1Notes on Filesystem Layout 2-------------------------- 3 4These notes describe what mkcramfs generates. Kernel requirements are 5a bit looser, e.g. it doesn't care if the <file_data> items are 6swapped around (though it does care that directory entries (inodes) in 7a given directory are contiguous, as this is used by readdir). 8 9All data is currently in host-endian format; neither mkcramfs nor the 10kernel ever do swabbing. (See section `Block Size' below.) 11 12<filesystem>: 13 <superblock> 14 <directory_structure> 15 <data> 16 17<superblock>: struct cramfs_super (see cramfs_fs.h). 18 19<directory_structure>: 20 For each file: 21 struct cramfs_inode (see cramfs_fs.h). 22 Filename. Not generally null-terminated, but it is 23 null-padded to a multiple of 4 bytes. 24 25The order of inode traversal is described as "width-first" (not to be 26confused with breadth-first); i.e. like depth-first but listing all of 27a directory's entries before recursing down its subdirectories: the 28same order as `ls -AUR' (but without the /^\..*:$/ directory header 29lines); put another way, the same order as `find -type d -exec 30ls -AU1 {} \;'. 31 32Beginning in 2.4.7, directory entries are sorted. This optimization 33allows cramfs_lookup to return more quickly when a filename does not 34exist, speeds up user-space directory sorts, etc. 35 36<data>: 37 One <file_data> for each file that's either a symlink or a 38 regular file of non-zero st_size. 39 40<file_data>: 41 nblocks * <block_pointer> 42 (where nblocks = (st_size - 1) / blksize + 1) 43 nblocks * <block> 44 padding to multiple of 4 bytes 45 46The i'th <block_pointer> for a file stores the byte offset of the 47*end* of the i'th <block> (i.e. one past the last byte, which is the 48same as the start of the (i+1)'th <block> if there is one). The first 49<block> immediately follows the last <block_pointer> for the file. 50<block_pointer>s are each 32 bits long. 51 52When the CRAMFS_FLAG_EXT_BLOCK_POINTERS capability bit is set, each 53<block_pointer>'s top bits may contain special flags as follows: 54 55CRAMFS_BLK_FLAG_UNCOMPRESSED (bit 31): 56 The block data is not compressed and should be copied verbatim. 57 58CRAMFS_BLK_FLAG_DIRECT_PTR (bit 30): 59 The <block_pointer> stores the actual block start offset and not 60 its end, shifted right by 2 bits. The block must therefore be 61 aligned to a 4-byte boundary. The block size is either blksize 62 if CRAMFS_BLK_FLAG_UNCOMPRESSED is also specified, otherwise 63 the compressed data length is included in the first 2 bytes of 64 the block data. This is used to allow discontiguous data layout 65 and specific data block alignments e.g. for XIP applications. 66 67 68The order of <file_data>'s is a depth-first descent of the directory 69tree, i.e. the same order as `find -size +0 \( -type f -o -type l \) 70-print'. 71 72 73<block>: The i'th <block> is the output of zlib's compress function 74applied to the i'th blksize-sized chunk of the input data if the 75corresponding CRAMFS_BLK_FLAG_UNCOMPRESSED <block_ptr> bit is not set, 76otherwise it is the input data directly. 77(For the last <block> of the file, the input may of course be smaller.) 78Each <block> may be a different size. (See <block_pointer> above.) 79 80<block>s are merely byte-aligned, not generally u32-aligned. 81 82When CRAMFS_BLK_FLAG_DIRECT_PTR is specified then the corresponding 83<block> may be located anywhere and not necessarily contiguous with 84the previous/next blocks. In that case it is minimally u32-aligned. 85If CRAMFS_BLK_FLAG_UNCOMPRESSED is also specified then the size is always 86blksize except for the last block which is limited by the file length. 87If CRAMFS_BLK_FLAG_DIRECT_PTR is set and CRAMFS_BLK_FLAG_UNCOMPRESSED 88is not set then the first 2 bytes of the block contains the size of the 89remaining block data as this cannot be determined from the placement of 90logically adjacent blocks. 