jim owens
2010-Jan-04 21:13 UTC
[RFC 06/12 PATCH] Btrfs: reduce memory use in compress/decompress workspace.
Compression and decompression can use a single z_stream and z_stream.workspace memory as the operations always begin from scratch and do not overlap in one thread''s execution. Signed-off-by: jim owens <jowens@hp.com> --- fs/btrfs/zlib.c | 146 ++++++++++++++++++++++++++----------------------------- 1 files changed, 69 insertions(+), 77 deletions(-) diff --git a/fs/btrfs/zlib.c b/fs/btrfs/zlib.c index 3e2b90e..d7bdce5 100644 --- a/fs/btrfs/zlib.c +++ b/fs/btrfs/zlib.c @@ -42,8 +42,7 @@ #define STREAM_END_SPACE 12 struct workspace { - z_stream inf_strm; - z_stream def_strm; + z_stream z_strm; char *buf; struct list_head list; }; @@ -91,16 +90,13 @@ again: goto fail; } - workspace->def_strm.workspace = vmalloc(zlib_deflate_workspacesize()); - if (!workspace->def_strm.workspace) { + workspace->z_strm.workspace = vmalloc(max(zlib_deflate_workspacesize(), + zlib_inflate_workspacesize())); + if (!workspace->z_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; @@ -109,9 +105,7 @@ again: return workspace; fail_kmalloc: - vfree(workspace->inf_strm.workspace); -fail_inflate: - vfree(workspace->def_strm.workspace); + vfree(workspace->z_strm.workspace); fail: kfree(workspace); atomic_dec(&alloc_workspace); @@ -135,8 +129,7 @@ static int free_workspace(struct workspace *workspace) return 0; } spin_unlock(&workspace_lock); - vfree(workspace->def_strm.workspace); - vfree(workspace->inf_strm.workspace); + vfree(workspace->z_strm.workspace); kfree(workspace->buf); kfree(workspace); @@ -156,8 +149,7 @@ static void free_workspaces(void) workspace = list_entry(idle_workspace.next, struct workspace, list); list_del(&workspace->list); - vfree(workspace->def_strm.workspace); - vfree(workspace->inf_strm.workspace); + vfree(workspace->z_strm.workspace); kfree(workspace->buf); kfree(workspace); atomic_dec(&alloc_workspace); @@ -211,14 +203,14 @@ int btrfs_zlib_compress_pages(struct address_space *mapping, if (IS_ERR(workspace)) return -1; - if (Z_OK != zlib_deflateInit(&workspace->def_strm, 3)) { + if (Z_OK != zlib_deflateInit(&workspace->z_strm, 3)) { printk(KERN_WARNING "deflateInit failed\n"); ret = -1; goto out; } - workspace->def_strm.total_in = 0; - workspace->def_strm.total_out = 0; + workspace->z_strm.total_in = 0; + workspace->z_strm.total_out = 0; in_page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); data_in = kmap(in_page); @@ -228,28 +220,28 @@ int btrfs_zlib_compress_pages(struct address_space *mapping, pages[0] = out_page; nr_pages = 1; - workspace->def_strm.next_in = data_in; - workspace->def_strm.next_out = cpage_out; - workspace->def_strm.avail_out = PAGE_CACHE_SIZE; - workspace->def_strm.avail_in = min(len, PAGE_CACHE_SIZE); + workspace->z_strm.next_in = data_in; + workspace->z_strm.next_out = cpage_out; + workspace->z_strm.avail_out = PAGE_CACHE_SIZE; + workspace->z_strm.avail_in = min(len, PAGE_CACHE_SIZE); out_written = 0; in_read = 0; - while (workspace->def_strm.total_in < len) { - ret = zlib_deflate(&workspace->def_strm, Z_SYNC_FLUSH); + while (workspace->z_strm.total_in < len) { + ret = zlib_deflate(&workspace->z_strm, Z_SYNC_FLUSH); if (ret != Z_OK) { printk(KERN_DEBUG "btrfs deflate in loop returned %d\n", ret); - zlib_deflateEnd(&workspace->def_strm); + zlib_deflateEnd(&workspace->z_strm); ret = -1; goto out; } /* we''re making it bigger, give up */ - if (workspace->def_strm.total_in > 8192 && - workspace->def_strm.total_in < - workspace->def_strm.total_out) { + if (workspace->z_strm.total_in > 8192 && + workspace->z_strm.total_in < + workspace->z_strm.total_out) { ret = -1; goto out; } @@ -257,7 +249,7 @@ int btrfs_zlib_compress_pages(struct address_space *mapping, * before the total_in so we will pull in a new page for * the stream end if required */ - if (workspace->def_strm.avail_out == 0) { + if (workspace->z_strm.avail_out == 0) { kunmap(out_page); if (nr_pages == nr_dest_pages) { out_page = NULL; @@ -268,19 +260,19 @@ int btrfs_zlib_compress_pages(struct address_space *mapping, cpage_out = kmap(out_page); pages[nr_pages] = out_page; nr_pages++; - workspace->def_strm.avail_out = PAGE_CACHE_SIZE; - workspace->def_strm.next_out = cpage_out; + workspace->z_strm.avail_out = PAGE_CACHE_SIZE; + workspace->z_strm.next_out = cpage_out; } /* we''re all done */ - if (workspace->def_strm.total_in >= len) + if (workspace->z_strm.total_in >= len) break; /* we''ve read in a full page, get a new one */ - if (workspace->def_strm.avail_in == 0) { - if (workspace->def_strm.total_out > max_out) + if (workspace->z_strm.avail_in == 0) { + if (workspace->z_strm.total_out > max_out) break; - bytes_left = len - workspace->def_strm.total_in; + bytes_left = len - workspace->z_strm.total_in; kunmap(in_page); page_cache_release(in_page); @@ -288,28 +280,28 @@ int btrfs_zlib_compress_pages(struct address_space *mapping, in_page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT); data_in = kmap(in_page); - workspace->def_strm.avail_in = min(bytes_left, + workspace->z_strm.avail_in = min(bytes_left, PAGE_CACHE_SIZE); - workspace->def_strm.next_in = data_in; + workspace->z_strm.next_in = data_in; } } - workspace->def_strm.avail_in = 0; - ret = zlib_deflate(&workspace->def_strm, Z_FINISH); - zlib_deflateEnd(&workspace->def_strm); + workspace->z_strm.avail_in = 0; + ret = zlib_deflate(&workspace->z_strm, Z_FINISH); + zlib_deflateEnd(&workspace->z_strm); if (ret != Z_STREAM_END) { ret = -1; goto out; } - if (workspace->def_strm.total_out >= workspace->def_strm.total_in) { + if (workspace->z_strm.total_out >= workspace->z_strm.total_in) { ret = -1; goto out; } ret = 0; - *total_out = workspace->def_strm.total_out; - *total_in = workspace->def_strm.total_in; + *total_out = workspace->z_strm.total_out; + *total_in = workspace->z_strm.total_in; out: *out_pages = nr_pages; if (out_page) @@ -370,13 +362,13 @@ int btrfs_zlib_decompress_biovec(struct page **pages_in, return -ENOMEM; data_in = kmap(pages_in[page_in_index]); - workspace->inf_strm.next_in = data_in; - workspace->inf_strm.avail_in = min_t(size_t, srclen, PAGE_CACHE_SIZE); - workspace->inf_strm.total_in = 0; + workspace->z_strm.next_in = data_in; + workspace->z_strm.avail_in = min_t(size_t, srclen, PAGE_CACHE_SIZE); + workspace->z_strm.total_in = 0; - workspace->inf_strm.total_out = 0; - workspace->inf_strm.next_out = workspace->buf; - workspace->inf_strm.avail_out = PAGE_CACHE_SIZE; + workspace->z_strm.total_out = 0; + workspace->z_strm.next_out = workspace->buf; + workspace->z_strm.avail_out = PAGE_CACHE_SIZE; page_out = bvec[page_out_index].bv_page; page_bytes_left = PAGE_CACHE_SIZE; pg_offset = 0; @@ -388,17 +380,17 @@ int btrfs_zlib_decompress_biovec(struct page **pages_in, !