[PATCH] swsusp: introduce the swap map structure
[safe/jmp/linux-2.6] / kernel / power / swsusp.c
index fc50b5d..b09bd7c 100644 (file)
@@ -1,8 +1,7 @@
 /*
  * linux/kernel/power/swsusp.c
  *
- * This file is to realize architecture-independent
- * machine suspend feature using pretty near only high-level routines
+ * This file provides code to write suspend image to swap and read it back.
  *
  * Copyright (C) 1998-2001 Gabor Kuti <seasons@fornax.hu>
  * Copyright (C) 1998,2001-2005 Pavel Machek <pavel@suse.cz>
@@ -31,8 +30,8 @@
  * Alex Badea <vampire@go.ro>:
  * Fixed runaway init
  *
- * Andreas Steinmetz <ast@domdv.de>:
- * Added encrypted suspend option
+ * Rafael J. Wysocki <rjw@sisk.pl>
+ * Added the swap map data structure and reworked the handling of swap
  *
  * More state savers are welcome. Especially for the scsi layer...
  *
 #include <linux/utsname.h>
 #include <linux/version.h>
 #include <linux/delay.h>
-#include <linux/reboot.h>
 #include <linux/bitops.h>
-#include <linux/vt_kern.h>
-#include <linux/kbd_kern.h>
-#include <linux/keyboard.h>
 #include <linux/spinlock.h>
 #include <linux/genhd.h>
 #include <linux/kernel.h>
 #include <linux/swapops.h>
 #include <linux/bootmem.h>
 #include <linux/syscalls.h>
-#include <linux/console.h>
 #include <linux/highmem.h>
 #include <linux/bio.h>
-#include <linux/mount.h>
 
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
 #include <asm/tlbflush.h>
 #include <asm/io.h>
 
-#include <linux/random.h>
-#include <linux/crypto.h>
-#include <asm/scatterlist.h>
-
 #include "power.h"
 
-#define CIPHER "aes"
-#define MAXKEY 32
-#define MAXIV  32
+#ifdef CONFIG_HIGHMEM
+int save_highmem(void);
+int restore_highmem(void);
+#else
+static int save_highmem(void) { return 0; }
+static int restore_highmem(void) { return 0; }
+#endif
 
 extern char resume_file[];
 
-/* Local variables that should not be affected by save */
-unsigned int nr_copy_pages __nosavedata = 0;
-
-/* Suspend pagedir is allocated before final copy, therefore it
-   must be freed after resume
-
-   Warning: this is evil. There are actually two pagedirs at time of
-   resume. One is "pagedir_save", which is empty frame allocated at
-   time of suspend, that must be freed. Second is "pagedir_nosave",
-   allocated at time of resume, that travels through memory not to
-   collide with anything.
-
-   Warning: this is even more evil than it seems. Pagedirs this file
-   talks about are completely different from page directories used by
-   MMU hardware.
- */
-suspend_pagedir_t *pagedir_nosave __nosavedata = NULL;
-suspend_pagedir_t *pagedir_save;
-
 #define SWSUSP_SIG     "S1SUSPEND"
 
 static struct swsusp_header {
-       char reserved[PAGE_SIZE - 20 - MAXKEY - MAXIV - sizeof(swp_entry_t)];
-       u8 key_iv[MAXKEY+MAXIV];
+       char reserved[PAGE_SIZE - 20 - sizeof(swp_entry_t)];
        swp_entry_t swsusp_info;
        char    orig_sig[10];
        char    sig[10];
@@ -129,131 +102,6 @@ static struct swsusp_info swsusp_info;
 static unsigned short swapfile_used[MAX_SWAPFILES];
 static unsigned short root_swap;
 
