Staging: virtual block device driver (ramzswap)
authorNitin Gupta <ngupta@vflare.org>
Tue, 22 Sep 2009 04:56:53 +0000 (10:26 +0530)
committerGreg Kroah-Hartman <gregkh@suse.de>
Fri, 11 Dec 2009 20:23:04 +0000 (12:23 -0800)
Creates RAM based block devices (/dev/ramzswapX) which can be
used (only) as swap disks. Pages swapped to these are compressed
and stored in memory itself.

The module is called ramzswap.ko. It depends on:
 - xvmalloc memory allocator (compiled with this driver)
 - lzo_compress.ko
 - lzo_decompress.ko

See ramzswap.txt for usage details.

Signed-off-by: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
drivers/staging/Kconfig
drivers/staging/Makefile
drivers/staging/ramzswap/Kconfig [new file with mode: 0644]
drivers/staging/ramzswap/Makefile [new file with mode: 0644]
drivers/staging/ramzswap/ramzswap_drv.c [new file with mode: 0644]
drivers/staging/ramzswap/ramzswap_drv.h [new file with mode: 0644]
drivers/staging/ramzswap/ramzswap_ioctl.h [new file with mode: 0644]

index 37ec2138d3854341b8f29ffcbd67ee08ffad82d2..f012304e22a8306d71c4ad452f595659f2e44d6d 100644 (file)
@@ -123,6 +123,8 @@ source "drivers/staging/sep/Kconfig"
 
 source "drivers/staging/iio/Kconfig"
 
+source "drivers/staging/ramzswap/Kconfig"
+
 source "drivers/staging/strip/Kconfig"
 
 source "drivers/staging/arlan/Kconfig"
index ff7222d7ed8f8bd7ae4ca012e7ca4b39fab2b87a..0c0dd53e5027a3094546f414d4d5bb48c6d69cbe 100644 (file)
@@ -43,6 +43,7 @@ obj-$(CONFIG_VME_BUS)         += vme/
 obj-$(CONFIG_RAR_REGISTER)     += rar/
 obj-$(CONFIG_DX_SEP)           += sep/
 obj-$(CONFIG_IIO)              += iio/
+obj-$(CONFIG_RAMZSWAP)         += ramzswap/
 obj-$(CONFIG_STRIP)            += strip/
 obj-$(CONFIG_ARLAN)            += arlan/
 obj-$(CONFIG_WAVELAN)          += wavelan/
diff --git a/drivers/staging/ramzswap/Kconfig b/drivers/staging/ramzswap/Kconfig
new file mode 100644 (file)
index 0000000..24e2569
--- /dev/null
@@ -0,0 +1,21 @@
+config RAMZSWAP
+       tristate "Compressed in-memory swap device (ramzswap)"
+       depends on SWAP
+       select LZO_COMPRESS
+       select LZO_DECOMPRESS
+       default n
+       help
+         Creates virtual block devices which can be used (only) as a swap
+         disks. Pages swapped to these disks are compressed and stored in
+         memory itself.
+
+         See ramzswap.txt for more information.
+         Project home: http://compcache.googlecode.com/
+
+config RAMZSWAP_STATS
+       bool "Enable ramzswap stats"
+       depends on RAMZSWAP
+       default y
+       help
+         Enable statistics collection for ramzswap. This adds only a minimal
+         overhead. In unsure, say Y.
diff --git a/drivers/staging/ramzswap/Makefile b/drivers/staging/ramzswap/Makefile
new file mode 100644 (file)
index 0000000..507d7dc
--- /dev/null
@@ -0,0 +1,3 @@
+ramzswap-objs  :=      ramzswap_drv.o xvmalloc.o
+
+obj-$(CONFIG_RAMZSWAP) +=      ramzswap.o
diff --git a/drivers/staging/ramzswap/ramzswap_drv.c b/drivers/staging/ramzswap/ramzswap_drv.c
new file mode 100644 (file)
index 0000000..b839f05
--- /dev/null
@@ -0,0 +1,1435 @@
+/*
+ * Compressed RAM based swap device
+ *
+ * Copyright (C) 2008, 2009  Nitin Gupta
+ *
+ * This code is released using a dual license strategy: BSD/GPL
+ * You can choose the licence that better fits your requirements.
+ *
+ * Released under the terms of 3-clause BSD License
+ * Released under the terms of GNU General Public License Version 2.0
+ *
+ * Project home: http://compcache.googlecode.com
+ */
+
+#define KMSG_COMPONENT "ramzswap"
+#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/bitops.h>
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/device.h>
+#include <linux/genhd.h>
+#include <linux/highmem.h>
+#include <linux/lzo.h>
+#include <linux/mutex.h>
+#include <linux/string.h>
+#include <linux/swap.h>
+#include <linux/swapops.h>
+#include <linux/vmalloc.h>
+#include <linux/version.h>
+
+#include "ramzswap_drv.h"
+
+/* Globals */
+static int ramzswap_major;
+static struct ramzswap *devices;
+
+/*
+ * Pages that compress to larger than this size are
+ * forwarded to backing swap, if present or stored
+ * uncompressed in memory otherwise.
+ */
+static unsigned int max_zpage_size;
+
+/* Module params (documentation at end) */
+static unsigned int num_devices;
+
+static int rzs_test_flag(struct ramzswap *rzs, u32 index,
+                       enum rzs_pageflags flag)
+{
+       return rzs->table[index].flags & BIT(flag);
+}
+
+static void rzs_set_flag(struct ramzswap *rzs, u32 index,
+                       enum rzs_pageflags flag)
+{
+       rzs->table[index].flags |= BIT(flag);
+}
+
+static void rzs_clear_flag(struct ramzswap *rzs, u32 index,
+                       enum rzs_pageflags flag)
+{
+       rzs->table[index].flags &= ~BIT(flag);
+}
+
+static int page_zero_filled(void *ptr)
+{
+       unsigned int pos;
+       unsigned long *page;
+
+       page = (unsigned long *)ptr;
+
+       for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
+               if (page[pos])
+                       return 0;
+       }
+
+       return 1;
+}
+
+/*
+ * memlimit cannot be greater than backing disk size.
