@DATABASE AmigaMail @NODE MAIN "II-5: AmigaDOS Packet Interface Specification" @TOC "Table_of_Contents/II" AmigaDOS Packet Interface Specification by John Toebes AmigaDOS communicates with file systems and other DOS handlers by sending and receiving packets. Opening and closing file handles (including console file handles), creating directories, and renaming disks all require DOS to tell a handler to perform these actions through sending a packet. The particular action a handler performs depends on the type of packet it receives. This article documents the standard AmigaDOS packet types. For information on how to use packets to communicate with handlers see the AmigaDOS Manual. Packets sent to a file system or handler can be divided into several basic categories: o @{"Basic Input/Output" link Basic} These actions deal with tranferring data to and from objects controlled by the handler. o @{"File/Directory Manipulation/Information" link DirFileMan} These actions are used to gain access to and manipulate the high level structures of the file system. o @{"Volume Manipulation/Information" link VolMan} These actions allow access to the specific volume controlled by the file system. o @{"Handler Maintenance and Control" link HandMain} These allow control over the handler/file system itself, independent of the actual volume or structure underneath. o @{"Handler Internal" link HandInt} These actions are never sent to the handler directly. Instead they are generally responses to IO requests made by the handler. The handler makes these responses look like packets in order to simplify processing. o @{"Obsolete Packets" link Obs} These packets are no longer valid for use by handlers and file systems. o @{"Console Only Packets" link ConOnly} These packets are specific to console handlers. File Systems can ignore these packets. Much of this information can be extracted from Developer Conference notes, The AmigaDOS Manual, and various Fred Fish disks. However, because there is no single complete reference to these packet types, a consolidated view of all the packets is presented here. Several structures are referenced here which can be found by looking at the include files and . (If you are using the 1.3 version of the include files, these are in the libraries directory instead of the dos directory). Before attempting to work with a file handler you should first become familiar with these files. Each packet type documented in this article is listed with its action name, its corresponding number, any AmigaDOS routines which uses this packet, and the list of parameters that the packets uses. The C variable types for the packet parameters are one of the following types: BPTR This is BCPL pointer (the address of the given object shifted right by 2). Note: this means that the object must be aligned on a longword boundary. LOCK This is a BPTR to a FileLock structure returned by a previous ACTION_LOCATE_OBJECT. A lock of 0 is legal, indicating the root of the volume for the handler. BSTR This is a BPTR to a string where the first byte indicates the number of characters in the string. This length byte is unsigned but because it is stored in a byte, the strings are limited to 255 characters in length. BOOL A 32-bit boolean value either containing DOSTRUE (-1) or DOSFALSE (0). Note: equality comparisons with DOSTRUE should be avoided. CODE A 32 bit error code as defined in the dos/dos.h include file. Handlers should not return error codes besides those defined in dos/dos.h. ARG1 The FileHandle->fh_Arg1 field. LONG A 32 bit integer value. Summary of Defined Packet Numbers This is a listing of all the DOS packets defined by Commodore. Packets 0-1999 are reserved for use by Commodore. Unless otherwise noted, packets 2050-2999 are reserved for use by third party developers (see chart below). The remaining packets are reserved for future expansion (Note: packets 2008, 2009, 4097, and 4098 are in use by Commodore). Decimal Hex Action #define 0 0x0000 ACTION_NIL 1 2 0x0002 ACTION_GET_BLOCK 3 4 0x0004 ACTION_SET_MAP 5 0x0005 ACTION_DIE 6 0x0006 ACTION_EVENT 7 0x0007 ACTION_CURRENT_VOLUME 8 0x0008 ACTION_LOCATE_OBJECT 9 0x0009 ACTION_RENAME_DISK 10-14 15 0x000F ACTION_FREE_LOCK 16 0x0010 ACTION_DELETE_OBJECT 17 0x0011 ACTION_RENAME_OBJECT 18 0x0012 ACTION_MORE_CACHE 19 0x0013 ACTION_COPY_DIR 20 0x0014 ACTION_WAIT_CHAR 21 0x0015 ACTION_SET_PROTECT 22 0x0016 ACTION_CREATE_DIR 23 0x0017 ACTION_EXAMINE_OBJECT 24 0x0018 ACTION_EXAMINE_NEXT 25 0x0019 ACTION_DISK_INFO 26 0x001A ACTION_INFO 27 0x001B ACTION_FLUSH 28 0x001C ACTION_SET_COMMENT 29 0x001D