]> git.karo-electronics.de Git - mv-sheeva.git/blobdiff - fs/btrfs/zlib.c
Merge tag 'v2.6.38' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[mv-sheeva.git] / fs / btrfs / zlib.c
index b9cd5445f71cac0c6b2d565f80d065091d65fa4b..f5ec2d44150df2a484d6f6b88f856ba9ba8b6464 100644 (file)
 #include <linux/bio.h>
 #include "compression.h"
 
-/* Plan: call deflate() with avail_in == *sourcelen,
-       avail_out = *dstlen - 12 and flush == Z_FINISH.
-       If it doesn't manage to finish, call it again with
-       avail_in == 0 and avail_out set to the remaining 12
-       bytes for it to clean up.
-   Q: Is 12 bytes sufficient?
-*/
-#define STREAM_END_SPACE 12
-
 struct workspace {
        z_stream inf_strm;
        z_stream def_strm;
@@ -48,152 +39,51 @@ struct workspace {
        struct list_head list;
 };
 
-static LIST_HEAD(idle_workspace);
-static DEFINE_SPINLOCK(workspace_lock);
-static unsigned long num_workspace;
-static atomic_t alloc_workspace = ATOMIC_INIT(0);
-static DECLARE_WAIT_QUEUE_HEAD(workspace_wait);
+static void zlib_free_workspace(struct list_head *ws)
+{
+       struct workspace *workspace = list_entry(ws, struct workspace, list);
 
-/*
- * this finds an available zlib workspace or allocates a new one
- * NULL or an ERR_PTR is returned if things go bad.
- */
-static struct workspace *find_zlib_workspace(void)
+       vfree(workspace->def_strm.workspace);
+       vfree(workspace->inf_strm.workspace);
+       kfree(workspace->buf);
+       kfree(workspace);
+}
+
+static struct list_head *zlib_alloc_workspace(void)
 {
        struct workspace *workspace;
-       int ret;
-       int cpus = num_online_cpus();
-
-again:
-       spin_lock(&workspace_lock);
-       if (!list_empty(&idle_workspace)) {
-               workspace = list_entry(idle_workspace.next, struct workspace,
-                                      list);
-               list_del(&workspace->list);
-               num_workspace--;
-               spin_unlock(&workspace_lock);
-               return workspace;
 
-       }
-       spin_unlock(&workspace_lock);
-       if (atomic_read(&alloc_workspace) > cpus) {
-               DEFINE_WAIT(wait);
-               prepare_to_wait(&workspace_wait, &wait, TASK_UNINTERRUPTIBLE);
-               if (atomic_read(&alloc_workspace) > cpus)
-                       schedule();
-               finish_wait(&workspace_wait, &wait);
-               goto again;
-       }
-       atomic_inc(&alloc_workspace);
        workspace = kzalloc(sizeof(*workspace), GFP_NOFS);
-       if (!workspace) {
-               ret = -ENOMEM;
-               goto fail;
-       }
+       if (!workspace)
+               return ERR_PTR(-ENOMEM);
 
        workspace->def_strm.workspace = vmalloc(zlib_deflate_workspacesize());
-       if (!workspace->def_strm.workspace) {
-               ret = -ENOMEM;
-               goto fail;
-       }
        workspace->inf_strm.workspace = vmalloc(zlib_inflate_workspacesize());
-       if (!workspace->inf_strm.workspace) {
-               ret = -ENOMEM;
-               goto fail_inflate;
-       }
        workspace->buf = kmalloc(PAGE_CACHE_SIZE, GFP_NOFS);
-       if (!workspace->buf) {
-               ret = -ENOMEM;
-               goto fail_kmalloc;
-       }
-       return workspace;
-
-fail_kmalloc:
-       vfree(workspace->inf_strm.workspace);
-fail_inflate:
-       vfree(workspace->def_strm.workspace);
-fail:
-       kfree(workspace);
-       atomic_dec(&alloc_workspace);
-       wake_up(&workspace_wait);
-       return ERR_PTR(ret);
-}
-
-/*
- * put a workspace struct back on the list or free it if we have enough
- * idle ones sitting around
- */
-static int free_workspace(struct workspace *workspace)
-{
-       spin_lock(&workspace_lock);
-       if (num_workspace < num_online_cpus()) {
-               list_add_tail(&workspace->list, &idle_workspace);
-               num_workspace++;
-               spin_unlock(&workspace_lock);
-               if (waitqueue_active(&workspace_wait))
-                       wake_up(&workspace_wait);
-               return 0;
-       }
-       spin_unlock(&workspace_lock);
-       vfree(workspace->def_strm.workspace);
-       vfree(workspace->inf_strm.workspace);
-       kfree(workspace->buf);
-       kfree(workspace);
+       if (!workspace->def_strm.workspace ||
+           !workspace->inf_strm.workspace || !workspace->buf)
+               goto fail;
 
