APT Cache File Format Jason Gunthorpe jgg@debian.org $Id: cache.sgml,v 1.11 2003/02/12 15:05:44 doogie Exp $ This document describes the complete implementation and format of the APT Cache file. The APT Cache file is a way for APT to parse and store a large number of package files for display in the UI. It's primary design goal is to make display of a single package in the tree very fast by pre-linking important things like dependencies and provides. The specification doubles as documentation for one of the in-memory structures used by the package library and the APT GUI. Copyright © Jason Gunthorpe, 1997-1998.

APT and this document are free software; you can redistribute them and/or modify them under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

For more details, on Debian GNU/Linux systems, see the file /usr/share/common-licenses/GPL for the full license. Introduction Purpose

This document describes the implementation of an architecture dependent binary cache file. The goal of this cache file is two fold, firstly to speed loading and processing of the package file array and secondly to reduce memory consumption of the package file array.

The implementation is aimed at an environment with many primary package files, for instance someone that has a Package file for their CD-ROM, a Package file for the latest version of the distribution on the CD-ROM and a package file for the development version. Always present is the information contained in the status file which might be considered a separate package file.

Please understand, this is designed as a -CACHE FILE- it is not meant to be used on any system other than the one it was created for. It is not meant to be authoritative either, i.e. if a system crash or software failure occurs it must be perfectly acceptable for the cache file to be in an inconsistent state. Furthermore at any time the cache file may be erased without losing any information.

Also the structures and storage layout is optimized for use by the APT GUI and may not be suitable for all purposes. However it should be possible to extend it with associate cache files that contain other information.

To keep memory use down the cache file only contains often used fields and fields that are inexpensive to store, the Package file has a full list of fields. Also the client may assume that all items are perfectly valid and need not perform checks against their correctness. Removal of information from the cache is possible, but blanks will be left in the file, and unused strings will also be present. The recommended implementation is to simply rebuild the cache each time any of the data files change. It is possible to add a new package file to the cache without any negative side effects. Note on Pointer access

Every item in every structure is stored as the index to that structure. What this means is that once the files is mmaped every data access has to go through a fixup stage to get a real memory pointer. This is done by taking the index, multiplying it by the type size and then adding it to the start address of the memory block. This sounds complex, but in C it is a single array dereference. Because all items are aligned to their size and indexes are stored as multiples of the size of the structure the format is immediately portable to all possible architectures - BUT the generated files are -NOT-.

This scheme allows code like this to be written: void *Map = mmap(...); Package *PkgList = (Package *)Map; Header *Head = (Header *)Map; char *Strings = (char *)Map; cout << (Strings + PkgList[Head->HashTable[0]]->Name) << endl;

Notice the lack of casting or multiplication. The net result is to return the name of the first package in the first hash bucket, without error checks.

The generator uses allocation pools to group similarly sized structures in large blocks to eliminate any alignment overhead. The generator also assures that no structures overlap and all indexes are unique. Although at first glance it may seem like there is the potential for two structures to exist at the same point the generator never allows this to happen. (See the discussion of free space pools) Structures Header

This is the first item in the file. struct Header { // Signature information unsigned long Signature; short MajorVersion; short MinorVersion; bool Dirty; // Size of structure values unsigned short HeaderSz; unsigned short PackageSz; unsigned short PackageFileSz; unsigned short VersionSz; unsigned short DependencySz; unsigned short ProvidesSz; unsigned short VerFileSz; // Structure counts unsigned long PackageCount; unsigned long VersionCount; unsigned long DependsCount; unsigned long PackageFileCount; // Offsets unsigned long FileList; // PackageFile unsigned long StringList; // StringItem unsigned long VerSysName; // StringTable unsigned long Architecture; // StringTable unsigned long MaxVerFileSize; // Allocation pools struct { unsigned long ItemSize; unsigned long Start; unsigned long Count; } Pools[7]; // Package name lookup unsigned long HashTable[2*1024]; // Package }; Signature This must contain the hex value 0x98FE76DC which is designed to verify that the system loading the image has the same byte order and byte size as the system saving the image MajorVersion MinorVersion These contain the version of the cache file, currently 0.2. Dirty Dirty is true if the cache file was opened for reading, the client expects to have written things to it and have not fully synced it. The file should be erased and rebuilt if it is true. HeaderSz PackageSz PackageFileSz VersionSz DependencySz VerFileSz ProvidesSz *Sz contains the sizeof() that particular structure. It is used as an extra consistency check on the structure of the file. If any of the size values do not exactly match what the client expects then the client should refuse the load the file. PackageCount VersionCount DependsCount PackageFileCount These indicate the number of each structure contained in the cache. PackageCount is especially useful for generating user state structures. See Package::Id for more info. VerSysName String representing the version system used for this cache Architecture Architecture the cache was built against. MaxVerFileSize The maximum size of a raw entry from the original Package file (i.e. VerFile::Size) is stored here. FileList This contains the index of the first PackageFile structure. The PackageFile structures are singly linked lists that represent all package files that have been merged into the cache. StringList This contains a list of all the unique strings (string item type strings) in the cache. The parser reads this list into memory so it can match strings against it. Pools The Pool structures manage the allocation pools that the generator uses. Start indicates the first byte of the pool, Count is the number of objects remaining in the pool and ItemSize is the structure size (alignment factor) of the pool. An ItemSize of 0 indicates the pool is empty. There should be the same number of pools as there are structure types. The generator stores this information so future additions can make use of any unused pool blocks. HashTable HashTable is a hash table that provides indexing for all of the packages. Each package name is inserted into the hash table using the following has function: unsigned long Hash(string Str) { unsigned long Hash = 0; for (const char *I = Str.begin(); I != Str.end(); I++) Hash += *I * ((Str.end() - I + 1)); return Hash % _count(Head.HashTable); }