+ */
+static void ramzswap_set_memlimit(struct ramzswap *rzs, size_t totalram_bytes)
+{
+       int memlimit_valid = 1;
+
+       if (!rzs->memlimit) {
+               pr_info("Memory limit not set.\n");
+               memlimit_valid = 0;
+       }
+
+       if (rzs->memlimit > rzs->disksize) {
+               pr_info("Memory limit cannot be greater than "
+                       "disksize: limit=%zu, disksize=%zu\n",
+                       rzs->memlimit, rzs->disksize);
+               memlimit_valid = 0;
+       }
+
+       if (!memlimit_valid) {
+               size_t mempart, disksize;
+               pr_info("Using default: smaller of (%u%% of RAM) and "
+                       "(backing disk size).\n",
+                       default_memlimit_perc_ram);
+               mempart = default_memlimit_perc_ram * (totalram_bytes / 100);
+               disksize = rzs->disksize;
+               rzs->memlimit = mempart > disksize ? disksize : mempart;
+       }
+
+       if (rzs->memlimit > totalram_bytes / 2) {
+               pr_info(
+               "Its not advisable setting limit more than half of "
+               "size of memory since we expect a 2:1 compression ratio. "
+               "Limit represents amount of *compressed* data we can keep "
+               "in memory!\n"
+               "\tMemory Size: %zu kB\n"
+               "\tLimit you selected: %zu kB\n"
+               "Continuing anyway ...\n",
+               totalram_bytes >> 10, rzs->memlimit >> 10
+               );
+       }
+
+       rzs->memlimit &= PAGE_MASK;
+       BUG_ON(!rzs->memlimit);
+}
+
+static void ramzswap_set_disksize(struct ramzswap *rzs, size_t totalram_bytes)
+{
+       if (!rzs->disksize) {
+               pr_info(
+               "disk size not provided. You can use disksize_kb module "
+               "param to specify size.\nUsing default: (%u%% of RAM).\n",
+               default_disksize_perc_ram
+               );
+               rzs->disksize = default_disksize_perc_ram *
+                                       (totalram_bytes / 100);
+       }
+
+       if (rzs->disksize > 2 * (totalram_bytes)) {
+               pr_info(
+               "There is little point creating a ramzswap of greater than "
+               "twice the size of memory since we expect a 2:1 compression "
+               "ratio. Note that ramzswap uses about 0.1%% of the size of "
+               "the swap device when not in use so a huge ramzswap is "
+               "wasteful.\n"
+               "\tMemory Size: %zu kB\n"
+               "\tSize you selected: %zu kB\n"
+               "Continuing anyway ...\n",
+               totalram_bytes >> 10, rzs->disksize
+               );
+       }
+
+       rzs->disksize &= PAGE_MASK;
+}
+
+/*
+ * Swap header (1st page of swap device) contains information
+ * to indentify it as a swap partition. Prepare such a header
+ * for ramzswap device (ramzswap0) so that swapon can identify
+ * it as swap partition. In case backing swap device is provided,
+ * copy its swap header.
+ */
+static int setup_swap_header(struct ramzswap *rzs, union swap_header *s)
+{
+       int ret = 0;
+       struct page *page;
+       struct address_space *mapping;
+       union swap_header *backing_swap_header;
+
+       /*
+        * There is no backing swap device. Create a swap header
+        * that is acceptable by swapon.
+        */
+       if (!rzs->backing_swap) {
+               s->info.version = 1;
+               s->info.last_page = (rzs->disksize >> PAGE_SHIFT) - 1;
+               s->info.nr_badpages = 0;
+               memcpy(s->magic.magic, "SWAPSPACE2", 10);
+               return 0;
+       }
+
+       /*
+        * We have a backing swap device. Copy its swap header
+        * to ramzswap device header. If this header contains
+        * invalid information (backing device not a swap
+        * partition, etc.), swapon will fail for ramzswap
+        * which is correct behavior - we don't want to swap
+        * over filesystem partition!
+        */
+
+       /* Read the backing swap header (code from sys_swapon) */
+       mapping = rzs->swap_file->f_mapping;
+       if (!mapping->a_ops->readpage) {
+               ret = -EINVAL;
+               goto out;
+       }
+
+       page = read_mapping_page(mapping, 0, rzs->swap_file);
+       if (IS_ERR(page)) {
+               ret = PTR_ERR(page);
+               goto out;
+       }
+
+       backing_swap_header = kmap(page);
+       memcpy(s, backing_swap_header, sizeof(*s));
+       if (s->info.nr_badpages) {
+               pr_info("Cannot use backing swap with bad pages (%u)\n",
+                       s->info.nr_badpages);
+               ret = -EINVAL;
+       }
+       /*
+        * ramzswap disksize equals number of usable pages in backing
+        * swap. Set last_page in swap header to match this disksize
+        * ('last_page' means 0-based index of last usable swap page).
+        */
+       s->info.last_page = (rzs->disksize >> PAGE_SHIFT) - 1;
+       kunmap(page);
+
+out:
+       return ret;
+}
+
+static void ramzswap_flush_dcache_page(struct page *page)
+{
+#ifdef CONFIG_ARM
+       int flag = 0;
+       /*
+        * Ugly hack to get flush_dcache_page() work on ARM.
+        * page_mapping(page) == NULL after clearing this swap cache flag.
+        * Without clearing this flag, flush_dcache_page() will simply set
+        * "PG_dcache_dirty" bit and return.
+        */
+       if (PageSwapCache(page)) {
+               flag = 1;
+               ClearPageSwapCache(page);
+       }
+#endif
+       flush_dcache_page(page);
+#ifdef CONFIG_ARM
+       if (flag)
+               SetPageSwapCache(page);
+#endif
+}
+
+void ramzswap_ioctl_get_stats(struct ramzswap *rzs,
+                       struct ramzswap_ioctl_stats *s)
+{
+       strncpy(s->backing_swap_name, rzs->backing_swap_name,
+               MAX_SWAP_NAME_LEN - 1);
+       s->backing_swap_name[MAX_SWAP_NAME_LEN - 1] = '\0';
+
+       s->disksize = rzs->disksize;
+       s->memlimit = rzs->memlimit;
+
+#if defined(CONFIG_RAMZSWAP_STATS)
+       {
+       struct ramzswap_stats *rs = &rzs->stats;
+       size_t succ_writes, mem_used;
+       unsigned int good_compress_perc = 0, no_compress_perc = 0;
+
+       mem_used = xv_get_total_size_bytes(rzs->mem_pool)
+                       + (rs->pages_expand << PAGE_SHIFT);
+       succ_writes = rs->num_writes - rs->failed_writes;
+
+       if (succ_writes && rs->pages_stored) {
+               good_compress_perc = rs->good_compress * 100
+                                       / rs->pages_stored;
+               no_compress_perc = rs->pages_expand * 100
+                                       / rs->pages_stored;
+       }
+
+       s->num_reads = rs->num_reads;
+       s->num_writes = rs->num_writes;
+       s->failed_reads = rs->failed_reads;
+       s->failed_writes = rs->failed_writes;
+       s->invalid_io = rs->invalid_io;
+       s->pages_zero = rs->pages_zero;
+
+       s->good_compress_pct = good_compress_perc;
+       s->pages_expand_pct = no_compress_perc;
+
+       s->pages_stored = rs->pages_stored;
+       s->pages_used = mem_used >> PAGE_SHIFT;
+       s->orig_data_size = rs->pages_stored << PAGE_SHIFT;
+       s->compr_data_size = rs->compr_size;
+       s->mem_used_total = mem_used;
+
+       s->bdev_num_reads = rs->bdev_num_reads;
+       s->bdev_num_writes = rs->bdev_num_writes;
+       }
+#endif /* CONFIG_RAMZSWAP_STATS */
+}
+
+static int add_backing_swap_extent(struct ramzswap *rzs,
+                               pgoff_t phy_pagenum,
+                               pgoff_t num_pages)
+{
+       unsigned int idx;
+       struct list_head *head;
+       struct page *curr_page, *new_page;
+       unsigned int extents_per_page = PAGE_SIZE /
+                               sizeof(struct ramzswap_backing_extent);
+
+       idx = rzs->num_extents % extents_per_page;
+       if (!idx) {
+               new_page = alloc_page(__GFP_ZERO);
+               if (!new_page)
+                       return -ENOMEM;
+
+               if (rzs->num_extents) {
+                       curr_page = virt_to_page(rzs->curr_extent);
+                       head = &curr_page->lru;
+               } else {
+                       head = &rzs->backing_swap_extent_list;
+               }
+
+               list_add(&new_page->lru, head);
+               rzs->curr_extent = page_address(new_page);
+       }
+
+       rzs->curr_extent->phy_pagenum = phy_pagenum;
+       rzs->curr_extent->num_pages = num_pages;
+
+       pr_debug("add_extent: idx=%u, phy_pgnum=%lu, num_pgs=%lu, "
+               "pg_last=%lu, curr_ext=%p\n", idx, phy_pagenum, num_pages,
+               phy_pagenum + num_pages - 1, rzs->curr_extent);
+
+       if (idx != extents_per_page - 1)
+               rzs->curr_extent++;
+
+       return 0;
+}
+
+static int setup_backing_swap_extents(struct ramzswap *rzs,
+                               struct inode *inode, unsigned long *num_pages)
+{
+       int ret = 0;
+       unsigned blkbits;
+       unsigned blocks_per_page;
+       pgoff_t contig_pages = 0, total_pages = 0;
+       pgoff_t pagenum = 0, prev_pagenum = 0;
+       sector_t probe_block = 0;
+       sector_t last_block;
+
+       blkbits = inode->i_blkbits;
+       blocks_per_page = PAGE_SIZE >> blkbits;
+
+       last_block = i_size_read(inode) >> blkbits;
+       while (probe_block + blocks_per_page <= last_block) {
+               unsigned block_in_page;
+               sector_t first_block;
+
+               first_block = bmap(inode, probe_block);
+               if (first_block == 0)
+                       goto bad_bmap;
+
+               /* It must be PAGE_SIZE aligned on-disk */
+               if (first_block & (blocks_per_page - 1)) {
+                       probe_block++;
+                       goto probe_next;
+               }
+
+               /* All blocks within this page must be contiguous on disk */
+               for (block_in_page = 1; block_in_page < blocks_per_page;
+                                       block_in_page++) {
+                       sector_t block;
+
+                       block = bmap(inode, probe_block + block_in_page);
+                       if (block == 0)
+                               goto bad_bmap;
+                       if (block != first_block + block_in_page) {
+                               /* Discontiguity */
+                               probe_block++;
+                               goto probe_next;
+                       }
+               }
+
+               /*
+                * We found a PAGE_SIZE length, PAGE_SIZE aligned
+                * run of blocks.