-static int write_page(unsigned long addr, swp_entry_t * loc);
-static int bio_read_page(pgoff_t page_off, void * page);
-
-static u8 key_iv[MAXKEY+MAXIV];
-
-#ifdef CONFIG_SWSUSP_ENCRYPT
-
-static int crypto_init(int mode, void **mem)
-{
-       int error = 0;
-       int len;
-       char *modemsg;
-       struct crypto_tfm *tfm;
-
-       modemsg = mode ? "suspend not possible" : "resume not possible";
-
-       tfm = crypto_alloc_tfm(CIPHER, CRYPTO_TFM_MODE_CBC);
-       if(!tfm) {
-               printk(KERN_ERR "swsusp: no tfm, %s\n", modemsg);
-               error = -EINVAL;
-               goto out;
-       }
-
-       if(MAXKEY < crypto_tfm_alg_min_keysize(tfm)) {
-               printk(KERN_ERR "swsusp: key buffer too small, %s\n", modemsg);
-               error = -ENOKEY;
-               goto fail;
-       }
-
-       if (mode)
-               get_random_bytes(key_iv, MAXKEY+MAXIV);
-
-       len = crypto_tfm_alg_max_keysize(tfm);
-       if (len > MAXKEY)
-               len = MAXKEY;
-
-       if (crypto_cipher_setkey(tfm, key_iv, len)) {
-               printk(KERN_ERR "swsusp: key setup failure, %s\n", modemsg);
-               error = -EKEYREJECTED;
-               goto fail;
-       }
-
-       len = crypto_tfm_alg_ivsize(tfm);
-
-       if (MAXIV < len) {
-               printk(KERN_ERR "swsusp: iv buffer too small, %s\n", modemsg);
-               error = -EOVERFLOW;
-               goto fail;
-       }
-
-       crypto_cipher_set_iv(tfm, key_iv+MAXKEY, len);
-
-       *mem=(void *)tfm;
-
-       goto out;
-
-fail:  crypto_free_tfm(tfm);
-out:   return error;
-}
-
-static __inline__ void crypto_exit(void *mem)
-{
-       crypto_free_tfm((struct crypto_tfm *)mem);
-}
-
-static __inline__ int crypto_write(struct pbe *p, void *mem)
-{
-       int error = 0;
-       struct scatterlist src, dst;
-
-       src.page   = virt_to_page(p->address);
-       src.offset = 0;
-       src.length = PAGE_SIZE;
-       dst.page   = virt_to_page((void *)&swsusp_header);
-       dst.offset = 0;
-       dst.length = PAGE_SIZE;
-
-       error = crypto_cipher_encrypt((struct crypto_tfm *)mem, &dst, &src,
-                                       PAGE_SIZE);
-
-       if (!error)
-               error = write_page((unsigned long)&swsusp_header,
-                               &(p->swap_address));
-       return error;
-}
-
-static __inline__ int crypto_read(struct pbe *p, void *mem)
-{
-       int error = 0;
-       struct scatterlist src, dst;
-
-       error = bio_read_page(swp_offset(p->swap_address), (void *)p->address);
-       if (!error) {
-               src.offset = 0;
-               src.length = PAGE_SIZE;
-               dst.offset = 0;
-               dst.length = PAGE_SIZE;
-               src.page = dst.page = virt_to_page((void *)p->address);
-
-               error = crypto_cipher_decrypt((struct crypto_tfm *)mem, &dst,
-                                               &src, PAGE_SIZE);
-       }
-       return error;
-}
-#else
-static __inline__ int crypto_init(int mode, void *mem)
-{
-       return 0;
-}
-
-static __inline__ void crypto_exit(void *mem)
-{
-}
-
-static __inline__ int crypto_write(struct pbe *p, void *mem)
-{
-       return write_page(p->address, &(p->swap_address));
-}
-
-static __inline__ int crypto_read(struct pbe *p, void *mem)
-{
-       return bio_read_page(swp_offset(p->swap_address), (void *)p->address);
-}
-#endif
-
 static int mark_swapfiles(swp_entry_t prev)
 {
        int error;
@@ -265,7 +113,6 @@ static int mark_swapfiles(swp_entry_t prev)
            !memcmp("SWAPSPACE2",swsusp_header.sig, 10)) {
                memcpy(swsusp_header.orig_sig,swsusp_header.sig, 10);
                memcpy(swsusp_header.sig,SWSUSP_SIG, 10);
-               memcpy(swsusp_header.key_iv, key_iv, MAXKEY+MAXIV);
                swsusp_header.swsusp_info = prev;
                error = rw_swap_page_sync(WRITE,
                                          swp_entry(root_swap, 0),
@@ -362,7 +209,7 @@ static void lock_swapdevices(void)
  *     This is a partial improvement, since we will at least return other
  *     errors, though we need to eventually fix the damn code.
  */
-static int write_page(unsigned long addr, swp_entry_t * loc)
+static int write_page(unsigned long addr, swp_entry_t *loc)
 {
        swp_entry_t entry;
        int error = 0;
@@ -382,55 +229,205 @@ static int write_page(unsigned long addr, swp_entry_t * loc)
 }
 
 /**
- *     data_free - Free the swap entries used by the saved image.
+ *     Swap map-handling functions
+ *
+ *     The swap map is a data structure used for keeping track of each page
+ *     written to the swap.  It consists of many swap_map_page structures
+ *     that contain each an array of MAP_PAGE_SIZE swap entries.
+ *     These structures are linked together with the help of either the
+ *     .next (in memory) or the .next_swap (in swap) member.
  *
- *     Walk the list of used swap entries and free each one.
- *     This is only used for cleanup when suspend fails.
+ *     The swap map is created during suspend.  At that time we need to keep
+ *     it in memory, because we have to free all of the allocated swap
+ *     entries if an error occurs.  The memory needed is preallocated
+ *     so that we know in advance if there's enough of it.
+ *
+ *     The first swap_map_page structure is filled with the swap entries that
+ *     correspond to the first MAP_PAGE_SIZE data pages written to swap and
+ *     so on.  After the all of the data pages have been written, the order
+ *     of the swap_map_page structures in the map is reversed so that they
+ *     can be read from swap in the original order.  This causes the data
+ *     pages to be loaded in exactly the same order in which they have been
+ *     saved.
+ *
+ *     During resume we only need to use one swap_map_page structure
+ *     at a time, which means that we only need to use two memory pages for
+ *     reading the image - one for reading the swap_map_page structures
+ *     and the second for reading the data pages from swap.
  */
-static void data_free(void)
+
+#define MAP_PAGE_SIZE  ((PAGE_SIZE - sizeof(swp_entry_t) - sizeof(void *)) \
+                       / sizeof(swp_entry_t))
+
+struct swap_map_page {
+       swp_entry_t             entries[MAP_PAGE_SIZE];
+       swp_entry_t             next_swap;
+       struct swap_map_page    *next;
+};
+
+static inline void free_swap_map(struct swap_map_page *swap_map)
 {
-       swp_entry_t entry;
-       struct pbe * p;
+       struct swap_map_page *swp;
 
