mirror of
https://github.com/moparisthebest/SickRage
synced 2024-11-16 22:35:03 -05:00
0d9fbc1ad7
This version of SickBeard uses both TVDB and TVRage to search and gather it's series data from allowing you to now have access to and download shows that you couldn't before because of being locked into only what TheTVDB had to offer. Also this edition is based off the code we used in our XEM editon so it does come with scene numbering support as well as all the other features our XEM edition has to offer. Please before using this with your existing database (sickbeard.db) please make a backup copy of it and delete any other database files such as cache.db and failed.db if present, we HIGHLY recommend starting out with no database files at all to make this a fresh start but the choice is at your own risk! Enjoy!
121 lines
5.0 KiB
Python
121 lines
5.0 KiB
Python
"""
|
|
ReiserFS file system version 3 parser (version 1, 2 and 4 are not supported).
|
|
|
|
Author: Frederic Weisbecker
|
|
Creation date: 8 december 2006
|
|
|
|
Sources:
|
|
- http://p-nand-q.com/download/rfstool/reiserfs_docs.html
|
|
- http://homes.cerias.purdue.edu/~florian/reiser/reiserfs.php
|
|
- file://usr/src/linux-2.6.16.19/include/linux/reiserfs_fs.h
|
|
|
|
NOTES:
|
|
|
|
The most part of the description of the structures, their fields and their
|
|
comments decribed here comes from the file include/linux/reiserfs_fs.h
|
|
- written by Hans reiser - located in the Linux kernel 2.6.16.19 and from
|
|
the Reiserfs explanations in
|
|
http://p-nand-q.com/download/rfstool/reiserfs_docs.html written by Gerson
|
|
Kurz.
|
|
"""
|
|
|
|
|
|
from lib.hachoir_parser import Parser
|
|
from lib.hachoir_core.field import (FieldSet, Enum,
|
|
UInt16, UInt32, String, RawBytes, NullBytes)
|
|
from lib.hachoir_core.endian import LITTLE_ENDIAN
|
|
|
|
class Journal_params(FieldSet):
|
|
static_size = 32*8
|
|
|
|
def createFields(self):
|
|
yield UInt32(self, "1st_block", "Journal 1st block number")
|
|
yield UInt32(self, "dev", "Journal device number")
|
|
yield UInt32(self, "size", "Size of the journal")
|
|
yield UInt32(self, "trans_max", "Max number of blocks in a transaction")
|
|
#TODO: Must be explained: it was sb_journal_block_count
|
|
yield UInt32(self, "magic", "Random value made on fs creation.")
|
|
yield UInt32(self, "max_batch", "Max number of blocks to batch into a trans")
|
|
yield UInt32(self, "max_commit_age", "In seconds, how old can an async commit be")
|
|
yield UInt32(self, "max_trans_age", "In seconds, how old can a transaction be")
|
|
|
|
|
|
def createDescription(self):
|
|
return "Parameters of the journal"
|
|
|
|
class SuperBlock(FieldSet):
|
|
static_size = 204*8
|
|
|
|
UMOUNT_STATE = { 1: "unmounted", 2: "not unmounted" }
|
|
HASH_FUNCTIONS = {
|
|
0: "UNSET_HASH",
|
|
1: "TEA_HASH",
|
|
2: "YURA_HASH",
|
|
3: "R5_HASH"
|
|
}
|
|
|
|
def createFields(self):
|
|
#TODO: This structure is normally divided in two parts:
|
|
# _reiserfs_super_block_v1
|
|
# _reiserfs_super_block
|
|
# It will be divided later to easily support older version of the first part
|
|
yield UInt32(self, "block_count", "Number of blocks")
|
|
yield UInt32(self, "free_blocks", "Number of free blocks")
|
|
yield UInt32(self, "root_block", "Root block number")
|
|
yield Journal_params(self, "Journal parameters")
|
|
yield UInt16(self, "blocksize", "Size of a block")
|
|
yield UInt16(self, "oid_maxsize", "Max size of object id array")
|
|
yield UInt16(self, "oid_cursize", "Current size of object id array")
|
|
yield Enum(UInt16(self, "umount_state", "Filesystem umounted or not"), self.UMOUNT_STATE)
|
|
yield String(self, "magic", 10, "Magic string", strip="\0")
|
|
#TODO: change the type of s_fs_state in Enum to have more details about this fsck state
|
|
yield UInt16(self, "fs_state", "Rebuilding phase of fsck ")
|
|
yield Enum(UInt32(self, "hash_function", "Hash function to sort names in a directory"), self.HASH_FUNCTIONS)
|
|
yield UInt16(self, "tree_height", "Height of disk tree")
|
|
yield UInt16(self, "bmap_nr", "Amount of bitmap blocks needed to address each block of file system")
|
|
#TODO: find a good description for this field
|
|
yield UInt16(self, "version", "Field only reliable on filesystem with non-standard journal")
|
|
yield UInt16(self, "reserved_for_journal", "Size in blocks of journal area on main device")
|
|
#TODO: same as above
|
|
yield UInt32(self, "inode_generation", "No description")
|
|
#TODO: same as above and should be an enum field
|
|
yield UInt32(self, "flags", "No description")
|
|
#TODO: Create a special Type to format this id
|
|
yield RawBytes(self, "uuid", 16, "Filesystem unique identifier")
|
|
yield String(self, "label", 16, "Filesystem volume label", strip="\0")
|
|
yield NullBytes(self, "unused", 88)
|
|
|
|
def createDescription(self):
|
|
return "Superblock: ReiserFs Filesystem"
|
|
|
|
class REISER_FS(Parser):
|
|
PARSER_TAGS = {
|
|
"id": "reiserfs",
|
|
"category": "file_system",
|
|
# 130 blocks before the journal +
|
|
# Minimal size of journal (513 blocks) +
|
|
# 1 block for the rest
|
|
# And The Minimal size of a block is 512 bytes
|
|
"min_size": (130+513+1) * (512*8),
|
|
"description": "ReiserFS file system"
|
|
}
|
|
endian = LITTLE_ENDIAN
|
|
|
|
# Offsets (in bytes) of important information
|
|
SUPERBLOCK_OFFSET = 64*1024
|
|
MAGIC_OFFSET = SUPERBLOCK_OFFSET + 52
|
|
|
|
def validate(self):
|
|
# Let's look at the magic field in the superblock
|
|
magic = self.stream.readBytes(self.MAGIC_OFFSET*8, 9).rstrip("\0")
|
|
if magic == "ReIsEr3Fs":
|
|
return True
|
|
if magic in ("ReIsEr2Fs", "ReIsErFs"):
|
|
return "Unsupported version of ReiserFs"
|
|
return "Invalid magic string"
|
|
|
|
def createFields(self):
|
|
yield NullBytes(self, "padding[]", self.SUPERBLOCK_OFFSET)
|
|
yield SuperBlock(self, "superblock")
|
|
|