ACTION_PARENT 30 0x001E ACTION_TIMER 31 0x001F ACTION_INHIBIT 32 0x0020 ACTION_DISK_TYPE 33 0x0021 ACTION_DISK_CHANGE 34 0x0022 ACTION_SET_DATE 35-39 40 0x0028 ACTION_SAME_LOCK 41-81 82 0x0052 ACTION_READ 83-86 87 0x0057 ACTION_WRITE 88-993 994 0x03E2 ACTION_SCREEN_MODE 995 0x03E3 ACTION_CHANGE_SIGNAL 996-1000 1001 0x03E9 ACTION_READ_RETURN 1002 0x03EA ACTION_WRITE_RETURN 1003 1004 0x03EC ACTION_FINDUPDATE 1005 0x03ED ACTION_FINDINPUT 1006 0x03EE ACTION_FINDOUTPUT 1007 0x03EF ACTION_END 1008 0x03F0 ACTION_SEEK 1009-1019 1020 0x03FC ACTION_FORMAT 1021 0x03FD ACTION_MAKE_LINK 1022 0x03FE ACTION_SET_FILE_SIZE 1023 0x03FF ACTION_WRITE_PROTECT 1024 0x0400 ACTION_READ_LINK 1025 1026 0x0402 ACTION_FH_FROM_LOCK 1027 0x0403 ACTION_IS_FILESYSTEM 1028 0x0404 ACTION_CHANGE_MODE 1029 1030 0x0406 ACTION_COPY_DIR_FH 1031 0x0407 ACTION_PARENT_FH 1032 1033 0x0409 ACTION_EXAMINE_ALL 1034 0x040A ACTION_EXAMINE_FH 1035-2007 2008 0x07D8 ACTION_LOCK_RECORD 2009 0x07D9 ACTION_FREE_RECORD 2010-2049 2050-2999 4097 0x1001 ACTION_ADD_NOTIFY 4098 0x1002 ACTION_REMOVE_NOTIFY 4099- @ENDNODE @NODE Basic "Basic Input/Output" The Basic Input/Output actions are supported by both handlers and file systems. In this way, the application can get a stream level access to both devices and files. One difference that arises between the two is that a handler will not necessarily support an ACTION_SEEK while it is generally expected for a file system to do so. These actions work based on a FileHandle which is filled in by one of the three forms of opens: ACTION_FINDINPUT 1005 Open(..., MODE_OLDFILE) ACTION_FINDOUTPUT 1006 Open(..., MODE_NEWFILE) ACTION_FINDUPDATE 1004 Open(..., MODE_READWRITE) ARG1: BPTR FileHandle to fill in ARG2: LOCK Lock on directory that ARG3 is relative to ARG3: BSTR Name of file to be opened (relative to ARG1) RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE All three actions use the lock (ARG2) as a base directory location from which to open the file. If this lock is NULL, then the file name (ARG3) is relative to the root of the current volume. Because of this, file names are not limited to a single file name but instead can include a volume name (followed by a colon) and multiple slashes allowing the file system to fully resolve the name. This eliminates the need for AmigaDOS or the application to parse names before sending them to the file system. Note that the lock in ARG2 must be associated with the file system in question. It is illegal to use a lock from another file system. The calling program owns the file handle (ARG1). The program must initialize the file handle before trying to open anything (in the case of a call to Open(), AmigaDOS allocates the file handle automatically and then frees it in Close() ). All fields must be zero except the fh_Pos and fh_End fields which should be set to -1. The Open() function fills in the fh_Type field with a pointer to the MsgPort of the handler process. Lastly, the handler must initialize fh_Arg1 with something that allows the handler to uniquely locate the object being opened (normally a file). This value is implementation specific. This field is passed to the READ/WRITE/SEEK/ END/TRUNCATE operations and not the file handle itself. FINDINPUT and FINDUPDATE are similar in that they only succeed if the file already exists. FINDINPUT will open with a shared lock while FINDUPDATE will open it with a shared lock but if the file doesn't exist, FINDUPDATE will create the file. FINDOUTPUT will always open the file (deleting any existing one) with an exclusive lock. ACTION_READ 'R' Read(...) ARG1: ARG1 fh_Arg1 field of the opened FileHandle ARG2: APTR Buffer to put data into ARG3: LONG Number of bytes to read RES1: LONG Number of bytes read. 0 indicates EOF. -1 indicates ERROR RES2: CODE Failure code if RES1 is -1 This action extracts data from the file (or input channel) at the current position. If fewer bytes remain in the file than requested, only those bytes remaining will be returned with the number of bytes stored in RES1. The handler indicates an error is indicated by placing a -1 in RES1 and the error code in RES2. If the read fails, the current file position remains unchanged. Note that a handler may return a smaller number of bytes than requested, even if not at the end of a file. This happens with interactive type file handles which may return one line at a time as the user hits return, for example the console handler, CON:. ACTION_WRITE 'W' Write(...) ARG1: ARG1 fh_Arg1 field of the opened file handle ARG2: APTR Buffer to write to the file handle ARG3: LONG Number of bytes