-       for_each_pbe(p, pagedir_nosave) {
-               entry = p->swap_address;
-               if (entry.val)
-                       swap_free(entry);
-               else
-                       break;
+       while (swap_map) {
+               swp = swap_map->next;
+               free_page((unsigned long)swap_map);
+               swap_map = swp;
        }
 }
 
+static struct swap_map_page *alloc_swap_map(unsigned int nr_pages)
+{
+       struct swap_map_page *swap_map, *swp;
+       unsigned n = 0;
+
+       if (!nr_pages)
+               return NULL;
+
+       pr_debug("alloc_swap_map(): nr_pages = %d\n", nr_pages);
+       swap_map = (struct swap_map_page *)get_zeroed_page(GFP_ATOMIC);
+       swp = swap_map;
+       for (n = MAP_PAGE_SIZE; n < nr_pages; n += MAP_PAGE_SIZE) {
+               swp->next = (struct swap_map_page *)get_zeroed_page(GFP_ATOMIC);
+               swp = swp->next;
+               if (!swp) {
+                       free_swap_map(swap_map);
+                       return NULL;
+               }
+       }
+       return swap_map;
+}
+
 /**
- *     data_write - Write saved image to swap.
- *
- *     Walk the list of pages in the image and sync each one to swap.
+ *     reverse_swap_map - reverse the order of pages in the swap map
+ *     @swap_map
  */
-static int data_write(void)
+
+static inline struct swap_map_page *reverse_swap_map(struct swap_map_page *swap_map)
 {
-       int error = 0, i = 0;
-       unsigned int mod = nr_copy_pages / 100;
-       struct pbe *p;
-       void *tfm;
+       struct swap_map_page *prev, *next;
+
+       prev = NULL;
+       while (swap_map) {
+               next = swap_map->next;
+               swap_map->next = prev;
+               prev = swap_map;
+               swap_map = next;
+       }
+       return prev;
+}
 
-       if ((error = crypto_init(1, &tfm)))
-               return error;
+/**
+ *     free_swap_map_entries - free the swap entries allocated to store
+ *     the swap map @swap_map (this is only called in case of an error)
+ */
+static inline void free_swap_map_entries(struct swap_map_page *swap_map)
+{
+       while (swap_map) {
+               if (swap_map->next_swap.val)
+                       swap_free(swap_map->next_swap);
+               swap_map = swap_map->next;
+       }
+}
 
-       if (!mod)
-               mod = 1;
+/**
+ *     save_swap_map - save the swap map used for tracing the data pages
+ *     stored in the swap
+ */
 
-       printk( "Writing data to swap (%d pages)...     ", nr_copy_pages );
-       for_each_pbe (p, pagedir_nosave) {
-               if (!(i%mod))
-                       printk( "\b\b\b\b%3d%%", i / mod );
-               if ((error = crypto_write(p, tfm))) {
-                       crypto_exit(tfm);
+static int save_swap_map(struct swap_map_page *swap_map, swp_entry_t *start)
+{
+       swp_entry_t entry = (swp_entry_t){0};
+       int error;
+
+       while (swap_map) {
+               swap_map->next_swap = entry;
+               if ((error = write_page((unsigned long)swap_map, &entry)))
                        return error;
-               }
-               i++;
+               swap_map = swap_map->next;
+       }
+       *start = entry;
+       return 0;
+}
+
+/**
+ *     free_image_entries - free the swap entries allocated to store
+ *     the image data pages (this is only called in case of an error)
+ */
+
+static inline void free_image_entries(struct swap_map_page *swp)
+{
+       unsigned k;
+
+       while (swp) {
+               for (k = 0; k < MAP_PAGE_SIZE; k++)
+                       if (swp->entries[k].val)
+                               swap_free(swp->entries[k]);
+               swp = swp->next;
        }
-       printk("\b\b\b\bdone\n");
-       crypto_exit(tfm);
+}
+
+/**
+ *     The swap_map_handle structure is used for handling the swap map in
+ *     a file-alike way
+ */
+
+struct swap_map_handle {
+       struct swap_map_page *cur;
+       unsigned int k;
+};
+
+static inline void init_swap_map_handle(struct swap_map_handle *handle,
+                                        struct swap_map_page *map)
+{
+       handle->cur = map;
+       handle->k = 0;
+}
+
+static inline int swap_map_write_page(struct swap_map_handle *handle,
+                                      unsigned long addr)
+{
+       int error;
+
+       error = write_page(addr, handle->cur->entries + handle->k);
+       if (error)
+               return error;
+       if (++handle->k >= MAP_PAGE_SIZE) {
+               handle->cur = handle->cur->next;
+               handle->k = 0;
+       }
+       return 0;
+}
+
+/**
+ *     save_image_data - save the data pages pointed to by the PBEs
+ *     from the list @pblist using the swap map handle @handle
+ *     (assume there are @nr_pages data pages to save)
+ */
+
+static int save_image_data(struct pbe *pblist,
+                           struct swap_map_handle *handle,
+                           unsigned int nr_pages)
+{
+       unsigned int m;
+       struct pbe *p;
+       int error = 0;
+
+       printk("Saving image data pages (%u pages) ...     ", nr_pages);
+       m = nr_pages / 100;
+       if (!m)
+               m = 1;
+       nr_pages = 0;
+       for_each_pbe (p, pblist) {
+               error = swap_map_write_page(handle, p->address);
+               if (error)
+                       break;
+               if (!(nr_pages % m))
+                       printk("\b\b\b\b%3d%%", nr_pages / m);
+               nr_pages++;
+       }
+       if (!error)
+               printk("\b\b\b\bdone\n");
        return error;
 }
 
