /* ready to stop flag */
1149         int ret;                                  /* return code */
1150         wait_queue_head_t go;                     /* start decompression */
1151         wait_queue_head_t done;                   /* decompression done */
1152         size_t unc_len;                           /* uncompressed length */
1153         size_t cmp_len;                           /* compressed length */
1154         unsigned char unc[UNC_SIZE];              /* uncompressed buffer */
1155         unsigned char cmp[CMP_SIZE];              /* compressed buffer */
1156 };
1157
1158 /*
1159  * Decompression function that runs in its own thread.
1160  */
1161 static int decompress_threadfn(void *data)
1162 {
1163         struct dec_data *d = data;
1164         unsigned int unc_len = 0;
1165
1166         while (1) {
1167                 wait_event(d->go, atomic_read_acquire(&d->ready) ||
1168                                   kthread_should_stop());
1169                 if (kthread_should_stop()) {
1170                         d->thr = NULL;
1171                         d->ret = -1;
1172                         atomic_set_release(&d->stop, 1);
1173                         wake_up(&d->done);
1174                         break;
1175                 }
1176                 atomic_set(&d->ready, 0);
1177
1178                 unc_len = UNC_SIZE;
1179                 d->ret = crypto_comp_decompress(d->cc, d->cmp + CMP_HEADER, d->cmp_len,
1180                                                 d->unc, &unc_len);
1181                 d->unc_len = unc_len;
1182
1183                 if (clean_pages_on_decompress)
1184                         flush_icache_range((unsigned long)d->unc,
1185                                            (unsigned long)d->unc + d->unc_len);
1186
1187                 atomic_set_release(&d->stop, 1);
1188                 wake_up(&d->done);
1189         }
1190         return 0;
1191 }
1192
1193 /**
1194  * load_compressed_image - Load compressed image data and decompress it.
1195  * @handle: Swap map handle to use for loading data.
1196  * @snapshot: Image to copy uncompressed data into.
1197  * @nr_to_read: Number of pages to load.
1198  */
1199 static int load_compressed_image(struct swap_map_handle *handle,
1200                                  struct snapshot_handle *snapshot,
1201                                  unsigned int nr_to_read)
1202 {
1203         unsigned int m;
1204         int ret = 0;
1205         int eof = 0;
1206         struct hib_bio_batch hb;
1207         ktime_t start;
1208         ktime_t stop;
1209         unsigned nr_pages;
1210         size_t off;
1211         unsigned i, thr, run_threads, nr_threads;
1212         unsigned ring = 0, pg = 0, ring_size = 0,
1213                  have = 0, want, need, asked = 0;
1214         unsigned long read_pages = 0;
1215         unsigned char **page = NULL;
1216         struct dec_data *data = NULL;
1217         struct crc_data *crc = NULL;
1218
1219         hib_init_batch(&hb);
1220
1221         /*
1222          * We'll limit the number of threads for decompression to limit memory
1223          * footprint.
1224          */
1225         nr_threads = num_online_cpus() - 1;
1226         nr_threads = clamp_val(nr_threads, 1, CMP_THREADS);
1227
1228         page = vmalloc(array_size(CMP_MAX_RD_PAGES, sizeof(*page)));
1229         if (!page) {
1230                 pr_err("Failed to allocate %s page\n", hib_comp_algo);
1231                 ret = -ENOMEM;
1232                 goto out_clean;
1233         }
1234
1235         data = vzalloc(array_size(nr_threads, sizeof(*data)));
1236         if (!data) {
1237                 pr_err("Failed to allocate %s data\n", hib_comp_algo);
1238                 ret = -ENOMEM;
1239                 goto out_clean;
1240         }
1241
1242         crc = kzalloc(sizeof(*crc), GFP_KERNEL);
1243         if (!crc) {
1244                 pr_err("Failed to allocate crc\n");
1245                 ret = -ENOMEM;
1246                 goto out_clean;
1247         }
1248
1249         clean_pages_on_decompress = true;
1250
1251         /*
1252          * Start the decompression threads.