-       atomic_dec(&alloc_workspace);
-       if (waitqueue_active(&workspace_wait))
-               wake_up(&workspace_wait);
-       return 0;
-}
+       INIT_LIST_HEAD(&workspace->list);
 
-/*
- * cleanup function for module exit
- */
-static void free_workspaces(void)
-{
-       struct workspace *workspace;
-       while (!list_empty(&idle_workspace)) {
-               workspace = list_entry(idle_workspace.next, struct workspace,
-                                      list);
-               list_del(&workspace->list);
-               vfree(workspace->def_strm.workspace);
-               vfree(workspace->inf_strm.workspace);
-               kfree(workspace->buf);
-               kfree(workspace);
-               atomic_dec(&alloc_workspace);
-       }
+       return &workspace->list;
+fail:
+       zlib_free_workspace(&workspace->list);
+       return ERR_PTR(-ENOMEM);
 }
 
-/*
- * given an address space and start/len, compress the bytes.
- *
- * pages are allocated to hold the compressed result and stored
- * in 'pages'
- *
- * out_pages is used to return the number of pages allocated.  There
- * may be pages allocated even if we return an error
- *
- * total_in is used to return the number of bytes actually read.  It
- * may be smaller then len if we had to exit early because we
- * ran out of room in the pages array or because we cross the
- * max_out threshold.
- *
- * total_out is used to return the total number of compressed bytes
- *
- * max_out tells us the max number of bytes that we're allowed to
- * stuff into pages
- */
-int btrfs_zlib_compress_pages(struct address_space *mapping,
-                             u64 start, unsigned long len,
-                             struct page **pages,
-                             unsigned long nr_dest_pages,
-                             unsigned long *out_pages,
-                             unsigned long *total_in,
-                             unsigned long *total_out,
-                             unsigned long max_out)
+static int zlib_compress_pages(struct list_head *ws,
+                              struct address_space *mapping,
+                              u64 start, unsigned long len,
+                              struct page **pages,
+                              unsigned long nr_dest_pages,
+                              unsigned long *out_pages,
+                              unsigned long *total_in,
+                              unsigned long *total_out,
+                              unsigned long max_out)
 {
+       struct workspace *workspace = list_entry(ws, struct workspace, list);
        int ret;
-       struct workspace *workspace;
        char *data_in;
        char *cpage_out;
        int nr_pages = 0;
@@ -205,10 +95,6 @@ int btrfs_zlib_compress_pages(struct address_space *mapping,
        *total_out = 0;
        *total_in = 0;
 
-       workspace = find_zlib_workspace();
-       if (IS_ERR(workspace))
-               return -1;
-
        if (Z_OK != zlib_deflateInit(&workspace->def_strm, 3)) {
                printk(KERN_WARNING "deflateInit failed\n");
                ret = -1;
@@ -222,6 +108,10 @@ int btrfs_zlib_compress_pages(struct address_space *mapping,
        data_in = kmap(in_page);
 
        out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
+       if (out_page == NULL) {
+               ret = -1;
+               goto out;
+       }
        cpage_out = kmap(out_page);
        pages[0] = out_page;
        nr_pages = 1;
@@ -260,6 +150,10 @@ int btrfs_zlib_compress_pages(struct address_space *mapping,
                                goto out;
                        }
                        out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
+                       if (out_page == NULL) {
+                               ret = -1;
+                               goto out;
+                       }
                        cpage_out = kmap(out_page);
                        pages[nr_pages] = out_page;
                        nr_pages++;
@@ -314,55 +208,26 @@ out:
                kunmap(in_page);
                page_cache_release(in_page);
        }
-       free_workspace(workspace);
        return ret;
 }
 
