Editing PFS

'''PFS''' (Pseudo/Playstation File System) is the file system used by (at least) downloadable content and games on the PS4. It is loosely based on the [[wikipedia:UFS|UFS]] (Unix File System) used in FreeBSD. PFS typically uses a 64kB block size, although the block size is configurable and specified in the file system header. The minimum block size is 4kB, and the maximum is 32MiB; the block size must be a power of 2.

= Structure =

There are four main sections in PFS:
* Header (superblock)
* Inode blocks
* Directory blocks
* Data blocks

== Header/Superblock ==

{| class="wikitable"
! Offset !! Value !! Size !! Notes
|-
| 0x00 || version || 0x8 || Always 1
|-
| 0x08 || format || 0x8 || Always 20130315
|-
| 0x10 || id || 0x8 || 
|-
| 0x18 || fmode || 0x1 || 
|-
| 0x19 || clean || 0x1 || 
|-
| 0x1A || ronly || 0x1 || If 1, this is a readonly filesystem
|-
| 0x1B || rsv || 0x1 || 
|-
| 0x1C || mode || 0x2 || Bit 0 = signed/unsigned, Bit 1 = 32/64 bit inodes
|-
| 0x1E || unknown || 0x2 || 
|-
| 0x20 || blocksz || 0x4 || The size of each block in the filesystem
|-
| 0x24 || nbackup || 0x4 || Seems to always be 0
|-
| 0x28 || nblock || 0x8 || Seems to always be 1
|-
| 0x30 || ndinode || 0x8 || Number of inodes in the inode blocks
|-
| 0x38 || ndblock || 0x8 || Number of data blocks
|-
| 0x40 || ndinodeblock || 0x8 || Number of inode blocks
|-
| 0x48 || superroot_ino || 0x8 || 
|}

<source lang="c">
typedef struct {
    int64 version;
    int64 magic;
    int32 id[2];
    char fmode;
    char clean;
    char ronly;
    char rsv;
    int16 mode;
    int16 unk1;
    int32 blocksz;
    int32 nbackup;
    int64 nblock;
    int64 ndinode;
    int64 ndblock;
    int64 ndinodeblock;
    int64 superroot_ino;
} PFS_HDR;
</source>

== Inodes ==

Inode table starts at the second block and continues for <tt>header.ndinodeblock</tt> blocks.
There are only <tt>header.blocksz / sizeof(di_d32)</tt> inodes per block. (inodes will never cross a block boundary)

For an explanation of inodes, direct and indirect blocks, see [[wikipedia:Inode_pointer_structure|inode pointer structure]] on Wikipedia. Most PFS images from PKGs have just one direct block pointer per file, and the file's data just takes up consecutive blocks. But if there is fragmentation on the file system, you will have to follow the direct and indirect block pointers.

There are actually 4 types of inodes that can be used in a PFS image, and the type is determined by the low two bits of the<tt>mode</tt> field in the superblock. Most PFS images that you see, such as pfs_image.dat, will be using mode 0, which refers to non-signed 32-bit inodes.

{| class="wikitable"
! Offset !! Value !! Size !! Notes
|-
| 0x00 || mode|| 0x2 || Inode mode (bitwise OR of flags; file=0x8000, directory=0x4000), low 9 bits are file permissions
|-
| 0x02 || nlink || 0x2 || Number of links to this inode. For files, this is 1. For dirs, this is 1 plus the number of subdirectories (the <tt>..</tt> entry links to the subdir's parent, increasing its <tt>nlink</tt> count).
|-
| 0x04 || flags|| 0x4 || Bitfield of flags for compressed, readonly, etc.
|-
| 0x08 || size || 0x8 || Size in bytes of the entity
|-
| 0x10 || size_compressed || 0x8 || same as <tt>size</tt> for uncompressed PFS
|-
| 0x18 || times|| 0x30 || Four 64-bit unix timestamps, followed by four 32-bit zeroes, which may be nanoseconds.
|-
| 0x48 || uid|| 0x4 || User ID (zero for game PFS images at least)
|-
| 0x4C || gid|| 0x4 || Group ID (zero for game PFS images at least)
|-
| 0x50 || spare || 0x10 || Probably the same as di_spare, always 0
|-
| 0x60 || blocks || 0x4 || Number of blocks occupied
|-
| 0x64 || db || 0x30 || Direct blocks
|-
| 0x94 || ib || 0x14 || Indirect blocks
|}