1253          */
1254         for (thr = 0; thr < nr_threads; thr++) {
1255                 init_waitqueue_head(&data[thr].go);
1256                 init_waitqueue_head(&data[thr].done);
1257
1258                 data[thr].cc = crypto_alloc_comp(hib_comp_algo, 0, 0);
1259                 if (IS_ERR_OR_NULL(data[thr].cc)) {
1260                         pr_err("Could not allocate comp stream %ld\n", PTR_ERR(data[thr].cc));
1261                         ret = -EFAULT;
1262                         goto out_clean;
1263                 }
1264
1265                 data[thr].thr = kthread_run(decompress_threadfn,
1266                                             &data[thr],
1267                                             "image_decompress/%u", thr);
1268                 if (IS_ERR(data[thr].thr)) {
1269                         data[thr].thr = NULL;
1270                         pr_err("Cannot start decompression threads\n");
1271                         ret = -ENOMEM;
1272                         goto out_clean;
1273                 }
1274         }
1275
1276         /*
1277          * Start the CRC32 thread.
1278          */
1279         init_waitqueue_head(&crc->go);
1280         init_waitqueue_head(&crc->done);
1281
1282         handle->crc32 = 0;
1283         crc->crc32 = &handle->crc32;
1284         for (thr = 0; thr < nr_threads; thr++) {
1285                 crc->unc[thr] = data[thr].unc;
1286                 crc->unc_len[thr] = &data[thr].unc_len;
1287         }
1288
1289         crc->thr = kthread_run(crc32_threadfn, crc, "image_crc32");
1290         if (IS_ERR(crc->thr)) {
1291                 crc->thr = NULL;
1292                 pr_err("Cannot start CRC32 thread\n");
1293                 ret = -ENOMEM;
1294                 goto out_clean;
1295         }
1296
1297         /*
1298          * Set the number of pages for read buffering.
1299          * This is complete guesswork, because we'll only know the real
1300          * picture once prepare_image() is called, which is much later on
1301          * during the image load phase. We'll assume the worst case and
1302          * say that none of the image pages are from high memory.
1303          */
1304         if (low_free_pages() > snapshot_get_image_size())
1305                 read_pages = (low_free_pages() - snapshot_get_image_size()) / 2;
1306         read_pages = clamp_val(read_pages, CMP_MIN_RD_PAGES, CMP_MAX_RD_PAGES);
1307
1308         for (i = 0; i < read_pages; i++) {
1309                 page[i] = (void *)__get_free_page(i < CMP_PAGES ?
1310                                                   GFP_NOIO | __GFP_HIGH :
1311                                                   GFP_NOIO | __GFP_NOWARN |
1312                                                   __GFP_NORETRY);
1313
1314                 if (!page[i]) {
1315                         if (i < CMP_PAGES) {
1316                                 ring_size = i;
1317                                 pr_err("Failed to allocate %s pages\n", hib_comp_algo);
1318                                 ret = -ENOMEM;
1319                                 goto out_clean;
1320                         } else {
1321                                 break;
1322                         }
1323                 }
1324         }
1325         want = ring_size = i;
1326
1327         pr_info("Using %u thread(s) for %s decompression\n", nr_threads, hib_comp_algo);
1328         pr_info("Loading and decompressing image data (%u pages)...\n",
1329                 nr_to_read);
1330         m = nr_to_read / 10;
1331         if (!m)
1332                 m = 1;
1333         nr_pages = 0;
1334         start = ktime_get();
1335
1336         ret = snapshot_write_next(snapshot);
1337         if (ret <= 0)
1338                 goto out_finish;
1339
1340         for(;;) {
1341                 for (i = 0; !eof && i < want; i++) {
1342                         ret = swap_read_page(handle, page[ring], &hb);
1343                         if (ret) {
1344                                 /*
1345                                  * On real read error, finish. On end of data,
1346                                  * set EOF flag and just exit the read loop.