-/*
- * pages_in is an array of pages with compressed data.
- *
- * disk_start is the starting logical offset of this array in the file
- *
- * bvec is a bio_vec of pages from the file that we want to decompress into
- *
- * vcnt is the count of pages in the biovec
- *
- * srclen is the number of bytes in pages_in
- *
- * The basic idea is that we have a bio that was created by readpages.
- * The pages in the bio are for the uncompressed data, and they may not
- * be contiguous.  They all correspond to the range of bytes covered by
- * the compressed extent.
- */
-int btrfs_zlib_decompress_biovec(struct page **pages_in,
-                             u64 disk_start,
-                             struct bio_vec *bvec,
-                             int vcnt,
-                             size_t srclen)
+static int zlib_decompress_biovec(struct list_head *ws, struct page **pages_in,
+                                 u64 disk_start,
+                                 struct bio_vec *bvec,
+                                 int vcnt,
+                                 size_t srclen)
 {
-       int ret = 0;
+       struct workspace *workspace = list_entry(ws, struct workspace, list);
+       int ret = 0, ret2;
        int wbits = MAX_WBITS;
-       struct workspace *workspace;
        char *data_in;
        size_t total_out = 0;
-       unsigned long page_bytes_left;
        unsigned long page_in_index = 0;
        unsigned long page_out_index = 0;
-       struct page *page_out;
        unsigned long total_pages_in = (srclen + PAGE_CACHE_SIZE - 1) /
                                        PAGE_CACHE_SIZE;
        unsigned long buf_start;
-       unsigned long buf_offset;
-       unsigned long bytes;
-       unsigned long working_bytes;
        unsigned long pg_offset;
-       unsigned long start_byte;
-       unsigned long current_buf_start;
-       char *kaddr;
-
-       workspace = find_zlib_workspace();
-       if (IS_ERR(workspace))
-               return -ENOMEM;
 
        data_in = kmap(pages_in[page_in_index]);
        workspace->inf_strm.next_in = data_in;
@@ -372,8 +237,6 @@ int btrfs_zlib_decompress_biovec(struct page **pages_in,
        workspace->inf_strm.total_out = 0;
        workspace->inf_strm.next_out = workspace->buf;
        workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
-       page_out = bvec[page_out_index].bv_page;
-       page_bytes_left = PAGE_CACHE_SIZE;
        pg_offset = 0;
 
        /* If it's deflate, and it's got no preset dictionary, then
@@ -389,107 +252,29 @@ int btrfs_zlib_decompress_biovec(struct page **pages_in,
 
        if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) {
                printk(KERN_WARNING "inflateInit failed\n");
-               ret = -1;
-               goto out;
+               return -1;
        }
        while (workspace->inf_strm.total_in < srclen) {
                ret = zlib_inflate(&workspace->inf_strm, Z_NO_FLUSH);
                if (ret != Z_OK && ret != Z_STREAM_END)
                        break;
-               /*
-                * buf start is the byte offset we're of the start of
-                * our workspace buffer
-                */
-               buf_start = total_out;
 
-               /* total_out is the last byte of the workspace buffer */
+               buf_start = total_out;
                total_out = workspace->inf_strm.total_out;
 
-               working_bytes = total_out - buf_start;
-
-               /*
-                * start byte is the first byte of the page we're currently
-                * copying into relative to the start of the compressed data.
-                */
-               start_byte = page_offset(page_out) - disk_start;
-
-               if (working_bytes == 0) {
-                       /* we didn't make progress in this inflate
-                        * call, we're done
-                        */
-                       if (ret != Z_STREAM_END)
-                               ret = -1;
+               /* we didn't make progress in this inflate call, we're done */
+               if (buf_start == total_out)
                        break;
-               }
 