<source lang="c">
typedef struct {
    // bitfields are from LSB to MSB
    struct {
        uint16 o_read : 1;
        uint16 o_write : 1;
        uint16 o_execute : 1;
        uint16 g_read : 1;
        uint16 g_write : 1;
        uint16 g_execute : 1;
        uint16 u_read : 1;
        uint16 u_write : 1;
        uint16 u_execute : 1;
        uint16 unk : 5;
        uint16 dir : 1;
        uint16 file : 1;
    } mode;
    uint16 nlink;
    struct {
        uint compressed : 1;
        uint unk : 3;
        uint readonly : 1;
        uint unk2 : 12;
        uint internal : 1;
    } flags;
    uint64 size;
    uint64 size_compressed;
    struct { 
        uint64 unix_time[4];
        uint32 time_nsec[4];
    } times;
    uint32 uid;
    uint32 gid;
    uint64 spare[2];
    uint32 blocks;
    int32 db[12];
    int32 ib[5];
} di_d32;
</source>

== Dirents ==

Each inode with a the <tt>mode.dir</tt> bit set (<tt>mode |= 0x4000</tt>) points to block(s) of dirents. Dirents contain the name and type of files, directories, symlinks, etc. Each directory will have an associated dirent block containing at least the '.' and '..' special files, along with all other files and sub-directories in that directory. Dirents are 8-byte aligned. The <tt>entsize</tt> value will say the total length of this dirent. There is typically padding after <tt>name</tt> which can just be skipped.

If the type of the dirent is 3, it is a directory. Its <tt>ino</tt> value indicates the inode number (0 being the first inode). That inode will point to the dirent block for that directory so you can continue down the file system tree.

If the type of the dirent is 2, it is a file. Its inode will point you to the location(s) of the file data.
{| class="wikitable"
! Offset !! Value !! Size !! Notes
|-
| 0x00 || ino || 0x4 || Inode index
|-
| 0x04 || type|| 0x4 || Type of entry. 2=file, 3=directory, 4= . (link to current dir) , 5= .. (link to parent)
|-
| 0x08 || namelen|| 0x4 || Length of filename (add 1 for 0-terminator)
|-
| 0x0C || entsize || 0x4 || Size of this whole struct, in bytes
|-
| 0x10 || name || namelen + 1 || Filename and 0-terminator
|-
| 0x11 + namelen || padding || variable || Padding so this structure is exactly <tt>entsize</tt> bytes.
|}

<source lang="c">
typedef struct {
    int32 ino;
    int32 type;
    int32 namelen;
    int32 entsize;
    char name[namelen+1];
} dirent;
</source>

== Finding the root ==

The filesystem tree starts with the superroot, a sort of meta-directory that contains the root directory within it, along with something called a "flat_path_table". The superroot's inode is typically the first (zeroeth) inode entry in the table, but you should check <tt>PFS_HDR.superroot_ino</tt> for the actual index. The true root of the filesystem is the "uroot" directory within the super root.

== flat_path_table ==

The flat_path_table is a file located above the root directory on every PFS image. It is simply a mapping of filename hashes to inode number, to increase the lookup speed for files:

<source lang="c">
typedef struct {
    uint32 filename_hash;
    uint32 inode;
} path_table_entry;
</source>

The hashes are sorted in ascending order in the file. The hash function is given below:

<source lang="c">
uint32_t fptbl_hash(const char* filename){
  int i;
  uint32_t hash = 0;
  for(i = 0; filename[i]; i++) {
    // convert character to uppercase
    char c = (filename[i] >= 'a' && filename[i] <= 'z') ? (filename[i] - 32) : filename[i];
    hash = c + 31 * hash;
  }
  return hash;
}
</source>

[https://www.google.com/patents/EP2878348A1?cl=en Patent explaining the flat path table]

= Tools =
* [https://github.com/maxton/GameArchives/releases/latest GameArchives/ArchiveExplorer by maxton] C# library/tool that supports opening and extracting from PFS images
* [https://github.com/maxton/MakePFS MakePFS by maxton] C# tool for creating PFS images