91 92 93Holes 94----- 95 96This kernel supports cramfs holes (i.e. [efficient representation of] 97blocks in uncompressed data consisting entirely of NUL bytes), but by 98default mkcramfs doesn't test for & create holes, since cramfs in 99kernels up to at least 2.3.39 didn't support holes. Run mkcramfs 100with -z if you want it to create files that can have holes in them. 101 102 103Tools 104----- 105 106The cramfs user-space tools, including mkcramfs and cramfsck, are 107located at <http://sourceforge.net/projects/cramfs/>. 108 109 110Future Development 111================== 112 113Block Size 114---------- 115 116(Block size in cramfs refers to the size of input data that is 117compressed at a time. It's intended to be somewhere around 118PAGE_SIZE for cramfs_read_folio's convenience.) 119 120The superblock ought to indicate the block size that the fs was 121written for, since comments in <linux/pagemap.h> indicate that 122PAGE_SIZE may grow in future (if I interpret the comment 123correctly). 124 125Currently, mkcramfs #define's PAGE_SIZE as 4096 and uses that 126for blksize, whereas Linux-2.3.39 uses its PAGE_SIZE, which in 127turn is defined as PAGE_SIZE (which can be as large as 32KB on arm). 128This discrepancy is a bug, though it's not clear which should be 129changed. 130 131One option is to change mkcramfs to take its PAGE_SIZE from 132<asm/page.h>. Personally I don't like this option, but it does 133require the least amount of change: just change `#define 134PAGE_SIZE (4096)' to `#include <asm/page.h>'. The disadvantage 135is that the generated cramfs cannot always be shared between different 136kernels, not even necessarily kernels of the same architecture if 137PAGE_SIZE is subject to change between kernel versions 138(currently possible with arm and ia64). 139 140The remaining options try to make cramfs more sharable. 141 142One part of that is addressing endianness. The two options here are 143`always use little-endian' (like ext2fs) or `writer chooses 144endianness; kernel adapts at runtime'. Little-endian wins because of 145code simplicity and little CPU overhead even on big-endian machines. 146 147The cost of swabbing is changing the code to use the le32_to_cpu 148etc. macros as used by ext2fs. We don't need to swab the compressed 149data, only the superblock, inodes and block pointers. 150 151 152The other part of making cramfs more sharable is choosing a block 153size. The options are: 154 155 1. Always 4096 bytes. 156 157 2. Writer chooses blocksize; kernel adapts but rejects blocksize > 158 PAGE_SIZE. 159 160 3. Writer chooses blocksize; kernel adapts even to blocksize > 161 PAGE_SIZE. 162 163It's easy enough to change the kernel to use a smaller value than 164PAGE_SIZE: just make cramfs_read_folio read multiple blocks. 165 166The cost of option 1 is that kernels with a larger PAGE_SIZE 167value don't get as good compression as they can. 168 169The cost of option 2 relative to option 1 is that the code uses 170variables instead of #define'd constants. The gain is that people 171with kernels having larger PAGE_SIZE can make use of that if 172they don't mind their cramfs being inaccessible to kernels with 173smaller PAGE_SIZE values. 174 175Option 3 is easy to implement if we don't mind being CPU-inefficient: 176e.g. get read_folio to decompress to a buffer of size MAX_BLKSIZE (which 177must be no larger than 32KB) and discard what it doesn't need. 178Getting read_folio to read into all the covered pages is harder. 179 180The main advantage of option 3 over 1, 2, is better compression. The 181cost is greater complexity. Probably not worth it, but I hope someone 182will disagree. (If it is implemented, then I'll re-use that code in 183e2compr.) 184 185 186Another cost of 2 and 3 over 1 is making mkcramfs use a different 187block size, but that just means adding and parsing a -b option. 188 189 190Inode Size 191---------- 192 193Given that cramfs will probably be used for CDs etc. as well as just 194silicon ROMs, it might make sense to expand the inode a little from 195its current 12 bytes. Inodes other than the root inode are followed 196by filename, so the expansion doesn't even have to be a multiple of 4 197bytes. 198