1347                                  */
1348                                 if (handle->cur &&
1349                                     handle->cur->entries[handle->k]) {
1350                                         goto out_finish;
1351                                 } else {
1352                                         eof = 1;
1353                                         break;
1354                                 }
1355                         }
1356                         if (++ring >= ring_size)
1357                                 ring = 0;
1358                 }
1359                 asked += i;
1360                 want -= i;
1361
1362                 /*
1363                  * We are out of data, wait for some more.
1364                  */
1365                 if (!have) {
1366                         if (!asked)
1367                                 break;
1368
1369                         ret = hib_wait_io(&hb);
1370                         if (ret)
1371                                 goto out_finish;
1372                         have += asked;
1373                         asked = 0;
1374                         if (eof)
1375                                 eof = 2;
1376                 }
1377
1378                 if (crc->run_threads) {
1379                         wait_event(crc->done, atomic_read_acquire(&crc->stop));
1380                         atomic_set(&crc->stop, 0);
1381                         crc->run_threads = 0;
1382                 }
1383
1384                 for (thr = 0; have && thr < nr_threads; thr++) {
1385                         data[thr].cmp_len = *(size_t *)page[pg];
1386                         if (unlikely(!data[thr].cmp_len ||
1387                                      data[thr].cmp_len >
1388                                         bytes_worst_compress(UNC_SIZE))) {
1389                                 pr_err("Invalid %s compressed length\n", hib_comp_algo);
1390                                 ret = -1;
1391                                 goto out_finish;
1392                         }
1393
1394                         need = DIV_ROUND_UP(data[thr].cmp_len + CMP_HEADER,
1395                                             PAGE_SIZE);
1396                         if (need > have) {
1397                                 if (eof > 1) {
1398                                         ret = -1;
1399                                         goto out_finish;
1400                                 }
1401                                 break;
1402                         }
1403
1404                         for (off = 0;
1405                              off < CMP_HEADER + data[thr].cmp_len;
1406                              off += PAGE_SIZE) {
1407                                 memcpy(data[thr].cmp + off,
1408                                        page[pg], PAGE_SIZE);
1409                                 have--;
1410                                 want++;
1411                                 if (++pg >= ring_size)
1412                                         pg = 0;
1413                         }
1414
1415                         atomic_set_release(&data[thr].ready, 1);
1416                         wake_up(&data[thr].go);
1417                 }
1418
1419                 /*
1420                  * Wait for more data while we are decompressing.
1421                  */
1422                 if (have < CMP_PAGES && asked) {
1423                         ret = hib_wait_io(&hb);
1424                         if (ret)
1425                                 goto out_finish;
1426                         have += asked;
1427                         asked = 0;
1428                         if (eof)
1429                                 eof = 2;
1430                 }
1431
1432                 for (run_threads = thr, thr = 0; thr < run_threads; thr++) {
1433                         wait_event(data[thr].done,
1434                                 atomic_read_acquire(&data[thr].stop));
1435                         atomic_set(&data[thr].stop, 0);
1436
1437                         ret = data[thr].ret;
1438
1439                         if (ret < 0) {
1440                                 pr_err("%s decompression failed\n", hib_comp_algo);
1441                                 goto out_finish;
1442                         }
1443
1444                         if (unlikely(!data[thr].unc_len ||
1445                                 data[thr].unc_len > UNC_SIZE ||
1446                                 data[thr].unc_len & (PAGE_SIZE - 1))) {
1447                                 pr_err("Invalid %s uncompressed length\n", hib_comp_algo);
1448                                 ret = -1;