+                */
+               pagenum = first_block >> (PAGE_SHIFT - blkbits);
+
+               if (total_pages && (pagenum != prev_pagenum + 1)) {
+                       ret = add_backing_swap_extent(rzs, prev_pagenum -
+                                       (contig_pages - 1), contig_pages);
+                       if (ret < 0)
+                               goto out;
+                       rzs->num_extents++;
+                       contig_pages = 0;
+               }
+               total_pages++;
+               contig_pages++;
+               prev_pagenum = pagenum;
+               probe_block += blocks_per_page;
+
+probe_next:
+               continue;
+       }
+
+       if (contig_pages) {
+               pr_debug("adding last extent: pagenum=%lu, "
+                       "contig_pages=%lu\n", pagenum, contig_pages);
+               ret = add_backing_swap_extent(rzs,
+                       prev_pagenum - (contig_pages - 1), contig_pages);
+               if (ret < 0)
+                       goto out;
+               rzs->num_extents++;
+       }
+       if (!rzs->num_extents) {
+               pr_err("No swap extents found!\n");
+               ret = -EINVAL;
+       }
+
+       if (!ret) {
+               *num_pages = total_pages;
+               pr_info("Found %lu extents containing %luk\n",
+                       rzs->num_extents, *num_pages << (PAGE_SHIFT - 10));
+       }
+       goto out;
+
+bad_bmap:
+       pr_err("Backing swapfile has holes\n");
+       ret = -EINVAL;
+out:
+       while (ret && !list_empty(&rzs->backing_swap_extent_list)) {
+               struct page *page;
+               struct list_head *entry = rzs->backing_swap_extent_list.next;
+               page = list_entry(entry, struct page, lru);
+               list_del(entry);
+               __free_page(page);
+       }
+       return ret;
+}
+
+static void map_backing_swap_extents(struct ramzswap *rzs)
+{
+       struct ramzswap_backing_extent *se;
+       struct page *table_page, *se_page;
+       unsigned long num_pages, num_table_pages, entry;
+       unsigned long se_idx, span;
+       unsigned entries_per_page = PAGE_SIZE / sizeof(*rzs->table);
+       unsigned extents_per_page = PAGE_SIZE / sizeof(*se);
+
+       /* True for block device */
+       if (!rzs->num_extents)
+               return;
+
+       se_page = list_entry(rzs->backing_swap_extent_list.next,
+                                       struct page, lru);
+       se = page_address(se_page);
+       span = se->num_pages;
+       num_pages = rzs->disksize >> PAGE_SHIFT;
+       num_table_pages = DIV_ROUND_UP(num_pages * sizeof(*rzs->table),
+                                                       PAGE_SIZE);
+
+       entry = 0;
+       se_idx = 0;
+       while (num_table_pages--) {
+               table_page = vmalloc_to_page(&rzs->table[entry]);
+               while (span <= entry) {
+                       se_idx++;
+                       if (se_idx == rzs->num_extents)
+                               BUG();
+
+                       if (!(se_idx % extents_per_page)) {
+                               se_page = list_entry(se_page->lru.next,
+                                               struct page, lru);
+                               se = page_address(se_page);
+                       } else
+                               se++;
+
+                       span += se->num_pages;
+               }
+               table_page->mapping = (struct address_space *)se;
+               table_page->private = se->num_pages - (span - entry);
+               pr_debug("map_table: entry=%lu, span=%lu, map=%p, priv=%lu\n",
+                       entry, span, table_page->mapping, table_page->private);
+               entry += entries_per_page;
+       }
+}
+
+/*
+ * Check if value of backing_swap module param is sane.
+ * Claim this device and set ramzswap size equal to
+ * size of this block device.
+ */
+static int setup_backing_swap(struct ramzswap *rzs)
+{
+       int ret = 0;
+       size_t disksize;
+       unsigned long num_pages = 0;
+       struct inode *inode;
+       struct file *swap_file;
+       struct address_space *mapping;
+       struct block_device *bdev = NULL;
+
+       if (!rzs->backing_swap_name[0]) {
+               pr_debug("backing_swap param not given\n");
+               goto out;
+       }
+
+       pr_info("Using backing swap device: %s\n", rzs->backing_swap_name);
+
+       swap_file = filp_open(rzs->backing_swap_name,
+                               O_RDWR | O_LARGEFILE, 0);
+       if (IS_ERR(swap_file)) {
+               pr_err("Error opening backing device: %s\n",
+                       rzs->backing_swap_name);
+               ret = -EINVAL;
+               goto out;
+       }
+
+       mapping = swap_file->f_mapping;
+       inode = mapping->host;
+
+       if (S_ISBLK(inode->i_mode)) {
+               bdev = I_BDEV(inode);
+               ret = bd_claim(bdev, setup_backing_swap);
+               if (ret < 0) {
+                       bdev = NULL;
+                       goto bad_param;
+               }
+               disksize = i_size_read(inode);
+       } else if (S_ISREG(inode->i_mode)) {
+               bdev = inode->i_sb->s_bdev;
+               if (IS_SWAPFILE(inode)) {
+                       ret = -EBUSY;
+                       goto bad_param;
+               }
+               ret = setup_backing_swap_extents(rzs, inode, &num_pages);
+               if (ret < 0)
+                       goto bad_param;
+               disksize = num_pages << PAGE_SHIFT;
+       } else {
+               goto bad_param;
+       }
+
+       rzs->swap_file = swap_file;
+       rzs->backing_swap = bdev;
+       rzs->disksize = disksize;
+       BUG_ON(!rzs->disksize);
+
+       return 0;
+
+bad_param:
+       if (bdev)
+               bd_release(bdev);
+       filp_close(swap_file, NULL);
+
+out:
+       rzs->backing_swap = NULL;
+       return ret;
+}
+
+/*
+ * Map logical page number 'pagenum' to physical page number
+ * on backing swap device. For block device, this is a nop.