(((data_in[0]<<8) + data_in[1]) % 31)) { wbits = -((data_in[0] >> 4) + 8); - workspace->inf_strm.next_in += 2; - workspace->inf_strm.avail_in -= 2; + workspace->z_strm.next_in += 2; + workspace->z_strm.avail_in -= 2; } - if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) { + if (Z_OK != zlib_inflateInit2(&workspace->z_strm, wbits)) { printk(KERN_WARNING "inflateInit failed\n"); ret = -1; goto out; } - while (workspace->inf_strm.total_in < srclen) { - ret = zlib_inflate(&workspace->inf_strm, Z_NO_FLUSH); + while (workspace->z_strm.total_in < srclen) { + ret = zlib_inflate(&workspace->z_strm, Z_NO_FLUSH); if (ret != Z_OK && ret != Z_STREAM_END) break; /* @@ -408,7 +400,7 @@ int btrfs_zlib_decompress_biovec(struct page **pages_in, buf_start = total_out; /* total_out is the last byte of the workspace buffer */ - total_out = workspace->inf_strm.total_out; + total_out = workspace->z_strm.total_out; working_bytes = total_out - buf_start; @@ -495,10 +487,10 @@ int btrfs_zlib_decompress_biovec(struct page **pages_in, } } next: - workspace->inf_strm.next_out = workspace->buf; - workspace->inf_strm.avail_out = PAGE_CACHE_SIZE; + workspace->z_strm.next_out = workspace->buf; + workspace->z_strm.avail_out = PAGE_CACHE_SIZE; - if (workspace->inf_strm.avail_in == 0) { + if (workspace->z_strm.avail_in == 0) { unsigned long tmp; kunmap(pages_in[page_in_index]); page_in_index++; @@ -507,9 +499,9 @@ next: break; } data_in = kmap(pages_in[page_in_index]); - workspace->inf_strm.next_in = data_in; - tmp = srclen - workspace->inf_strm.total_in; - workspace->inf_strm.avail_in = min(tmp, + workspace->z_strm.next_in = data_in; + tmp = srclen - workspace->z_strm.total_in; + workspace->z_strm.avail_in = min(tmp, PAGE_CACHE_SIZE); } } @@ -518,7 +510,7 @@ next: else ret = 0; done: - zlib_inflateEnd(&workspace->inf_strm); + zlib_inflateEnd(&workspace->z_strm); if (data_in) kunmap(pages_in[page_in_index]); out: @@ -550,13 +542,13 @@ int btrfs_zlib_decompress(unsigned char *data_in, 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; + workspace->z_strm.next_in = data_in; + workspace->z_strm.avail_in = srclen; + workspace->z_strm.total_in = 0; - workspace->inf_strm.next_out = workspace->buf; - workspace->inf_strm.avail_out = PAGE_CACHE_SIZE; - workspace->inf_strm.total_out = 0; + workspace->z_strm.next_out = workspace->buf; + workspace->z_strm.avail_out = PAGE_CACHE_SIZE; + workspace->z_strm.total_out = 0; /* If it''s deflate, and it''s got no preset dictionary, then we can tell zlib to skip the adler32 check. */ if (srclen > 2 && !(data_in[1] & PRESET_DICT) && @@ -564,11 +556,11 @@ int btrfs_zlib_decompress(unsigned char *data_in, !(((data_in[0]<<8) + data_in[1]) % 31)) { wbits = -((data_in[0] >> 4) + 8); - workspace->inf_strm.next_in += 2; - workspace->inf_strm.avail_in -= 2; + workspace->z_strm.next_in += 2; + workspace->z_strm.avail_in -= 2; } - if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) { + if (Z_OK != zlib_inflateInit2(&workspace->z_strm, wbits)) { printk(KERN_WARNING "inflateInit failed\n"); ret = -1; goto out; @@ -580,12 +572,12 @@ int btrfs_zlib_decompress(unsigned char *data_in, unsigned long bytes; unsigned long pg_offset = 0; - ret = zlib_inflate(&workspace->inf_strm, Z_NO_FLUSH); + ret = zlib_inflate(&workspace->z_strm, Z_NO_FLUSH); if (ret != Z_OK && ret != Z_STREAM_END) break; buf_start = total_out; - total_out = workspace->inf_strm.total_out; + total_out = workspace->z_strm.total_out; if (total_out == buf_start) { ret = -1; @@ -611,8 +603,8 @@ int btrfs_zlib_decompress(unsigned char *data_in, pg_offset += bytes; bytes_left -= bytes; next: - workspace->inf_strm.next_out = workspace->buf; - workspace->inf_strm.avail_out = PAGE_CACHE_SIZE; + workspace->z_strm.next_out = workspace->buf; + workspace->z_strm.avail_out = PAGE_CACHE_SIZE; } if (ret != Z_STREAM_END && bytes_left != 0) @@ -620,7 +612,7 @@ next: else ret = 0; - zlib_inflateEnd(&workspace->inf_strm); + zlib_inflateEnd(&workspace->z_strm); out: free_workspace(workspace); return ret; -- 1.5.6.3 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html