@@ -446,19 +443,20 @@ static void dump_info(void)
        pr_debug(" swsusp: UTS Domain: %s\n",swsusp_info.uts.domainname);
        pr_debug(" swsusp: CPUs: %d\n",swsusp_info.cpus);
        pr_debug(" swsusp: Image: %ld Pages\n",swsusp_info.image_pages);
-       pr_debug(" swsusp: Pagedir: %ld Pages\n",swsusp_info.pagedir_pages);
+       pr_debug(" swsusp: Total: %ld Pages\n", swsusp_info.pages);
 }
 
-static void init_header(void)
+static void init_header(unsigned int nr_pages)
 {
        memset(&swsusp_info, 0, sizeof(swsusp_info));
        swsusp_info.version_code = LINUX_VERSION_CODE;
        swsusp_info.num_physpages = num_physpages;
        memcpy(&swsusp_info.uts, &system_utsname, sizeof(system_utsname));
 
-       swsusp_info.suspend_pagedir = pagedir_nosave;
        swsusp_info.cpus = num_online_cpus();
-       swsusp_info.image_pages = nr_copy_pages;
+       swsusp_info.image_pages = nr_pages;
+       swsusp_info.pages = nr_pages +
+               ((nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT);
 }
 
 static int close_swap(void)
@@ -477,67 +475,54 @@ static int close_swap(void)
 }
 
 /**
- *     free_pagedir_entries - Free pages used by the page directory.
- *
- *     This is used during suspend for error recovery.
+ *     pack_orig_addresses - the .orig_address fields of the PBEs from the
+ *     list starting at @pbe are stored in the array @buf[] (1 page)
  */
 
-static void free_pagedir_entries(void)
+static inline struct pbe *pack_orig_addresses(unsigned long *buf,
+                                              struct pbe *pbe)
 {
-       int i;
+       int j;
 
-       for (i = 0; i < swsusp_info.pagedir_pages; i++)
-               swap_free(swsusp_info.pagedir[i]);
+       for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
+               buf[j] = pbe->orig_address;
+               pbe = pbe->next;
+       }
+       if (!pbe)
+               for (; j < PAGE_SIZE / sizeof(long); j++)
+                       buf[j] = 0;
+       return pbe;
 }
 
-
 /**
- *     write_pagedir - Write the array of pages holding the page directory.
- *     @last:  Last swap entry we write (needed for header).
+ *     save_image_metadata - save the .orig_address fields of the PBEs
+ *     from the list @pblist using the swap map handle @handle
  */
 
-static int write_pagedir(void)
+static int save_image_metadata(struct pbe *pblist,
+                               struct swap_map_handle *handle)
 {
+       unsigned long *buf;
+       unsigned int n = 0;
+       struct pbe *p;
        int error = 0;
-       unsigned n = 0;
-       struct pbe * pbe;
 
-       printk( "Writing pagedir...");
-       for_each_pb_page (pbe, pagedir_nosave) {
-               if ((error = write_page((unsigned long)pbe, &swsusp_info.pagedir[n++])))
-                       return error;
+       printk("Saving image metadata ... ");
+       buf = (unsigned long *)get_zeroed_page(GFP_ATOMIC);
+       if (!buf)
+               return -ENOMEM;
+       p = pblist;
+       while (p) {
+               p = pack_orig_addresses(buf, p);
+               error = swap_map_write_page(handle, (unsigned long)buf);
+               if (error)
+                       break;
+               n++;
        }
-
-       swsusp_info.pagedir_pages = n;
-       printk("done (%u pages)\n", n);
-       return error;
-}
-
-/**
- *     write_suspend_image - Write entire image and metadata.
- *
- */
-static int write_suspend_image(void)
-{
-       int error;
-
-       init_header();
-       if ((error = data_write()))
-               goto FreeData;
-
-       if ((error = write_pagedir()))
-               goto FreePagedir;
-
-       if ((error = close_swap()))
-               goto FreePagedir;
- Done:
-       memset(key_iv, 0, MAXKEY+MAXIV);
+       free_page((unsigned long)buf);
+       if (!error)
+               printk("done (%u pages saved)\n", n);
        return error;
- FreePagedir:
-       free_pagedir_entries();
- FreeData:
-       data_free();
-       goto Done;
 }
 
 /**
@@ -550,39 +535,86 @@ static int write_suspend_image(void)
  *     We should only consider resume_device.
  */
 
-int enough_swap(void)
+static int enough_swap(unsigned int nr_pages)
 {
        struct sysinfo i;
 
        si_swapinfo(&i);
        pr_debug("swsusp: available swap: %lu pages\n", i.freeswap);
-       return i.freeswap > (nr_copy_pages + PAGES_FOR_IO +
-               nr_copy_pages/PBES_PER_PAGE + !!(nr_copy_pages%PBES_PER_PAGE));
+       return i.freeswap > (nr_pages + PAGES_FOR_IO +
+               (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 }
 