+ */
+u32 map_backing_swap_page(struct ramzswap *rzs, u32 pagenum)
+{
+       u32 skip_pages, entries_per_page;
+       size_t delta, se_offset, skipped;
+       struct page *table_page, *se_page;
+       struct ramzswap_backing_extent *se;
+
+       if (!rzs->num_extents)
+               return pagenum;
+
+       entries_per_page = PAGE_SIZE / sizeof(*rzs->table);
+
+       table_page = vmalloc_to_page(&rzs->table[pagenum]);
+       se = (struct ramzswap_backing_extent *)table_page->mapping;
+       se_page = virt_to_page(se);
+
+       skip_pages = pagenum - (pagenum / entries_per_page * entries_per_page);
+       se_offset = table_page->private + skip_pages;
+
+       if (se_offset < se->num_pages)
+               return se->phy_pagenum + se_offset;
+
+       skipped = se->num_pages - table_page->private;
+       do {
+               struct ramzswap_backing_extent *se_base;
+               u32 se_entries_per_page = PAGE_SIZE / sizeof(*se);
+
+               /* Get next swap extent */
+               se_base = (struct ramzswap_backing_extent *)
+                                               page_address(se_page);
+               if (se - se_base == se_entries_per_page - 1) {
+                       se_page = list_entry(se_page->lru.next,
+                                               struct page, lru);
+                       se = page_address(se_page);
+               } else {
+                       se++;
+               }
+
+               skipped += se->num_pages;
+       } while (skipped < skip_pages);
+
+       delta = skipped - skip_pages;
+       se_offset = se->num_pages - delta;
+
+       return se->phy_pagenum + se_offset;
+}
+
+static void ramzswap_free_page(struct ramzswap *rzs, size_t index)
+{
+       u32 clen;
+       void *obj;
+
+       struct page *page = rzs->table[index].page;
+       u32 offset = rzs->table[index].offset;
+
+       if (unlikely(!page)) {
+               if (rzs_test_flag(rzs, index, RZS_ZERO)) {
+                       rzs_clear_flag(rzs, index, RZS_ZERO);
+                       stat_dec(rzs->stats.pages_zero);
+               }
+               return;
+       }
+
+       if (unlikely(rzs_test_flag(rzs, index, RZS_UNCOMPRESSED))) {
+               clen = PAGE_SIZE;
+               __free_page(page);
+               rzs_clear_flag(rzs, index, RZS_UNCOMPRESSED);
+               stat_dec(rzs->stats.pages_expand);
+               goto out;
+       }
+
+       obj = kmap_atomic(page, KM_USER0) + offset;
+       clen = xv_get_object_size(obj) - sizeof(struct zobj_header);
+       kunmap_atomic(obj, KM_USER0);
+
+       xv_free(rzs->mem_pool, page, offset);
+       if (clen <= PAGE_SIZE / 2)
+               stat_dec(rzs->stats.good_compress);
+
+out:
+       rzs->stats.compr_size -= clen;
+       stat_dec(rzs->stats.pages_stored);
+
+       rzs->table[index].page = NULL;
+       rzs->table[index].offset = 0;
+}
+
+static int handle_zero_page(struct bio *bio)
+{
+       void *user_mem;
+       struct page *page = bio->bi_io_vec[0].bv_page;
+
+       user_mem = kmap_atomic(page, KM_USER0);
+       memset(user_mem, 0, PAGE_SIZE);
+       kunmap_atomic(user_mem, KM_USER0);
+
+       ramzswap_flush_dcache_page(page);
+
+       set_bit(BIO_UPTODATE, &bio->bi_flags);
+       bio_endio(bio, 0);
+       return 0;
+}
+
+static int handle_uncompressed_page(struct ramzswap *rzs, struct bio *bio)
+{
+       u32 index;
+       struct page *page;
+       unsigned char *user_mem, *cmem;
+
+       page = bio->bi_io_vec[0].bv_page;
+       index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
+
+       user_mem = kmap_atomic(page, KM_USER0);
+       cmem = kmap_atomic(rzs->table[index].page, KM_USER1) +
+                       rzs->table[index].offset;
+
+       memcpy(user_mem, cmem, PAGE_SIZE);
+       kunmap_atomic(user_mem, KM_USER0);
+       kunmap_atomic(cmem, KM_USER1);
+
+       ramzswap_flush_dcache_page(page);
+
+       set_bit(BIO_UPTODATE, &bio->bi_flags);
+       bio_endio(bio, 0);
+       return 0;
+}
+
+
+/*
+ * Called when request page is not present in ramzswap.
+ * Its either in backing swap device (if present) or
+ * this is an attempt to read before any previous write
+ * to this location - this happens due to readahead when
+ * swap device is read from user-space (e.g. during swapon)
+ */
+static int handle_ramzswap_fault(struct ramzswap *rzs, struct bio *bio)
+{
+       /*
+        * Always forward such requests to backing swap
+        * device (if present)
+        */
+       if (rzs->backing_swap) {
+               u32 pagenum;
+               stat_dec(rzs->stats.num_reads);
+               stat_inc(rzs->stats.bdev_num_reads);
+               bio->bi_bdev = rzs->backing_swap;
+
+               /*
+                * In case backing swap is a file, find the right offset within
+                * the file corresponding to logical position 'index'. For block
+                * device, this is a nop.
+                */
+               pagenum = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
+               bio->bi_sector = map_backing_swap_page(rzs, pagenum)
+                                       << SECTORS_PER_PAGE_SHIFT;
+               return 1;
+       }
+
+       /*
+        * Its unlikely event in case backing dev is
+        * not present
+        */
+       pr_debug("Read before write on swap device: "
+               "sector=%lu, size=%u, offset=%u\n",
+               (ulong)(bio->bi_sector), bio->bi_size,
+               bio->bi_io_vec[0].bv_offset);
+
+       /* Do nothing. Just return success */
+       set_bit(BIO_UPTODATE, &bio->bi_flags);
+       bio_endio(bio, 0);
+       return 0;
+}
+
+static int ramzswap_read(struct ramzswap *rzs, struct bio *bio)
+{
+       int ret;
+       u32 index;
+       size_t clen;
+       struct page *page;
+       struct zobj_header *zheader;
+       unsigned char *user_mem, *cmem;
+
+       stat_inc(rzs->stats.num_reads);
+
+       page = bio->bi_io_vec[0].bv_page;
+       index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
+
+       if (rzs_test_flag(rzs, index, RZS_ZERO))
+               return handle_zero_page(bio);
+
+       /* Requested page is not present in compressed area */
+       if (!rzs->table[index].page)
+               return handle_ramzswap_fault(rzs, bio);
+
+       /* Page is stored uncompressed since its incompressible */
+       if (unlikely(rzs_test_flag(rzs, index, RZS_UNCOMPRESSED)))
+               return handle_uncompressed_page(rzs, bio);
+
+       user_mem = kmap_atomic(page, KM_USER0);
+       clen = PAGE_SIZE;
+
+       cmem = kmap_atomic(rzs->table[index].page, KM_USER1) +
+                       rzs->table[index].offset;
+
+       ret = lzo1x_decompress_safe(
+               cmem + sizeof(*zheader),
+               xv_get_object_size(cmem) - sizeof(*zheader),
+               user_mem, &clen);
+
+       kunmap_atomic(user_mem, KM_USER0);
+       kunmap_atomic(cmem, KM_USER1);
+
+       /* should NEVER happen */
+       if (unlikely(ret != LZO_E_OK)) {
+               pr_err("Decompression failed! err=%d, page=%u\n",
+                       ret, index);
+               stat_inc(rzs->stats.failed_reads);
+               goto out;
+       }
+
+       ramzswap_flush_dcache_page(page);
+
+       set_bit(BIO_UPTODATE, &bio->bi_flags);
+       bio_endio(bio, 0);
+       return 0;
+
+out:
+       bio_io_error(bio);
+       return 0;
+}
+
+static int ramzswap_write(struct ramzswap *rzs, struct bio *bio)
+{
+       int ret, fwd_write_request = 0;
+       u32 offset, index;
+       size_t clen;
+       struct zobj_header *zheader;
+       struct page *page, *page_store;
+       unsigned char *user_mem, *cmem, *src;
+
+       stat_inc(rzs->stats.num_writes);
+
+       page = bio->bi_io_vec[0].bv_page;
+       index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
+
+       src = rzs->compress_buffer;
+
+       /*
+        * System swaps to same sector again when the stored page
+        * is no longer referenced by any process. So, its now safe
+        * to free the memory that was allocated for this page.