By iterating over each entry in the hash table it is possible to iterate over the entire list of packages. Hash Collisions are handled with a singly linked list of packages based at the hash item. The linked list contains only packages that match the hashing function. Package

This contains information for a single unique package. There can be any number of versions of a given package. Package exists in a singly linked list of package records starting at the hash index of the name in the Header->HashTable. struct Pacakge { // Pointers unsigned long Name; // Stringtable unsigned long VersionList; // Version unsigned long CurrentVer; // Version unsigned long Section; // StringTable (StringItem) // Linked lists unsigned long NextPackage; // Package unsigned long RevDepends; // Dependency unsigned long ProvidesList; // Provides // Install/Remove/Purge etc unsigned char SelectedState; // What unsigned char InstState; // Flags unsigned char CurrentState; // State // Unique ID for this pkg unsigned short ID; unsigned long Flags; }; Name Name of the package. VersionList Base of a singly linked list of version structures. Each structure represents a unique version of the package. The version structures contain links into PackageFile and the original text file as well as detailed information about the size and dependencies of the specific package. In this way multiple versions of a package can be cleanly handled by the system. Furthermore, this linked list is guaranteed to be sorted from Highest version to lowest version with no duplicate entries. CurrentVer CurrentVer is an index to the installed version, either can be 0. Section This indicates the deduced section. It should be "Unknown" or the section of the last parsed item. NextPackage Next link in this hash item. This linked list is based at Header.HashTable and contains only packages with the same hash value. RevDepends Reverse Depends is a linked list of all dependencies linked to this package. ProvidesList This is a linked list of all provides for this package name. SelectedState InstState CurrentState These correspond to the 3 items in the Status field found in the status file. See the section on defines for the possible values.

SelectedState is the state that the user wishes the package to be in.

InstState is the installation state of the package. This normally should be OK, but if the installation had an accident it may be otherwise.

CurrentState indicates if the package is installed, partially installed or not installed. ID ID is a value from 0 to Header->PackageCount. It is a unique value assigned by the generator. This allows clients to create an array of size PackageCount and use it to store state information for the package map. For instance the status file emitter uses this to track which packages have been emitted already. Flags Flags are some useful indicators of the package's state. PackageFile

This contains information for a single package file. Package files are referenced by Version structures. This is a singly linked list based from Header.FileList struct PackageFile { // Names unsigned long FileName; // Stringtable unsigned long Archive; // Stringtable unsigned long Component; // Stringtable unsigned long Version; // Stringtable unsigned long Origin; // Stringtable unsigned long Label; // Stringtable unsigned long Architecture; // Stringtable unsigned long Site; // Stringtable unsigned long IndexType; // Stringtable unsigned long Size; // Linked list unsigned long NextFile; // PackageFile unsigned short ID; unsigned long Flags; time_t mtime; // Modification time }; FileName Refers the the physical disk file that this PacakgeFile represents. Archive Component Version Origin Label Architecture NotAutomatic This is the release information. Please see the files document for a description of what the release information means. Site The site the index file was fetched from. IndexType A string indicating what sort of index file this is. Size Size is provided as a simple check to ensure that the package file has not been altered. ID See Package::ID. Flags Provides some flags for the PackageFile, see the section on defines. mtime Modification time for the file at time of cache generation. Version

This contains the information for a single version of a package. This is a single linked list based from Package.Versionlist.