+/**
+ *     write_suspend_image - Write entire image and metadata.
+ */
+static int write_suspend_image(struct pbe *pblist, unsigned int nr_pages)
+{
+       struct swap_map_page *swap_map;
+       struct swap_map_handle handle;
+       int error;
+
+       if (!enough_swap(nr_pages)) {
+               printk(KERN_ERR "swsusp: Not enough free swap\n");
+               return -ENOSPC;
+       }
+
+       init_header(nr_pages);
+       swap_map = alloc_swap_map(swsusp_info.pages);
+       if (!swap_map)
+               return -ENOMEM;
+       init_swap_map_handle(&handle, swap_map);
+
+       error = save_image_metadata(pblist, &handle);
+       if (!error)
+               error = save_image_data(pblist, &handle, nr_pages);
+       if (error)
+               goto Free_image_entries;
+
+       swap_map = reverse_swap_map(swap_map);
+       error = save_swap_map(swap_map, &swsusp_info.start);
+       if (error)
+               goto Free_map_entries;
+
+       error = close_swap();
+       if (error)
+               goto Free_map_entries;
+
+Free_swap_map:
+       free_swap_map(swap_map);
+       return error;
+
+Free_map_entries:
+       free_swap_map_entries(swap_map);
+Free_image_entries:
+       free_image_entries(swap_map);
+       goto Free_swap_map;
+}
 
 /* It is important _NOT_ to umount filesystems at this point. We want
  * them synced (in case something goes wrong) but we DO not want to mark
  * filesystem clean: it is not. (And it does not matter, if we resume
  * correctly, we'll mark system clean, anyway.)
  */
-int swsusp_write(void)
+int swsusp_write(struct pbe *pblist, unsigned int nr_pages)
 {
        int error;
-       device_resume();
+
+       if ((error = swsusp_swap_check())) {
+               printk(KERN_ERR "swsusp: cannot find swap device, try swapon -a.\n");
+               return error;
+       }
        lock_swapdevices();
-       error = write_suspend_image();
+       error = write_suspend_image(pblist, nr_pages);
        /* This will unlock ignored swap devices since writing is finished */
        lock_swapdevices();
        return error;
-
 }
 
-
-
 int swsusp_suspend(void)
 {
        int error;
+
        if ((error = arch_prepare_suspend()))
                return error;
        local_irq_disable();
@@ -594,15 +626,12 @@ int swsusp_suspend(void)
         */
        if ((error = device_power_down(PMSG_FREEZE))) {
                printk(KERN_ERR "Some devices failed to power down, aborting suspend\n");
-               local_irq_enable();
-               return error;
+               goto Enable_irqs;
        }
 
-       if ((error = swsusp_swap_check())) {
-               printk(KERN_ERR "swsusp: cannot find swap device, try swapon -a.\n");
-               device_power_up();
-               local_irq_enable();
-               return error;
+       if ((error = save_highmem())) {
+               printk(KERN_ERR "swsusp: Not enough free pages for highmem\n");
+               goto Restore_highmem;
        }
 
        save_processor_state();
@@ -610,8 +639,10 @@ int swsusp_suspend(void)
                printk(KERN_ERR "Error %d suspending\n", error);
        /* Restore control flow magically appears here */
        restore_processor_state();
+Restore_highmem:
        restore_highmem();
        device_power_up();
+Enable_irqs:
        local_irq_enable();
        return error;
 }
@@ -629,6 +660,11 @@ int swsusp_resume(void)
         * execution continues at place where swsusp_arch_suspend was called
          */
        BUG_ON(!error);
+       /* The only reason why swsusp_arch_resume() can fail is memory being
+        * very tight, so we have to free it as soon as we can to avoid
+        * subsequent failures
+        */
+       swsusp_free();
        restore_processor_state();
        restore_highmem();
        touch_softlockup_watchdog();
@@ -638,159 +674,42 @@ int swsusp_resume(void)
 }
 
 /**
- *     On resume, for storing the PBE list and the image,
- *     we can only use memory pages that do not conflict with the pages
- *     which had been used before suspend.
- *
- *     We don't know which pages are usable until we allocate them.
- *
- *     Allocated but unusable (ie eaten) memory pages are linked together
- *     to create a list, so that we can free them easily
- *
- *     We could have used a type other than (void *)
- *     for this purpose, but ...
+ *     mark_unsafe_pages - mark the pages that cannot be used for storing
+ *     the image during resume, because they conflict with the pages that
+ *     had been used before suspend
  */
-static void **eaten_memory = NULL;
-
-static inline void eat_page(void *page)
-{
-       void **c;
-
-       c = eaten_memory;
-       eaten_memory = page;
-       *eaten_memory = c;
-}
 