+        */
+       if (rzs->table[index].page)
+               ramzswap_free_page(rzs, index);
+
+       /*
+        * No memory ia allocated for zero filled pages.
+        * Simply clear zero page flag.
+        */
+       if (rzs_test_flag(rzs, index, RZS_ZERO)) {
+               stat_dec(rzs->stats.pages_zero);
+               rzs_clear_flag(rzs, index, RZS_ZERO);
+       }
+
+       mutex_lock(&rzs->lock);
+
+       user_mem = kmap_atomic(page, KM_USER0);
+       if (page_zero_filled(user_mem)) {
+               kunmap_atomic(user_mem, KM_USER0);
+               mutex_unlock(&rzs->lock);
+               stat_inc(rzs->stats.pages_zero);
+               rzs_set_flag(rzs, index, RZS_ZERO);
+
+               set_bit(BIO_UPTODATE, &bio->bi_flags);
+               bio_endio(bio, 0);
+               return 0;
+       }
+
+       if (rzs->backing_swap &&
+               (rzs->stats.compr_size > rzs->memlimit - PAGE_SIZE)) {
+               kunmap_atomic(user_mem, KM_USER0);
+               mutex_unlock(&rzs->lock);
+               fwd_write_request = 1;
+               goto out;
+       }
+
+       ret = lzo1x_1_compress(user_mem, PAGE_SIZE, src, &clen,
+                               rzs->compress_workmem);
+
+       kunmap_atomic(user_mem, KM_USER0);
+
+       if (unlikely(ret != LZO_E_OK)) {
+               mutex_unlock(&rzs->lock);
+               pr_err("Compression failed! err=%d\n", ret);
+               stat_inc(rzs->stats.failed_writes);
+               goto out;
+       }
+
+       /*
+        * Page is incompressible. Forward it to backing swap
+        * if present. Otherwise, store it as-is (uncompressed)
+        * since we do not want to return too many swap write
+        * errors which has side effect of hanging the system.
+        */
+       if (unlikely(clen > max_zpage_size)) {
+               if (rzs->backing_swap) {
+                       mutex_unlock(&rzs->lock);
+                       fwd_write_request = 1;
+                       goto out;
+               }
+
+               clen = PAGE_SIZE;
+               page_store = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
+               if (unlikely(!page_store)) {
+                       mutex_unlock(&rzs->lock);
+                       pr_info("Error allocating memory for incompressible "
+                               "page: %u\n", index);
+                       stat_inc(rzs->stats.failed_writes);
+                       goto out;
+               }
+
+               offset = 0;
+               rzs_set_flag(rzs, index, RZS_UNCOMPRESSED);
+               stat_inc(rzs->stats.pages_expand);
+               rzs->table[index].page = page_store;
+               src = kmap_atomic(page, KM_USER0);
+               goto memstore;
+       }
+
+       if (xv_malloc(rzs->mem_pool, clen + sizeof(*zheader),
+                       &rzs->table[index].page, &offset,
+                       GFP_NOIO | __GFP_HIGHMEM)) {
+               mutex_unlock(&rzs->lock);
+               pr_info("Error allocating memory for compressed "
+                       "page: %u, size=%zu\n", index, clen);
+               stat_inc(rzs->stats.failed_writes);
+               if (rzs->backing_swap)
+                       fwd_write_request = 1;
+               goto out;
+       }
+
+memstore:
+       rzs->table[index].offset = offset;
+
+       cmem = kmap_atomic(rzs->table[index].page, KM_USER1) +
+                       rzs->table[index].offset;
+
+#if 0
+       /* Back-reference needed for memory defragmentation */
+       if (!rzs_test_flag(rzs, index, RZS_UNCOMPRESSED)) {
+               zheader = (struct zobj_header *)cmem;
+               zheader->table_idx = index;
+               cmem += sizeof(*zheader);
+       }
+#endif
+
+       memcpy(cmem, src, clen);
+
+       kunmap_atomic(cmem, KM_USER1);
+       if (unlikely(rzs_test_flag(rzs, index, RZS_UNCOMPRESSED)))
+               kunmap_atomic(src, KM_USER0);
+
+       /* Update stats */
+       rzs->stats.compr_size += clen;
+       stat_inc(rzs->stats.pages_stored);
+       if (clen <= PAGE_SIZE / 2)
+               stat_inc(rzs->stats.good_compress);
+
+       mutex_unlock(&rzs->lock);
+
+       set_bit(BIO_UPTODATE, &bio->bi_flags);
+       bio_endio(bio, 0);
+       return 0;
+
+out:
+       if (fwd_write_request) {
+               stat_inc(rzs->stats.bdev_num_writes);
+               bio->bi_bdev = rzs->backing_swap;
+#if 0
+               /*
+                * TODO: We currently have linear mapping of ramzswap and
+                * backing swap sectors. This is not desired since we want
+                * to optimize writes to backing swap to minimize disk seeks
+                * or have effective wear leveling (for SSDs). Also, a
+                * non-linear mapping is required to implement compressed
+                * on-disk swapping.
+                */
+                bio->bi_sector = get_backing_swap_page()
+                                       << SECTORS_PER_PAGE_SHIFT;
+#endif
+               /*
+                * In case backing swap is a file, find the right offset within
+                * the file corresponding to logical position 'index'. For block
+                * device, this is a nop.
+                */
+               bio->bi_sector = map_backing_swap_page(rzs, index)
+                                       << SECTORS_PER_PAGE_SHIFT;
+               return 1;
+       }
+
+       bio_io_error(bio);
+       return 0;
+}
+
+
+/*
+ * Check if request is within bounds and page aligned.
+ */
+static inline int valid_swap_request(struct ramzswap *rzs, struct bio *bio)
+{
+       if (unlikely(
+               (bio->bi_sector >= (rzs->disksize >> SECTOR_SHIFT)) ||
+               (bio->bi_sector & (SECTORS_PER_PAGE - 1)) ||
+               (bio->bi_vcnt != 1) ||
+               (bio->bi_size != PAGE_SIZE) ||
+               (bio->bi_io_vec[0].bv_offset != 0))) {
+
+               return 0;
+       }
+
+       /* swap request is valid */
+       return 1;
+}
+
+/*
+ * Handler function for all ramzswap I/O requests.