The version list is always sorted from highest version to lowest version by the generator. Also there may not be any duplicate entries in the list (same VerStr). struct Version { unsigned long VerStr; // Stringtable unsigned long Section; // StringTable (StringItem) unsigned long Arch; // StringTable // Lists unsigned long FileList; // VerFile unsigned long NextVer; // Version unsigned long DependsList; // Dependency unsigned long ParentPkg; // Package unsigned long ProvidesList; // Provides unsigned long Size; unsigned long InstalledSize; unsigned long Hash; unsigned short ID; unsigned char Priority; }; VerStr This is the complete version string. FileList References the all the PackageFile's that this version came out of. FileList can be used to determine what distribution(s) the Version applies to. If FileList is 0 then this is a blank version. The structure should also have a 0 in all other fields excluding VerStr and Possibly NextVer. Section This string indicates which section it is part of. The string should be contained in the StringItem list. Arch Architecture the package was compiled for. NextVer Next step in the linked list. DependsList This is the base of the dependency list. ParentPkg This links the version to the owning package, allowing reverse dependencies to determine the package. ProvidesList Head of the linked list of Provides::NextPkgProv, forward provides. Size InstalledSize The archive size for this version. For Debian this is the size of the .deb file. Installed size is the uncompressed size for this version Hash This is a characteristic value representing this package. No two packages in existence should have the same VerStr and Hash with different contents. ID See Package::ID. Priority This is the parsed priority value of the package. Dependency

Dependency contains the information for a single dependency record. The records are split up like this to ease processing by the client. The base of list linked list is Version.DependsList. All forms of dependencies are recorded here including Conflicts, Breaks, Suggests and Recommends.

Multiple depends on the same package must be grouped together in the Dependency lists. Clients should assume this is always true. struct Dependency { unsigned long Version; // Stringtable unsigned long Package; // Package unsigned long NextDepends; // Dependency unsigned long NextRevDepends; // Reverse dependency linking unsigned long ParentVer; // Upwards parent version link // Specific types of depends unsigned char Type; unsigned char CompareOp; unsigned short ID; }; Version The string form of the version that the dependency is applied against. Package The index of the package file this depends applies to. If the package file does not already exist when the dependency is inserted a blank one (no version records) should be created. NextDepends Linked list based off a Version structure of all the dependencies in that version. NextRevDepends Reverse dependency linking, based off a Package structure. This linked list is a list of all packages that have a depends line for a given package. ParentVer Parent version linking, allows the reverse dependency list to link back to the version and package that the dependency are for. Type Describes weather it is depends, predepends, recommends, suggests, etc. CompareOp Describes the comparison operator specified on the depends line. If the high bit is set then it is a logical or with the previous record. ID See Package::ID. Provides

Provides handles virtual packages. When a Provides: line is encountered a new provides record is added associating the package with a virtual package name. The provides structures are linked off the package structures. This simplifies the analysis of dependencies and other aspects A provides refers to a specific version of a specific package, not all versions need to provide that provides.

There is a linked list of provided package names started from each version that provides packages. This is the forwards provides mechanism. struct Provides { unsigned long ParentPkg; // Package unsigned long Version; // Version unsigned long ProvideVersion; // Stringtable unsigned long NextProvides; // Provides unsigned long NextPkgProv; // Provides }; ParentPkg The index of the package that head of this linked list is in. ParentPkg->Name is the name of the provides. Version The index of the version this provide line applies to. ProvideVersion Each provides can specify a version in the provides line. This version allows dependencies to depend on specific versions of a Provides, as well as allowing Provides to override existing packages. This is experimental. NextProvides Next link in the singly linked list of provides (based off package) NextPkgProv Next link in the singly linked list of provides for 'Version'. VerFile

VerFile associates a version with a PackageFile, this allows a full description of all Versions in all files (and hence all sources) under consideration. struct pkgCache::VerFile { unsigned long File; // PackageFile unsigned long NextFile; // PkgVerFile unsigned long Offset; unsigned short Size; } File The index of the package file that this version was found in. NextFile The next step in the linked list. Offset Size These describe the exact position in the package file for the section from this version. StringItem