to write RES1: LONG Number of bytes written. RES2: CODE Failure code if RES1 not the same as ARG3 This action copies data into the file (or output channel) at the current position. The file is automatically extended if the write passes the end of the file. The handler indicates failure by returning a byte count in RES1 that differs from the number of bytes requested in ARG3. In the case of a failure, the handler does not update the current file position (although the file may have been extended and some data overwritten) so that an application can safely retry the operation. ACTION_SEEK 1008 Seek(...) ARG1: ARG1 fh_Arg1 field of the opened FileHandle ARG2: LONG New Position ARG3: LONG Mode: OFFSET_BEGINNING,OFFSET_END, or OFFSET_CURRENT RES1: LONG Old Position. -1 indicates an error RES2: CODE Failure code if RES1 = -1 This packet sets the current file position. The new position (ARG2) is relative to either the beginning of the file (OFFSET_BEGINNING), the end of the file (OFFSET_END), or the current file position (OFFSET_CURRENT), depending on the mode set in ARG3. Note that ARG2 can be negative. The handler returns the previous file position in RES1. Any attempt to seek past the end of the file will result in an error and will leave the current file position in an unknown location. ACTION_END 1007 Close(...) ARG1: ARG1 fh_Arg1 field of the opened FileHandle RES1: LONG DOSTRUE This packet closes an open file handle. This function generally returns a DOSTRUE as there is little the application can do to recover from a file closing failure. If an error is returned under 2.0, DOS will not deallocate the file handle. Under 1.3, it does not check the result. ACTION_LOCK_RECORD 2008 LockRecord(fh,pos,len,mod,tim) ARG1: BPTR FileHandle to lock record in ARG2: LONG Start position (in bytes) of record in the file ARG3: LONG Length (in bytes) of record to be locked ARG4: LONG Mode 0 = Exclusive 1 = Immediate Exclusive (timeout is ignored) 2 = Shared 3 = Immediate Shared (timeout is ignored) ARG5: LONG Timeout period in AmigaDOS ticks (0 is legal) RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE This function locks an area of a file in either a sharable (indicating read-only) or exclusive (indicating read/write) mode. Several sharable record locks from different file handles can exist simultaneously on a particular file area but only one file handle can have exclusive record locks on a particular area at a time. The ``exclusivity'' of an exclusive file lock only applies to record locks from other file handles, not to record locks within the file handle. One file handle can have any number of overlapping exclusive record locks. In the event of overlapping lock ranges, the entire range must be lockable before the request can succeed. The timeout period (ARG5) is the number of AmigaDOS ticks (1/50 second) to wait for success before failing the operation. ACTION_FREE_RECORD 2009 UnLockRecord(file,pos,len) ARG1: BPTR FileHandle to unlock record in ARG2: LONG Start position (in bytes) of record in the file ARG3: LONG Length of record (in bytes) to be unlocked RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE This function unlocks any previous record lock. If the given range does not represent one that is currently locked in the file, ACTION_FREE_RECORD returns an error. In the event of multiple locks on a given area, only one lock is freed. ACTION_SET_FILE_SIZE 1022 SetFileSize(file,off,mode) ARG1: BPTR FileHandle of opened file to modify ARG2: LONG New end of file location based on mode ARG3: LONG Mode. One of OFFSET_CURRENT, OFFSET_BEGIN, or OFFSET_END RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE This function is used to change the physical size of an opened file. ARG2, the new end-of-file position, is relative to either the current file position (OFFSET_CURRENT), the beginning of the file (OFFSET_BEGIN), or the end of the file (OFFSET_END), depending on the mode set in ARG3. The current file position will not change unless the current file position is past the new end-of-file position. In this case, the new file position will move to the new end of the file. If there are other open file handles on this file, ACTION_SET_FILE_SIZE sets the end-of-file for these alternate file handles to either their respective current file position or to the new end-of-file position of the file handle in ARG1, whichever makes the file appear longer. @ENDNODE @NODE DirFileMan "Directory/File Manipulation/Information" The directory/file actions permits an application to make queries about and modifications to handler