-unsigned long get_usable_page(gfp_t gfp_mask)
-{
-       unsigned long m;
-
-       m = get_zeroed_page(gfp_mask);
-       while (!PageNosaveFree(virt_to_page(m))) {
-               eat_page((void *)m);
-               m = get_zeroed_page(gfp_mask);
-               if (!m)
-                       break;
-       }
-       return m;
-}
-
-void free_eaten_memory(void)
-{
-       unsigned long m;
-       void **c;
-       int i = 0;
-
-       c = eaten_memory;
-       while (c) {
-               m = (unsigned long)c;
-               c = *c;
-               free_page(m);
-               i++;
-       }
-       eaten_memory = NULL;
-       pr_debug("swsusp: %d unused pages freed\n", i);
-}
-
-/**
- *     check_pagedir - We ensure here that pages that the PBEs point to
- *     won't collide with pages where we're going to restore from the loaded
- *     pages later
- */
-
-static int check_pagedir(struct pbe *pblist)
-{
-       struct pbe *p;
-
-       /* This is necessary, so that we can free allocated pages
-        * in case of failure
-        */
-       for_each_pbe (p, pblist)
-               p->address = 0UL;
-
-       for_each_pbe (p, pblist) {
-               p->address = get_usable_page(GFP_ATOMIC);
-               if (!p->address)
-                       return -ENOMEM;
-       }
-       return 0;
-}
-
-/**
- *     swsusp_pagedir_relocate - It is possible, that some memory pages
- *     occupied by the list of PBEs collide with pages where we're going to
- *     restore from the loaded pages later.  We relocate them here.
- */
-
-static struct pbe * swsusp_pagedir_relocate(struct pbe *pblist)
+static void mark_unsafe_pages(struct pbe *pblist)
 {
        struct zone *zone;
        unsigned long zone_pfn;
-       struct pbe *pbpage, *tail, *p;
-       void *m;
-       int rel = 0, error = 0;
+       struct pbe *p;
 
        if (!pblist) /* a sanity check */
-               return NULL;
-
-       pr_debug("swsusp: Relocating pagedir (%lu pages to check)\n",
-                       swsusp_info.pagedir_pages);
-
-       /* Set page flags */
+               return;
 
+       /* Clear page flags */
        for_each_zone (zone) {
-               for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
-                       SetPageNosaveFree(pfn_to_page(zone_pfn +
+               for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
+                       if (pfn_valid(zone_pfn + zone->zone_start_pfn))
+                               ClearPageNosaveFree(pfn_to_page(zone_pfn +
                                        zone->zone_start_pfn));
        }
 
-       /* Clear orig addresses */
-
+       /* Mark orig addresses */
        for_each_pbe (p, pblist)
-               ClearPageNosaveFree(virt_to_page(p->orig_address));
-
-       tail = pblist + PB_PAGE_SKIP;
-
-       /* Relocate colliding pages */
+               SetPageNosaveFree(virt_to_page(p->orig_address));
 
-       for_each_pb_page (pbpage, pblist) {
-               if (!PageNosaveFree(virt_to_page((unsigned long)pbpage))) {
-                       m = (void *)get_usable_page(GFP_ATOMIC | __GFP_COLD);
-                       if (!m) {
-                               error = -ENOMEM;
-                               break;
-                       }
-                       memcpy(m, (void *)pbpage, PAGE_SIZE);
-                       if (pbpage == pblist)
-                               pblist = (struct pbe *)m;
-                       else
-                               tail->next = (struct pbe *)m;
-
-                       eat_page((void *)pbpage);
-                       pbpage = (struct pbe *)m;
-
-                       /* We have to link the PBEs again */
-
-                       for (p = pbpage; p < pbpage + PB_PAGE_SKIP; p++)
-                               if (p->next) /* needed to save the end */
-                                       p->next = p + 1;
+}
 
-                       rel++;
-               }
-               tail = pbpage + PB_PAGE_SKIP;
+static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
+{
+       /* We assume both lists contain the same number of elements */
+       while (src) {
+               dst->orig_address = src->orig_address;
+               dst = dst->next;
+               src = src->next;
        }
-
-       if (error) {
-               printk("\nswsusp: Out of memory\n\n");
-               free_pagedir(pblist);
-               free_eaten_memory();
-               pblist = NULL;
-               /* Is this even worth handling? It should never ever happen, and we
-                  have just lost user's state, anyway... */
-       } else
-               printk("swsusp: Relocated %d pages\n", rel);
-
-       return pblist;
 }
 
 /*
@@ -804,7 +723,7 @@ static struct pbe * swsusp_pagedir_relocate(struct pbe *pblist)
 
 static atomic_t io_done = ATOMIC_INIT(0);
 
-static int end_io(struct bio * bio, unsigned int num, int err)
+static int end_io(struct bio *bio, unsigned int num, int err)
 {
        if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
                panic("I/O error reading memory image");
@@ -812,7 +731,7 @@ static int end_io(struct bio * bio, unsigned int num, int err)
        return 0;
 }
 
-static struct block_device * resume_bdev;
+static struct block_device *resume_bdev;
 
 /**
  *     submit - submit BIO request.
@@ -825,10 +744,10 @@ static struct block_device * resume_bdev;
  *     Then submit it and wait.
  */
 
-static int submit(int rw, pgoff_t page_off, void * page)
+static int submit(int rw, pgoff_t page_off, void *page)
 {
        int error = 0;
-       struct bio * bio;
+       struct bio *bio;
 
        bio = bio_alloc(GFP_ATOMIC, 1);
        if (!bio)
@@ -857,22 +776,77 @@ static int submit(int rw, pgoff_t page_off, void * page)
        return error;
 }
 