-               /* we haven't yet hit data corresponding to this page */
-               if (total_out <= start_byte)
-                       goto next;
-
-               /*
-                * the start of the data we care about is offset into
-                * the middle of our working buffer
-                */
-               if (total_out > start_byte && buf_start < start_byte) {
-                       buf_offset = start_byte - buf_start;
-                       working_bytes -= buf_offset;
-               } else {
-                       buf_offset = 0;
-               }
-               current_buf_start = buf_start;
-
-               /* copy bytes from the working buffer into the pages */
-               while (working_bytes > 0) {
-                       bytes = min(PAGE_CACHE_SIZE - pg_offset,
-                                   PAGE_CACHE_SIZE - buf_offset);
-                       bytes = min(bytes, working_bytes);
-                       kaddr = kmap_atomic(page_out, KM_USER0);
-                       memcpy(kaddr + pg_offset, workspace->buf + buf_offset,
-                              bytes);
-                       kunmap_atomic(kaddr, KM_USER0);
-                       flush_dcache_page(page_out);
-
-                       pg_offset += bytes;
-                       page_bytes_left -= bytes;
-                       buf_offset += bytes;
-                       working_bytes -= bytes;
-                       current_buf_start += bytes;
-
-                       /* check if we need to pick another page */
-                       if (page_bytes_left == 0) {
-                               page_out_index++;
-                               if (page_out_index >= vcnt) {
-                                       ret = 0;
-                                       goto done;
-                               }
-
-                               page_out = bvec[page_out_index].bv_page;
-                               pg_offset = 0;
-                               page_bytes_left = PAGE_CACHE_SIZE;
-                               start_byte = page_offset(page_out) - disk_start;
-
-                               /*
-                                * make sure our new page is covered by this
-                                * working buffer
-                                */
-                               if (total_out <= start_byte)
-                                       goto next;
-
-                               /* the next page in the biovec might not
-                                * be adjacent to the last page, but it
-                                * might still be found inside this working
-                                * buffer.  bump our offset pointer
-                                */
-                               if (total_out > start_byte &&
-                                   current_buf_start < start_byte) {
-                                       buf_offset = start_byte - buf_start;
-                                       working_bytes = total_out - start_byte;
-                                       current_buf_start = buf_start +
-                                               buf_offset;
-                               }
-                       }
+               ret2 = btrfs_decompress_buf2page(workspace->buf, buf_start,
+                                                total_out, disk_start,
+                                                bvec, vcnt,
+                                                &page_out_index, &pg_offset);
+               if (ret2 == 0) {
+                       ret = 0;
+                       goto done;
                }
-next:
+
                workspace->inf_strm.next_out = workspace->buf;
                workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
 
@@ -516,35 +301,21 @@ done:
        zlib_inflateEnd(&workspace->inf_strm);
        if (data_in)
                kunmap(pages_in[page_in_index]);
-out:
-       free_workspace(workspace);
        return ret;
 }
 
-/*
- * a less complex decompression routine.  Our compressed data fits in a
- * single page, and we want to read a single page out of it.
- * start_byte tells us the offset into the compressed data we're interested in
- */
-int btrfs_zlib_decompress(unsigned char *data_in,
-                         struct page *dest_page,
-                         unsigned long start_byte,
-                         size_t srclen, size_t destlen)
+static int zlib_decompress(struct list_head *ws, unsigned char *data_in,
+                          struct page *dest_page,
+                          unsigned long start_byte,
+                          size_t srclen, size_t destlen)
 {
+       struct workspace *workspace = list_entry(ws, struct workspace, list);
        int ret = 0;
        int wbits = MAX_WBITS;
-       struct workspace *workspace;
        unsigned long bytes_left = destlen;
        unsigned long total_out = 0;
        char *kaddr;
 
-       if (destlen > PAGE_CACHE_SIZE)
-               return -ENOMEM;
-
-       workspace = find_zlib_workspace();
-       if (IS_ERR(workspace))
-               return -ENOMEM;
-
        workspace->inf_strm.next_in = data_in;
        workspace->inf_strm.avail_in = srclen;
        workspace->inf_strm.total_in = 0;
@@ -565,8 +336,7 @@ int btrfs_zlib_decompress(unsigned char *data_in,
 
        if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) {
                printk(KERN_WARNING "inflateInit failed\n");
-               ret = -1;
-               goto out;
+               return -1;
        }
 
        while (bytes_left > 0) {
@@ -616,12 +386,13 @@ next:
                ret = 0;
 
        zlib_inflateEnd(&workspace->inf_strm);
-out:
-       free_workspace(workspace);
        return ret;
 }
 
-void btrfs_zlib_exit(void)
-{
-    free_workspaces();
-}
+struct btrfs_compress_op btrfs_zlib_compress = {
+       .alloc_workspace        = zlib_alloc_workspace,
+       .free_workspace         = zlib_free_workspace,
+       .compress_pages         = zlib_compress_pages,
+       .decompress_biovec      = zlib_decompress_biovec,
+       .decompress             = zlib_decompress,
+};