1449                                 goto out_finish;
1450                         }
1451
1452                         for (off = 0;
1453                              off < data[thr].unc_len; off += PAGE_SIZE) {
1454                                 memcpy(data_of(*snapshot),
1455                                        data[thr].unc + off, PAGE_SIZE);
1456
1457                                 if (!(nr_pages % m))
1458                                         pr_info("Image loading progress: %3d%%\n",
1459                                                 nr_pages / m * 10);
1460                                 nr_pages++;
1461
1462                                 ret = snapshot_write_next(snapshot);
1463                                 if (ret <= 0) {
1464                                         crc->run_threads = thr + 1;
1465                                         atomic_set_release(&crc->ready, 1);
1466                                         wake_up(&crc->go);
1467                                         goto out_finish;
1468                                 }
1469                         }
1470                 }
1471
1472                 crc->run_threads = thr;
1473                 atomic_set_release(&crc->ready, 1);
1474                 wake_up(&crc->go);
1475         }
1476
1477 out_finish:
1478         if (crc->run_threads) {
1479                 wait_event(crc->done, atomic_read_acquire(&crc->stop));
1480                 atomic_set(&crc->stop, 0);
1481         }
1482         stop = ktime_get();
1483         if (!ret) {
1484                 pr_info("Image loading done\n");
1485                 ret = snapshot_write_finalize(snapshot);
1486                 if (!ret && !snapshot_image_loaded(snapshot))
1487                         ret = -ENODATA;
1488                 if (!ret) {
1489                         if (swsusp_header->flags & SF_CRC32_MODE) {
1490                                 if(handle->crc32 != swsusp_header->crc32) {
1491                                         pr_err("Invalid image CRC32!\n");
1492                                         ret = -ENODATA;
1493                                 }
1494                         }
1495                 }
1496         }
1497         swsusp_show_speed(start, stop, nr_to_read, "Read");
1498 out_clean:
1499         hib_finish_batch(&hb);
1500         for (i = 0; i < ring_size; i++)
1501                 free_page((unsigned long)page[i]);
1502         if (crc) {
1503                 if (crc->thr)
1504                         kthread_stop(crc->thr);
1505                 kfree(crc);
1506         }
1507         if (data) {
1508                 for (thr = 0; thr < nr_threads; thr++) {
1509                         if (data[thr].thr)
1510                                 kthread_stop(data[thr].thr);
1511                         if (data[thr].cc)
1512                                 crypto_free_comp(data[thr].cc);
1513                 }
1514                 vfree(data);
1515         }
1516         vfree(page);
1517
1518         return ret;
1519 }
1520
1521 /**
1522  *      swsusp_read - read the hibernation image.
1523  *      @flags_p: flags passed by the "frozen" kernel in the image header should
1524  *                be written into this memory location
1525  */
1526
1527 int swsusp_read(unsigned int *flags_p)
1528 {
1529         int error;
1530         struct swap_map_handle handle;
1531         struct snapshot_handle snapshot;
1532         struct swsusp_info *header;
1533
1534         memset(&snapshot, 0, sizeof(struct snapshot_handle));
1535         error = snapshot_write_next(&snapshot);
1536         if (error < (int)PAGE_SIZE)
1537                 return error < 0 ? error : -EFAULT;
1538         header = (struct swsusp_info *)data_of(snapshot);
1539         error = get_swap_reader(&handle, flags_p);
1540         if (error)
1541                 goto end;
1542         if (!error)
1543                 error = swap_read_page(&handle, header, NULL);
1544         if (!error) {
1545                 error = (*flags_p & SF_NOCOMPRESS_MODE) ?
1546                         load_image(&handle, &snapshot, header->pages - 1) :
1547                         load_compressed_image(&handle, &snapshot, header->pages - 1);
1548         }
1549         swap_reader_finish(&handle);
1550 end:
1551         if (!error)
1552                 pr_debug("Image successfully loaded\n");
1553         else
1554                 pr_debug("Error %d resuming\n", error);
1555         return error;
1556 }
1557
1558 static void *swsusp_holder;
1559
1560 /**
1561  * swsusp_check - Open the resume device and check for the swsusp signature.