-static int bio_read_page(pgoff_t page_off, void * page)
+static int bio_read_page(pgoff_t page_off, void *page)
 {
        return submit(READ, page_off, page);
 }
 
-static int bio_write_page(pgoff_t page_off, void * page)
+static int bio_write_page(pgoff_t page_off, void *page)
 {
        return submit(WRITE, page_off, page);
 }
 
+/**
+ *     The following functions allow us to read data using a swap map
+ *     in a file-alike way
+ */
+
+static inline void release_swap_map_reader(struct swap_map_handle *handle)
+{
+       if (handle->cur)
+               free_page((unsigned long)handle->cur);
+       handle->cur = NULL;
+}
+
+static inline int get_swap_map_reader(struct swap_map_handle *handle,
+                                      swp_entry_t start)
+{
+       int error;
+
+       if (!swp_offset(start))
+               return -EINVAL;
+       handle->cur = (struct swap_map_page *)get_zeroed_page(GFP_ATOMIC);
+       if (!handle->cur)
+               return -ENOMEM;
+       error = bio_read_page(swp_offset(start), handle->cur);
+       if (error) {
+               release_swap_map_reader(handle);
+               return error;
+       }
+       handle->k = 0;
+       return 0;
+}
+
+static inline int swap_map_read_page(struct swap_map_handle *handle, void *buf)
+{
+       unsigned long offset;
+       int error;
+
+       if (!handle->cur)
+               return -EINVAL;
+       offset = swp_offset(handle->cur->entries[handle->k]);
+       if (!offset)
+               return -EINVAL;
+       error = bio_read_page(offset, buf);
+       if (error)
+               return error;
+       if (++handle->k >= MAP_PAGE_SIZE) {
+               handle->k = 0;
+               offset = swp_offset(handle->cur->next_swap);
+               if (!offset)
+                       release_swap_map_reader(handle);
+               else
+                       error = bio_read_page(offset, handle->cur);
+       }
+       return error;
+}
+
 /*
  * Sanity check if this image makes sense with this kernel/swap context
  * I really don't think that it's foolproof but more than nothing..
  */
 
-static const char * sanity_check(void)
+static const char *sanity_check(void)
 {
        dump_info();
        if (swsusp_info.version_code != LINUX_VERSION_CODE)
@@ -895,10 +869,9 @@ static const char * sanity_check(void)
        return NULL;
 }
 
-
 static int check_header(void)
 {
-       const char * reason = NULL;
+       const char *reason = NULL;
        int error;
 
        if ((error = bio_read_page(swp_offset(swsusp_header.swsusp_info), &swsusp_info)))
@@ -909,7 +882,6 @@ static int check_header(void)
                printk(KERN_ERR "swsusp: Resume mismatch: %s\n",reason);
                return -EPERM;
        }
-       nr_copy_pages = swsusp_info.image_pages;
        return error;
 }
 
@@ -922,14 +894,12 @@ static int check_sig(void)
                return error;
        if (!memcmp(SWSUSP_SIG, swsusp_header.sig, 10)) {
                memcpy(swsusp_header.sig, swsusp_header.orig_sig, 10);
-               memcpy(key_iv, swsusp_header.key_iv, MAXKEY+MAXIV);
-               memset(swsusp_header.key_iv, 0, MAXKEY+MAXIV);
 
                /*
                 * Reset swap signature now.
                 */
                error = bio_write_page(0, &swsusp_header);
-       } else { 
+       } else {
                return -EINVAL;
        }
        if (!error)
@@ -938,83 +908,88 @@ static int check_sig(void)
 }
 
 /**
- *     data_read - Read image pages from swap.
- *
- *     You do not need to check for overlaps, check_pagedir()
- *     already did that.
+ *     load_image_data - load the image data using the swap map handle
+ *     @handle and store them using the page backup list @pblist
+ *     (assume there are @nr_pages pages to load)
  */
 
-static int data_read(struct pbe *pblist)
+static int load_image_data(struct pbe *pblist,
+                           struct swap_map_handle *handle,
+                           unsigned int nr_pages)
 {
-       struct pbe * p;
-       int error = 0;
-       int i = 0;
-       int mod = swsusp_info.image_pages / 100;
-       void *tfm;
-
-       if ((error = crypto_init(0, &tfm)))
-               return error;
-
-       if (!mod)
-               mod = 1;
-
-       printk("swsusp: Reading image data (%lu pages):     ",
-                       swsusp_info.image_pages);
-
-       for_each_pbe (p, pblist) {
-               if (!(i % mod))
-                       printk("\b\b\b\b%3d%%", i / mod);
-
-               if ((error = crypto_read(p, tfm))) {
-                       crypto_exit(tfm);
-                       return error;
-               }
+       int error;
+       unsigned int m;
+       struct pbe *p;
 
-               i++;
+       if (!pblist)
+               return -EINVAL;
+       printk("Loading image data pages (%u pages) ...     ", nr_pages);
+       m = nr_pages / 100;
+       if (!m)
+               m = 1;
+       nr_pages = 0;
+       p = pblist;
+       while (p) {
+               error = swap_map_read_page(handle, (void *)p->address);
+               if (error)
+                       break;
+               p = p->next;
+               if (!(nr_pages % m))
+                       printk("\b\b\b\b%3d%%", nr_pages / m);
+               nr_pages++;
        }
-       printk("\b\b\b\bdone\n");
-       crypto_exit(tfm);
+       if (!error)
+               printk("\b\b\b\bdone\n");
        return error;
 }
 