objects. These packets perform functions such as creating subdirectories, resolving links, and filling in FileInfoBlock structures for specific files. ACTION_LOCATE_OBJECT 8 Lock(...) ARG1: LOCK Lock on directory to which ARG2 is relative ARG2: BSTR Name (possibly with a path) of object to lock ARG3: LONG Mode: ACCESS_READ/SHARED_LOCK, ACCESS_WRITE/EXCLUSIVE_LOCK RES1: LOCK Lock on requested object or 0 to indicate failure RES2: CODE Failure code if RES1 = 0 The AmigaDOS function Lock() uses this action to create its locks. Given a name for the object, which may include a path, (ARG2) and a lock on a directory from which to look for the name (and path), ACTION_LOCATE_OBJECT will locate the object within the file system and create a FileLock structure associated with the object. If the directory lock in ARG1 is NULL, the name is relative to the root of the file handler's volume (a.k.a. ``:''). The memory for the FileLock structure returned in RES1 is maintained by the handler and freed by an ACTION_FREE_LOCK. Although it's not a requirement, if an handler expects to support the pre-1.3 Format command, it must accept any illegal mode as ACCESS_READ. A handler can create an exclusive lock only if there are no other outstanding locks on the given object. Once created, an exclusive lock prevents any other locks from being created for that object. In general, a handler uses the FileLock->fl_Key field to uniquely identify an object. Note that some applications rely on this (although a handler is not required to implement this packet). The fl_Volume field of the returned FileLock structure should point to the DOS device list's volume entry for the volume on which the lock exists. In addition, there are several diagnostic programs that expect all locks for a volume to be chained together off the dl_LockList field in the volume entry. Note that relying on this chaining is not safe, and can cause serious problems including a system crash. No application should use it. ACTION_COPY_DIR 19 DupLock(...) ARG1: LOCK Lock to duplicate RES1: LOCK Duplicated Lock or 0 to indicate failure RES2: CODE Failure code if RES1 = 0 This action's name is misleading as it does not manipulate directories. Instead, it creates a copy of a shared lock. The copy is subsequently freed with an ACTION_FREE_LOCK. Note that it is valid to pass a NULL lock. Currently, the DupLock() call always returns 0 if passed a 0, although a handler is not required to return a 0. ACTION_FREE_LOCK 15 UnLock(...) ARG1: LOCK Lock to free RES1: BOOL TRUE This action frees the lock passed to it. The AmigaDOS function Unlock() uses this packet. If passed a NULL lock, the handler should return success. ACTION_EXAMINE_OBJECT 23 Examine(...) ARG1: LOCK Lock of object to examine ARG2: BPTR FileInfoBlock to fill in RES1: BOOL Success/failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 = DOSFALSE This action fills in the FileInfoBlock with information about the locked object. The Examine() function uses this packet. This packet is actually used for two different types of operations. It is called to obtain information about a given object while in other cases, it is called to prepare for a sequence of EXAMINE_NEXT operations in order to traverse a directory. This seemingly simple operation is not without its quirks. One in particular is the FileInfoBlock->fib_Comment field. This field used to be 116 bytes long, but was changed to 80 bytes in release 1.2. The extra 36 bytes lie in the fib_Reserved field. Another quirk of this packet is that both the fib_EntryType and the fib_DirEntryType fields must be set to the same value, as some programs look at one field while other programs look at the other. File systems should use the same values for fib_DirEntryType as the ROM file system and ram-handler do. These are as follows: ST_ROOT 1 ST_USERDIR 2 ST_SOFTLINK 3 NOTE: this Shows up as a directory unless checked for explicitly ST_LINKDIR 4 ST_FILE -3 ST_LINKFILE -4 Also note that for directories, handlers must use numbers greater than 0, since some programs test to see if fib_DirEntryType is greater than zero, ignoring the case where fib_DirEntryType equals 0. Handlers should avoid using 0 because it is not interpreted consistently. ACTION_EXAMINE_NEXT 24 ExNext(...) ARG1: LOCK Lock on directory being examined ARG2: BPTR BPTR FileInfoBlock RES1: BOOL Success/failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 = DOSFALSE The ExNext() function uses this packet to obtain information on all the objects in a directory. ACTION_EXAMINE fills in a FileInfoBlock structure describing the