+ */
+static int ramzswap_make_request(struct request_queue *queue, struct bio *bio)
+{
+       int ret = 0;
+       struct ramzswap *rzs = queue->queuedata;
+
+       if (unlikely(!rzs->init_done)) {
+               bio_io_error(bio);
+               return 0;
+       }
+
+       if (!valid_swap_request(rzs, bio)) {
+               stat_inc(rzs->stats.invalid_io);
+               bio_io_error(bio);
+               return 0;
+       }
+
+       switch (bio_data_dir(bio)) {
+       case READ:
+               ret = ramzswap_read(rzs, bio);
+               break;
+
+       case WRITE:
+               ret = ramzswap_write(rzs, bio);
+               break;
+       }
+
+       return ret;
+}
+
+static void reset_device(struct ramzswap *rzs)
+{
+       int is_backing_blkdev = 0;
+       size_t index, num_pages;
+       unsigned entries_per_page;
+       unsigned long num_table_pages, entry = 0;
+
+       if (rzs->backing_swap && !rzs->num_extents)
+               is_backing_blkdev = 1;
+
+       num_pages = rzs->disksize >> PAGE_SHIFT;
+
+       /* Free various per-device buffers */
+       kfree(rzs->compress_workmem);
+       free_pages((unsigned long)rzs->compress_buffer, 1);
+
+       rzs->compress_workmem = NULL;
+       rzs->compress_buffer = NULL;
+
+       /* Free all pages that are still in this ramzswap device */
+       for (index = 0; index < num_pages; index++) {
+               struct page *page;
+               u16 offset;
+
+               page = rzs->table[index].page;
+               offset = rzs->table[index].offset;
+
+               if (!page)
+                       continue;
+
+               if (unlikely(rzs_test_flag(rzs, index, RZS_UNCOMPRESSED)))
+                       __free_page(page);
+               else
+                       xv_free(rzs->mem_pool, page, offset);
+       }
+
+       entries_per_page = PAGE_SIZE / sizeof(*rzs->table);
+       num_table_pages = DIV_ROUND_UP(num_pages * sizeof(*rzs->table),
+                                       PAGE_SIZE);
+       /*
+        * Set page->mapping to NULL for every table page.
+        * Otherwise, we will hit bad_page() during free.
+        */
+       while (rzs->num_extents && num_table_pages--) {
+               struct page *page;
+               page = vmalloc_to_page(&rzs->table[entry]);
+               page->mapping = NULL;
+               entry += entries_per_page;
+       }
+       vfree(rzs->table);
+       rzs->table = NULL;
+
+       xv_destroy_pool(rzs->mem_pool);
+       rzs->mem_pool = NULL;
+
+       /* Free all swap extent pages */
+       while (!list_empty(&rzs->backing_swap_extent_list)) {
+               struct page *page;
+               struct list_head *entry;
+               entry = rzs->backing_swap_extent_list.next;
+               page = list_entry(entry, struct page, lru);
+               list_del(entry);
+               __free_page(page);
+       }
+       INIT_LIST_HEAD(&rzs->backing_swap_extent_list);
+       rzs->num_extents = 0;
+
+       /* Close backing swap device, if present */
+       if (rzs->backing_swap) {
+               if (is_backing_blkdev)
+                       bd_release(rzs->backing_swap);
+               filp_close(rzs->swap_file, NULL);
+               rzs->backing_swap = NULL;
+       }
+
+       /* Reset stats */
+       memset(&rzs->stats, 0, sizeof(rzs->stats));
+
+       rzs->disksize = 0;
+       rzs->memlimit = 0;
+
+       /* Back to uninitialized state */
+       rzs->init_done = 0;
+}
+
+static int ramzswap_ioctl_init_device(struct ramzswap *rzs)
+{
+       int ret;
+       size_t num_pages;
+       struct page *page;
+       union swap_header *swap_header;
+
+       if (rzs->init_done) {
+               pr_info("Device already initialized!\n");
+               return -EBUSY;
+       }
+
+       ret = setup_backing_swap(rzs);
+       if (ret)
+               goto fail;
+
+       if (rzs->backing_swap)
+               ramzswap_set_memlimit(rzs, totalram_pages << PAGE_SHIFT);
+       else
+               ramzswap_set_disksize(rzs, totalram_pages << PAGE_SHIFT);
+
+       rzs->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
+       if (!rzs->compress_workmem) {
+               pr_err("Error allocating compressor working memory!\n");
+               ret = -ENOMEM;
+               goto fail;
+       }
+
+       rzs->compress_buffer = (void *)__get_free_pages(__GFP_ZERO, 1);
+       if (!rzs->compress_buffer) {
+               pr_err("Error allocating compressor buffer space\n");
+               ret = -ENOMEM;
+               goto fail;
+       }
+
+       num_pages = rzs->disksize >> PAGE_SHIFT;
+       rzs->table = vmalloc(num_pages * sizeof(*rzs->table));
+       if (!rzs->table) {
+               pr_err("Error allocating ramzswap address table\n");
+               /* To prevent accessing table entries during cleanup */
+               rzs->disksize = 0;
+               ret = -ENOMEM;
+               goto fail;
+       }
+       memset(rzs->table, 0, num_pages * sizeof(*rzs->table));
+
+       map_backing_swap_extents(rzs);
+
+       page = alloc_page(__GFP_ZERO);
+       if (!page) {
+               pr_err("Error allocating swap header page\n");
+               ret = -ENOMEM;
+               goto fail;
+       }
+       rzs->table[0].page = page;
+       rzs_set_flag(rzs, 0, RZS_UNCOMPRESSED);
+
+       swap_header = kmap(page);
+       ret = setup_swap_header(rzs, swap_header);
+       kunmap(page);
+       if (ret) {
+               pr_err("Error setting swap header\n");
+               goto fail;
+       }
+
+       set_capacity(rzs->disk, rzs->disksize >> SECTOR_SHIFT);
+
+       /*
+        * We have ident mapping of sectors for ramzswap and
+        * and the backing swap device. So, this queue flag
+        * should be according to backing dev.