 /**
- *     read_pagedir - Read page backup list pages from swap
+ *     unpack_orig_addresses - copy the elements of @buf[] (1 page) to
+ *     the PBEs in the list starting at @pbe
  */
 
-static int read_pagedir(struct pbe *pblist)
+static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
+                                                struct pbe *pbe)
 {
-       struct pbe *pbpage, *p;
-       unsigned i = 0;
-       int error;
+       int j;
 
-       if (!pblist)
-               return -EFAULT;
+       for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
+               pbe->orig_address = buf[j];
+               pbe = pbe->next;
+       }
+       return pbe;
+}
 
-       printk("swsusp: Reading pagedir (%lu pages)\n",
-                       swsusp_info.pagedir_pages);
+/**
+ *     load_image_metadata - load the image metadata using the swap map
+ *     handle @handle and put them into the PBEs in the list @pblist
+ */
 
-       for_each_pb_page (pbpage, pblist) {
-               unsigned long offset = swp_offset(swsusp_info.pagedir[i++]);
+static int load_image_metadata(struct pbe *pblist, struct swap_map_handle *handle)
+{
+       struct pbe *p;
+       unsigned long *buf;
+       unsigned int n = 0;
+       int error = 0;
 
-               error = -EFAULT;
-               if (offset) {
-                       p = (pbpage + PB_PAGE_SKIP)->next;
-                       error = bio_read_page(offset, (void *)pbpage);
-                       (pbpage + PB_PAGE_SKIP)->next = p;
-               }
+       printk("Loading image metadata ... ");
+       buf = (unsigned long *)get_zeroed_page(GFP_ATOMIC);
+       if (!buf)
+               return -ENOMEM;
+       p = pblist;
+       while (p) {
+               error = swap_map_read_page(handle, buf);
                if (error)
                        break;
+               p = unpack_orig_addresses(buf, p);
+               n++;
        }
-
-       if (error)
-               free_pagedir(pblist);
-       else
-               BUG_ON(i != swsusp_info.pagedir_pages);
-
+       free_page((unsigned long)buf);
+       if (!error)
+               printk("done (%u pages loaded)\n", n);
        return error;
 }
 
-
 static int check_suspend_image(void)
 {
        int error = 0;
@@ -1028,38 +1003,39 @@ static int check_suspend_image(void)
        return 0;
 }
 
-static int read_suspend_image(void)
+static int read_suspend_image(struct pbe **pblist_ptr)
 {
        int error = 0;
-       struct pbe *p;
+       struct pbe *p, *pblist;
+       struct swap_map_handle handle;
+       unsigned int nr_pages = swsusp_info.image_pages;
 
-       if (!(p = alloc_pagedir(nr_copy_pages)))
+       p = alloc_pagedir(nr_pages, GFP_ATOMIC, 0);
+       if (!p)
                return -ENOMEM;
-
-       if ((error = read_pagedir(p)))
+       error = get_swap_map_reader(&handle, swsusp_info.start);
+       if (error)
+               /* The PBE list at p will be released by swsusp_free() */
                return error;
-
-       create_pbe_list(p, nr_copy_pages);
-
-       if (!(pagedir_nosave = swsusp_pagedir_relocate(p)))
-               return -ENOMEM;
-
-       /* Allocate memory for the image and read the data from swap */
-
-       error = check_pagedir(pagedir_nosave);
-
-       if (!error)
-               error = data_read(pagedir_nosave);
-
-       if (error) { /* We fail cleanly */
-               free_eaten_memory();
-               for_each_pbe (p, pagedir_nosave)
-                       if (p->address) {
-                               free_page(p->address);
-                               p->address = 0UL;
-                       }
-               free_pagedir(pagedir_nosave);
+       error = load_image_metadata(p, &handle);
+       if (!error) {
+               mark_unsafe_pages(p);
+               pblist = alloc_pagedir(nr_pages, GFP_ATOMIC, 1);
+               if (pblist)
+                       copy_page_backup_list(pblist, p);
+               free_pagedir(p);
+               if (!pblist)
+                       error = -ENOMEM;
+
+               /* Allocate memory for the image and read the data from swap */
+               if (!error)
+                       error = alloc_data_pages(pblist, GFP_ATOMIC, 1);
+               if (!error)
+                       error = load_image_data(pblist, &handle, nr_pages);
+               if (!error)
+                       *pblist_ptr = pblist;
        }
+       release_swap_map_reader(&handle);
        return error;
 }
 
@@ -1091,7 +1067,7 @@ int swsusp_check(void)
  *     swsusp_read - Read saved image from swap.
  */
 
-int swsusp_read(void)
+int swsusp_read(struct pbe **pblist_ptr)
 {
        int error;
 
@@ -1100,9 +1076,8 @@ int swsusp_read(void)
                return PTR_ERR(resume_bdev);
        }
 
-       error = read_suspend_image();
+       error = read_suspend_image(pblist_ptr);
        blkdev_put(resume_bdev);
-       memset(key_iv, 0, MAXKEY+MAXIV);
 
        if (!error)
                pr_debug("swsusp: Reading resume file was successful\n");