1562  * @exclusive: Open the resume device exclusively.
1563  */
1564
1565 int swsusp_check(bool exclusive)
1566 {
1567         void *holder = exclusive ? &swsusp_holder : NULL;
1568         int error;
1569
1570         hib_resume_bdev_file = bdev_file_open_by_dev(swsusp_resume_device,
1571                                 BLK_OPEN_READ, holder, NULL);
1572         if (!IS_ERR(hib_resume_bdev_file)) {
1573                 clear_page(swsusp_header);
1574                 error = hib_submit_io(REQ_OP_READ, swsusp_resume_block,
1575                                         swsusp_header, NULL);
1576                 if (error)
1577                         goto put;
1578
1579                 if (!memcmp(HIBERNATE_SIG, swsusp_header->sig, 10)) {
1580                         memcpy(swsusp_header->sig, swsusp_header->orig_sig, 10);
1581                         swsusp_header_flags = swsusp_header->flags;
1582                         /* Reset swap signature now */
1583                         error = hib_submit_io(REQ_OP_WRITE | REQ_SYNC,
1584                                                 swsusp_resume_block,
1585                                                 swsusp_header, NULL);
1586                 } else {
1587                         error = -EINVAL;
1588                 }
1589                 if (!error && swsusp_header->flags & SF_HW_SIG &&
1590                     swsusp_header->hw_sig != swsusp_hardware_signature) {
1591                         pr_info("Suspend image hardware signature mismatch (%08x now %08x); aborting resume.\n",
1592                                 swsusp_header->hw_sig, swsusp_hardware_signature);
1593                         error = -EINVAL;
1594                 }
1595
1596 put:
1597                 if (error)
1598                         bdev_fput(hib_resume_bdev_file);
1599                 else
1600                         pr_debug("Image signature found, resuming\n");
1601         } else {
1602                 error = PTR_ERR(hib_resume_bdev_file);
1603         }
1604
1605         if (error)
1606                 pr_debug("Image not found (code %d)\n", error);
1607
1608         return error;
1609 }
1610
1611 /**
1612  * swsusp_close - close resume device.
1613  */
1614
1615 void swsusp_close(void)
1616 {
1617         if (IS_ERR(hib_resume_bdev_file)) {
1618                 pr_debug("Image device not initialised\n");
1619                 return;
1620         }
1621
1622         fput(hib_resume_bdev_file);
1623 }
1624
1625 /**
1626  *      swsusp_unmark - Unmark swsusp signature in the resume device
1627  */
1628
1629 #ifdef CONFIG_SUSPEND
1630 int swsusp_unmark(void)
1631 {
1632         int error;
1633
1634         hib_submit_io(REQ_OP_READ, swsusp_resume_block,
1635                         swsusp_header, NULL);
1636         if (!memcmp(HIBERNATE_SIG,swsusp_header->sig, 10)) {
1637                 memcpy(swsusp_header->sig,swsusp_header->orig_sig, 10);
1638                 error = hib_submit_io(REQ_OP_WRITE | REQ_SYNC,
1639                                         swsusp_resume_block,
1640                                         swsusp_header, NULL);
1641         } else {
1642                 pr_err("Cannot find swsusp signature!\n");
1643                 error = -ENODEV;
1644         }
1645
1646         /*
1647          * We just returned from suspend, we don't need the image any more.
1648          */
1649         free_all_swap_pages(root_swap);
1650
1651         return error;
1652 }
1653 #endif
1654
1655 static int __init swsusp_header_init(void)
1656 {
1657         swsusp_header = (struct swsusp_header*) __get_free_page(GFP_KERNEL);
1658         if (!swsusp_header)
1659                 panic("Could not allocate memory for swsusp_header\n");
1660         return 0;
1661 }
1662
1663 core_initcall(swsusp_header_init);