+        */
+       if (!rzs->backing_swap ||
+                       blk_queue_nonrot(rzs->backing_swap->bd_disk->queue))
+               queue_flag_set_unlocked(QUEUE_FLAG_NONROT, rzs->disk->queue);
+
+       rzs->mem_pool = xv_create_pool();
+       if (!rzs->mem_pool) {
+               pr_err("Error creating memory pool\n");
+               ret = -ENOMEM;
+               goto fail;
+       }
+
+       /*
+        * Pages that compress to size greater than this are forwarded
+        * to physical swap disk (if backing dev is provided)
+        * TODO: make this configurable
+        */
+       if (rzs->backing_swap)
+               max_zpage_size = max_zpage_size_bdev;
+       else
+               max_zpage_size = max_zpage_size_nobdev;
+       pr_debug("Max compressed page size: %u bytes\n", max_zpage_size);
+
+       rzs->init_done = 1;
+
+       pr_debug("Initialization done!\n");
+       return 0;
+
+fail:
+       reset_device(rzs);
+
+       pr_err("Initialization failed: err=%d\n", ret);
+       return ret;
+}
+
+static int ramzswap_ioctl_reset_device(struct ramzswap *rzs)
+{
+       if (rzs->init_done)
+               reset_device(rzs);
+
+       return 0;
+}
+
+static int ramzswap_ioctl(struct block_device *bdev, fmode_t mode,
+                       unsigned int cmd, unsigned long arg)
+{
+       int ret = 0;
+       size_t disksize_kb, memlimit_kb;
+
+       struct ramzswap *rzs = bdev->bd_disk->private_data;
+
+       switch (cmd) {
+       case RZSIO_SET_DISKSIZE_KB:
+               if (rzs->init_done) {
+                       ret = -EBUSY;
+                       goto out;
+               }
+               if (copy_from_user(&disksize_kb, (void *)arg,
+                                               _IOC_SIZE(cmd))) {
+                       ret = -EFAULT;
+                       goto out;
+               }
+               rzs->disksize = disksize_kb << 10;
+               pr_info("Disk size set to %zu kB\n", disksize_kb);
+               break;
+
+       case RZSIO_SET_MEMLIMIT_KB:
+               if (rzs->init_done) {
+                       /* TODO: allow changing memlimit */
+                       ret = -EBUSY;
+                       goto out;
+               }
+               if (copy_from_user(&memlimit_kb, (void *)arg,
+                                               _IOC_SIZE(cmd))) {
+                       ret = -EFAULT;
+                       goto out;
+               }
+               rzs->memlimit = memlimit_kb << 10;
+               pr_info("Memory limit set to %zu kB\n", memlimit_kb);
+               break;
+
+       case RZSIO_SET_BACKING_SWAP:
+               if (rzs->init_done) {
+                       ret = -EBUSY;
+                       goto out;
+               }
+
+               if (copy_from_user(&rzs->backing_swap_name, (void *)arg,
+                                               _IOC_SIZE(cmd))) {
+                       ret = -EFAULT;
+                       goto out;
+               }
+               rzs->backing_swap_name[MAX_SWAP_NAME_LEN - 1] = '\0';
+               pr_info("Backing swap set to %s\n", rzs->backing_swap_name);
+               break;
+
+       case RZSIO_GET_STATS:
+       {
+               struct ramzswap_ioctl_stats *stats;
+               if (!rzs->init_done) {
+                       ret = -ENOTTY;
+                       goto out;
+               }
+               stats = kzalloc(sizeof(*stats), GFP_KERNEL);
+               if (!stats) {
+                       ret = -ENOMEM;
+                       goto out;
+               }
+               ramzswap_ioctl_get_stats(rzs, stats);
+               if (copy_to_user((void *)arg, stats, sizeof(*stats))) {
+                       kfree(stats);
+                       ret = -EFAULT;
+                       goto out;
+               }
+               kfree(stats);
+               break;
+       }
+       case RZSIO_INIT:
+               ret = ramzswap_ioctl_init_device(rzs);
+               break;
+
+       case RZSIO_RESET:
+               /* Do not reset an active device! */
+               if (bdev->bd_holders) {
+                       ret = -EBUSY;
+                       goto out;
+               }
+               ret = ramzswap_ioctl_reset_device(rzs);
+               break;
+
+       default:
+               pr_info("Invalid ioctl %u\n", cmd);
+               ret = -ENOTTY;
+       }
+
+out:
+       return ret;
+}
+
+static struct block_device_operations ramzswap_devops = {
+       .ioctl = ramzswap_ioctl,
+       .owner = THIS_MODULE,
+};
+
+static void create_device(struct ramzswap *rzs, int device_id)
+{
+       mutex_init(&rzs->lock);
+       INIT_LIST_HEAD(&rzs->backing_swap_extent_list);
+
+       rzs->queue = blk_alloc_queue(GFP_KERNEL);
+       if (!rzs->queue) {
+               pr_err("Error allocating disk queue for device %d\n",
+                       device_id);
+               return;
+       }
+
+       blk_queue_make_request(rzs->queue, ramzswap_make_request);
+       rzs->queue->queuedata = rzs;
+
+        /* gendisk structure */
+       rzs->disk = alloc_disk(1);
+       if (!rzs->disk) {
+               blk_cleanup_queue(rzs->queue);
+               pr_warning("Error allocating disk structure for device %d\n",
+                       device_id);
+               return;
+       }
+
+       rzs->disk->major = ramzswap_major;
+       rzs->disk->first_minor = device_id;
+       rzs->disk->fops = &ramzswap_devops;
+       rzs->disk->queue = rzs->queue;
+       rzs->disk->private_data = rzs;
+       snprintf(rzs->disk->disk_name, 16, "ramzswap%d", device_id);
+
+       /*
+        * Actual capacity set using RZSIO_SET_DISKSIZE_KB ioctl
+        * or set equal to backing swap device (if provided)
+        */
+       set_capacity(rzs->disk, 0);
+       add_disk(rzs->disk);
+
+       rzs->init_done = 0;
+}
+
+static void destroy_device(struct ramzswap *rzs)
+{
+       if (rzs->disk) {
+               del_gendisk(rzs->disk);
+               put_disk(rzs->disk);
+       }
+
+       if (rzs->queue)
+               blk_cleanup_queue(rzs->queue);
+}
+
+static int __init ramzswap_init(void)
+{
+       int i, ret;
+
+       if (num_devices > max_num_devices) {
+               pr_warning("Invalid value for num_devices: %u\n",
+                               num_devices);
+               return -EINVAL;
+       }
+
+       ramzswap_major = register_blkdev(0, "ramzswap");
+       if (ramzswap_major <= 0) {
+               pr_warning("Unable to get major number\n");
+               return -EBUSY;
+       }
+
+       if (!num_devices) {
+               pr_info("num_devices not specified. Using default: 1\n");
+               num_devices = 1;
+       }
+
+       /* Allocate the device array and initialize each one */
+       pr_info("Creating %u devices ...\n", num_devices);
+       devices = kzalloc(num_devices * sizeof(struct ramzswap), GFP_KERNEL);
+       if (!devices) {
+               ret = -ENOMEM;
+               goto out;
+       }
+
+       for (i = 0; i < num_devices; i++)
+               create_device(&devices[i], i);
+
+       return 0;
+out:
+       unregister_blkdev(ramzswap_major, "ramzswap");
+       return ret;
+}
+
+static void __exit ramzswap_exit(void)
+{
+       int i;
+       struct ramzswap *rzs;
+
+       for (i = 0; i < num_devices; i++) {
+               rzs = &devices[i];
+
+               destroy_device(rzs);
+               if (rzs->init_done)
+                       reset_device(rzs);
+       }
+
+       unregister_blkdev(ramzswap_major, "ramzswap");
+
+       kfree(devices);
+       pr_debug("Cleanup done!\n");
+}
+
+module_param(num_devices, uint, 0);
+MODULE_PARM_DESC(num_devices, "Number of ramzswap devices");
+
+module_init(ramzswap_init);
+module_exit(ramzswap_exit);
+
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
+MODULE_DESCRIPTION("Compressed RAM Based Swap Device");
diff --git a/drivers/staging/ramzswap/ramzswap_drv.h b/drivers/staging/ramzswap/ramzswap_drv.h
new file mode 100644 (file)
index 0000000..a6ea240
--- /dev/null
@@ -0,0 +1,171 @@
+/*
+ * Compressed RAM based swap device
+ *
+ * Copyright (C) 2008, 2009  Nitin Gupta
+ *
+ * This code is released using a dual license strategy: BSD/GPL
+ * You can choose the licence that better fits your requirements.
+ *
+ * Released under the terms of 3-clause BSD License
+ * Released under the terms of GNU General Public License Version 2.0
+ *
+ * Project home: http://compcache.googlecode.com
+ */
+
+#ifndef _RAMZSWAP_DRV_H_
+#define _RAMZSWAP_DRV_H_
+
+#include "ramzswap_ioctl.h"
+#include "xvmalloc.h"
+
+/*
+ * Some arbitrary value. This is just to catch
+ * invalid value for num_devices module parameter.