StringItem is used for generating single instances of strings. Some things like Section Name are are useful to have as unique tags. It is part of a linked list based at Header::StringList. struct StringItem { unsigned long String; // Stringtable unsigned long NextItem; // StringItem }; String The string this refers to. NextItem Next link in the chain. StringTable

All strings are simply inlined any place in the file that is natural for the writer. The client should make no assumptions about the positioning of strings. All stringtable values point to a byte offset from the start of the file that a null terminated string will begin. Defines

Several structures use variables to indicate things. Here is a list of all of them. Definitions for Dependency::Type

#define pkgDEP_Depends 1 #define pkgDEP_PreDepends 2 #define pkgDEP_Suggests 3 #define pkgDEP_Recommends 4 #define pkgDEP_Conflicts 5 #define pkgDEP_Replaces 6 #define pkgDEP_Breaks 8 Definitions for Dependency::CompareOp

#define pkgOP_OR 0x10 #define pkgOP_LESSEQ 0x1 #define pkgOP_GREATEREQ 0x2 #define pkgOP_LESS 0x3 #define pkgOP_GREATER 0x4 #define pkgOP_EQUALS 0x5 The lower 4 bits are used to indicate what operator is being specified and the upper 4 bits are flags. pkgOP_OR indicates that the next package is or'd with the current package. Definitions for Package::SelectedState

#define pkgSTATE_Unkown 0 #define pkgSTATE_Install 1 #define pkgSTATE_Hold 2 #define pkgSTATE_DeInstall 3 #define pkgSTATE_Purge 4 Definitions for Package::InstState

#define pkgSTATE_Ok 0 #define pkgSTATE_ReInstReq 1 #define pkgSTATE_Hold 2 #define pkgSTATE_HoldReInstReq 3 Definitions for Package::CurrentState

#define pkgSTATE_NotInstalled 0 #define pkgSTATE_UnPacked 1 #define pkgSTATE_HalfConfigured 2 #define pkgSTATE_UnInstalled 3 #define pkgSTATE_HalfInstalled 4 #define pkgSTATE_ConfigFiles 5 #define pkgSTATE_Installed 6 #define pkgSTATE_TriggersAwaited 7 #define pkgSTATE_TriggersPending 8 Definitions for Package::Flags

#define pkgFLAG_Auto (1 << 0) #define pkgFLAG_New (1 << 1) #define pkgFLAG_Obsolete (1 << 2) #define pkgFLAG_Essential (1 << 3) #define pkgFLAG_ImmediateConf (1 << 4) Definitions for Version::Priority

Zero is used for unparsable or absent Priority fields. #define pkgPRIO_Important 1 #define pkgPRIO_Required 2 #define pkgPRIO_Standard 3 #define pkgPRIO_Optional 4 #define pkgPRIO_Extra 5 Definitions for PackageFile::Flags

#define pkgFLAG_NotSource (1 << 0) #define pkgFLAG_NotAutomatic (1 << 1) Notes on the Generator

The pkgCache::MergePackageFile function is currently the only generator of the cache file. It implements a conversion from the normal textual package file into the cache file.

The generator assumes any package declaration with a Status: line is a 'Status of the package' type of package declaration. A Package with a Target-Version field should also really have a status field. The processing of a Target-Version field can create a place-holder Version structure that is empty to refer to the specified version (See Version for info on what a empty Version looks like). The Target-Version syntax allows the specification of a specific version and a target distribution.

Different section names on different versions is supported, but I do not expect to use it. To simplify the GUI it will merely use the section in the Package structure. This should be okay as I hope sections do not change much.

The generator goes through a number of post processing steps after producing a disk file. It sorts all of the version lists to be in descending order and then generates the reverse dependency lists for all of the packages. ID numbers and count values are also generated in the post processing step.

It is possible to extend many of the structures in the cache with extra data. This is done by using the ID member. ID will be a unique number from 0 to Header->??Count. For example struct MyPkgData; MyPkgData *Data = new MyPkgData[Header->PackageCount]; Data[Package->ID]->Item = 0; This provides a one way reference between package structures and user data. To get a two way reference would require a member inside the MyPkgData structure.

The generators use of free space pools tend to make the package file quite large, and quite full of blank space. This could be fixed with sparse files. Future Directions

Some good directions to take the cache file is into a cache directory that contains many associated caches that cache other important bits of information. (/var/cache/apt, FHS2)

Caching of the info/*.list is an excellent place to start, by generating all the list files into a tree structure and reverse linking them to the package structures in the main cache file major speed gains in dpkg might be achieved.