+ */
+static const unsigned max_num_devices = 32;
+
+/*
+ * Stored at beginning of each compressed object.
+ *
+ * It stores back-reference to table entry which points to this
+ * object. This is required to support memory defragmentation or
+ * migrating compressed pages to backing swap disk.
+ */
+struct zobj_header {
+#if 0
+       u32 table_idx;
+#endif
+};
+
+/*-- Configurable parameters */
+
+/* Default ramzswap disk size: 25% of total RAM */
+static const unsigned default_disksize_perc_ram = 25;
+static const unsigned default_memlimit_perc_ram = 15;
+
+/*
+ * Max compressed page size when backing device is provided.
+ * Pages that compress to size greater than this are sent to
+ * physical swap disk.
+ */
+static const unsigned max_zpage_size_bdev = PAGE_SIZE / 2;
+
+/*
+ * Max compressed page size when there is no backing dev.
+ * Pages that compress to size greater than this are stored
+ * uncompressed in memory.
+ */
+static const unsigned max_zpage_size_nobdev = PAGE_SIZE / 4 * 3;
+
+/*
+ * NOTE: max_zpage_size_{bdev,nobdev} sizes must be
+ * less than or equal to:
+ *   XV_MAX_ALLOC_SIZE - sizeof(struct zobj_header)
+ * since otherwise xv_malloc would always return failure.
+ */
+
+/*-- End of configurable params */
+
+#define SECTOR_SHIFT           9
+#define SECTOR_SIZE            (1 << SECTOR_SHIFT)
+#define SECTORS_PER_PAGE_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
+#define SECTORS_PER_PAGE       (1 << SECTORS_PER_PAGE_SHIFT)
+
+/* Debugging and Stats */
+#if defined(CONFIG_RAMZSWAP_STATS)
+#define stat_inc(stat) ((stat)++)
+#define stat_dec(stat) ((stat)--)
+#else
+#define stat_inc(x)
+#define stat_dec(x)
+#endif
+
+/* Flags for ramzswap pages (table[page_no].flags) */
+enum rzs_pageflags {
+       /* Page is stored uncompressed */
+       RZS_UNCOMPRESSED,
+
+       /* Page consists entirely of zeros */
+       RZS_ZERO,
+
+       __NR_RZS_PAGEFLAGS,
+};
+
+/*-- Data structures */
+
+/*
+ * Allocated for each swap slot, indexed by page no.
+ * These table entries must fit exactly in a page.
+ */
+struct table {
+       struct page *page;
+       u16 offset;
+       u8 count;       /* object ref count (not yet used) */
+       u8 flags;
+} __attribute__((aligned(4)));;
+
+/*
+ * Swap extent information in case backing swap is a regular
+ * file. These extent entries must fit exactly in a page.
+ */
+struct ramzswap_backing_extent {
+       pgoff_t phy_pagenum;
+       pgoff_t num_pages;
+} __attribute__((aligned(4)));
+
+struct ramzswap_stats {
+       /* basic stats */
+       size_t compr_size;      /* compressed size of pages stored -
+                                * needed to enforce memlimit */
+       /* more stats */
+#if defined(CONFIG_RAMZSWAP_STATS)
+       u64 num_reads;          /* failed + successful */
+       u64 num_writes;         /* --do-- */
+       u64 failed_reads;       /* can happen when memory is too low */
+       u64 failed_writes;      /* should NEVER! happen */
+       u64 invalid_io;         /* non-swap I/O requests */
+       u32 pages_zero;         /* no. of zero filled pages */
+       u32 pages_stored;       /* no. of pages currently stored */
+       u32 good_compress;      /* % of pages with compression ratio<=50% */
+       u32 pages_expand;       /* % of incompressible pages */
+       u64 bdev_num_reads;     /* no. of reads on backing dev */
+       u64 bdev_num_writes;    /* no. of writes on backing dev */
+#endif
+};
+
+struct ramzswap {
+       struct xv_pool *mem_pool;
+       void *compress_workmem;
+       void *compress_buffer;
+       struct table *table;
+       struct mutex lock;
+       struct request_queue *queue;
+       struct gendisk *disk;
+       int init_done;
+       /*
+        * This is limit on compressed data size (stats.compr_size)
+        * Its applicable only when backing swap device is present.
+        */
+       size_t memlimit;        /* bytes */
+       /*
+        * This is limit on amount of *uncompressed* worth of data
+        * we can hold. When backing swap device is provided, it is
+        * set equal to device size.
+        */
+       size_t disksize;        /* bytes */
+
+       struct ramzswap_stats stats;
+
+       /* backing swap device info */
+       struct ramzswap_backing_extent *curr_extent;
+       struct list_head backing_swap_extent_list;
+       unsigned long num_extents;
+       char backing_swap_name[MAX_SWAP_NAME_LEN];
+       struct block_device *backing_swap;
+       struct file *swap_file;
+};
+
+/*-- */
+
+#endif
+
diff --git a/drivers/staging/ramzswap/ramzswap_ioctl.h b/drivers/staging/ramzswap/ramzswap_ioctl.h
new file mode 100644 (file)
index 0000000..c713a09
--- /dev/null
@@ -0,0 +1,49 @@
+/*
+ * Compressed RAM based swap device
+ *
+ * Copyright (C) 2008, 2009  Nitin Gupta
+ *
+ * This code is released using a dual license strategy: BSD/GPL
+ * You can choose the licence that better fits your requirements.
+ *
+ * Released under the terms of 3-clause BSD License
+ * Released under the terms of GNU General Public License Version 2.0
+ *
+ * Project home: http://compcache.googlecode.com
+ */
+
+#ifndef _RAMZSWAP_IOCTL_H_
+#define _RAMZSWAP_IOCTL_H_
+
+#define MAX_SWAP_NAME_LEN 128
+
+struct ramzswap_ioctl_stats {
+       char backing_swap_name[MAX_SWAP_NAME_LEN];
+       u64 memlimit;           /* only applicable if backing swap present */
+       u64 disksize;           /* user specified or equal to backing swap
+                                * size (if present) */
+       u64 num_reads;          /* failed + successful */
+       u64 num_writes;         /* --do-- */
+       u64 failed_reads;       /* can happen when memory is too low */
+       u64 failed_writes;      /* should NEVER! happen */
+       u64 invalid_io;         /* non-swap I/O requests */
+       u32 pages_zero;         /* no. of zero filled pages */
+       u32 good_compress_pct;  /* no. of pages with compression ratio<=50% */
+       u32 pages_expand_pct;   /* no. of incompressible pages */
+       u32 pages_stored;
+       u32 pages_used;
+       u64 orig_data_size;
+       u64 compr_data_size;
+       u64 mem_used_total;
+       u64 bdev_num_reads;     /* no. of reads on backing dev */
+       u64 bdev_num_writes;    /* no. of writes on backing dev */
+} __attribute__ ((packed, aligned(4)));
+
+#define RZSIO_SET_DISKSIZE_KB  _IOW('z', 0, size_t)
+#define RZSIO_SET_MEMLIMIT_KB  _IOW('z', 1, size_t)
+#define RZSIO_SET_BACKING_SWAP _IOW('z', 2, unsigned char[MAX_SWAP_NAME_LEN])
+#define RZSIO_GET_STATS                _IOR('z', 3, struct ramzswap_ioctl_stats)
+#define RZSIO_INIT             _IO('z', 4)
+#define RZSIO_RESET            _IO('z', 5)
+
+#endif