first file or directory stored in the directory referred to in the lock in ARG1. ACTION_EXAMINE_NEXT is used to find out about the rest of the files and directories stored in the ARG1 directory. ARG2 contains a pointer to a valid FileInfoBlock field that was filled in by either an ACTION_EXAMINE or a previous ACTION_EXAMINE_NEXT call. It uses this structure to find the next entry in the directory. This packets writes over the old FileInfoBlock with information on the next file or directory in the ARG2 directory. ACTION_EXAMINE_NEXT returns a failure code of ERROR_NO_MORE_ENTRIES when there are no more files or directories left to be examined. Unfortunately, like ACTION_EXAMINE, this packet has its own peculiarities. Among the quirks that ACTION_EXAMINE_NEXT must account for are: · The situation where an application calls ACTION_EXAMINE_NEXT one or more times and then stops invoking it before encountering the end of the directory. · The situation where a FileInfoBlock passed to ACTION_EXAMINE_NEXT is not the same as the one passed to ACTION_EXAMINE or even the previous EXAMINE_NEXT operation. Instead, it is a copy of the FileInfoBlock with only the fib_DiskKey and the first 30 bytes of the fib_FileName fields copied over. This is now considered to be illegal and will not work in the future. Any new code should not be written in this manner. · Because a handler can receive other packet types between ACTION_EXAMINE_NEXT operations, the ACTION_EXAMINE_NEXT function must handle any special cases that may result. · The LOCK passed to ACTION_EXAMINE_NEXT is not always the same lock used in previous operations. It is however a lock on the same object. Because of these problems, ACTION_EXAMINE_NEXT is probably the trickiest action to write in any handler. Failure to handle any of the above cases can be quite disastrous. ACTION_CREATE_DIR 22 CreateDir(...) ARG1: LOCK Lock to which ARG2 is relative ARG2: BSTR Name of new directory (relative to ARG1) RES1: LOCK Lock on new directory RES2: CODE Failure code if RES1 = DOSFALSE ACTION_DELETE_OBJECT 16 DeleteFile(...) ARG1: LOCK Lock to which ARG2 is relative ARG2: BSTR Name of object to delete (relative to ARG1) RES1: BOOL Success/failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 = DOSFALSE ACTION_RENAME_OBJECT 17 Rename(...) ARG1: LOCK Lock to which ARG2 is relative ARG2: BSTR Name of object to rename (relative to ARG1) ARG3: LOCK Lock associated with target directory ARG4: BSTR Requested new name for the object RES1: BOOL Success/failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 = DOSFALSE These three actions perform most of the work behind the AmigaDOS commands MakeDir, Delete, and Rename (for single files). These packets take as their parameters a lock describing where the file is and a name relative to that lock. It is the responsibility of the file system to ensure that the operation is not going to cause adverse effects. In particular, the RENAME_OBJECT action allows moving files across directory bounds and as such must ensure that it doesn't create hidden directory loops by renaming a directory into a child of itself. For Directory objects, the DELETE_OBJECT action must ensure that the directory is empty before allowing the operation. ACTION_PARENT 29 Parent(...) ARG1: LOCK Lock on object to get the parent of RES1: LOCK Parent Lock RES2: CODE Failure code if RES1 = 0 This action receives a lock on an object and creates a shared lock on the object's parent. If the original object has no parent, then a lock of 0 is returned. Note that this operation is typically used in the process of constructing the absolute path name of a given object. ACTION_SET_PROTECT 21 SetProtection(...) ARG1: Unused ARG2: LOCK Lock to which ARG3 is relative ARG3: BSTR Name of object (relative to ARG2) ARG4: LONG Mask of new protection bits RES1: BOOL Success/failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 = DOSFALSE This action allows an application to modify the protection bits of an object. The 4 lowest order bits (RWED) are a bit peculiar. If their respective bit is set, that operation is not allowed (i.e. if a file's delete bit is set the file is not deleteable). By default, files are created with the RWED bits set and all others cleared. Additionally, any action which modifies a file is required to clear the A (archive) bit. See the dos/dos.h include file for the definitions of the bit fields. ACTION_SET_COMMENT 28 SetComment(...) ARG1: Unused ARG2: LOCK Lock to which ARG3 is relative ARG3: BSTR Name of object (relative to ARG2) ARG4: BSTR New Comment string RES1: BOOL Success/failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 = DOSFALSE This action allows an application to set the comment string of an object. If the object does not exist then DOSFALSE will be returned in RES1 with the failure code in RES2. The comment string is limited to 79 characters. ACTION_SET_DATE 34 SetFileDate(...) in 2.0 ARG1: Unused ARG2: LOCK Lock to which ARG3 is relative ARG3: BSTR Name of Object (relative to ARG2) ARG4: CPTR DateStamp RES1: BOOL Success/failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 = DOSFALSE This action allows an application to set an object's creation date. ACTION_FH_FROM_LOCK 1026 OpenFromLock(lock) ARG1: BPTR BPTR to file handle to fill in ARG2: LOCK Lock of file to open RES1: BOOL Success/failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 = NULL This action open a file from a given lock. If this action is successful, the file system will essentially steal the lock so a program should not use it anymore. If ACTION_FH_FROM_LOCK fails, the lock is still usable by an application. ACTION_SAME_LOCK 40 SameLock(lock1,lock2) ARG1: BPTR Lock 1 to compare ARG2: BPTR Lock 2 to compare RES1: LONG Result of comparison, one of DOSTRUE if locks are for the same object DOSFALSE if locks are on different objects RES2: CODE Failure code if RES1 is LOCK_DIFFERENT This action compares the targets of two locks. If they point to the same object, ACTION_SAME_LOCK should return LOCK_SAME. ACTION_MAKE_LINK 1021 MakeLink(name,targ,mode) ARG1: BPTR Lock on directory ARG2 is relative to ARG2: BSTR Name of the link to be created (relative to ARG1) ARG3: BPTR Lock on target object or name (for soft links). ARG4: LONG Mode of link, either LINK_SOFT or LINK_HARD RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE This packet causes the file system to create a link to an already existing file or directory. There are two kinds of links, hard links and soft links. The basic difference between them is that a file system resolves a hard link itself, while the file system passes a string back to DOS telling it where to find a soft linked file or directory. To the packet level programmer, there is essentially no difference between referencing a file by its original name or by its hard link name. In the case of a hard link, ARG3 is a lock on the file or directory that the link is ``linked'' to, while in a soft link, ARG3 is a pointer (CPTR) to a C-style string. In an over-simplified model of the ROM file system, when asked to locate a file, the system scans a disk looking for a file header with a specific (file) name. That file header points to the actual file data somewhere on the disk. With hard links, more than one file header can point to the same file data, so data can be referenced by more than one name. When the user tries to delete a hard link to a file, the system first checks to see if there are any other hard links to the file. If there are, only the hard link is deleted, the actual file data the hard link used to reference remains, so the existing hard links can still use it. In the case where the original link (not a hard or soft link) to a file is deleted, the file system will make one of its hard links the new ``real'' link to the file. Hard links can exist on directories as well. Because hard links ``link'' directly to the underlying media, hard links in one file system cannot reference objects in another file system. Soft links are resolved through DOS calls. When the file system scans a disk for a file or directory name and finds that the name is a soft link, it returns an error code (ERROR_IS_SOFT_LINK). If this happens, the application must ask the file system to tell it what the link the link refers to by calling ACTION_READ_LINK. Soft Links are stored on the media, but instead of pointing directly to data on the disk, a soft link contains a path to its object. This path can be relative to the lock in ARG1, relative to the volume (where the string will be prepended by a colon ':'), or an absolute path. An absolute path contains the name of another volume, so a soft link can reference files and directories on other disks. ACTION_READ_LINK 1024 ReadLink(port,lck,nam,buf,len) ARG1: BPTR Lock on directory that ARG2 is relative to ARG2: CPTR Path and name of link (relative to ARG1). NOTE: This is a C string not a BSTR ARG3: APTR Buffer for new path string ARG4: LONG Size of buffer in bytes RES1: LONG Actual length of returned string, -2 if there isn't enough space in buffer,or -1 for other errors RES2: CODE Failure code This action reads a link and returns a path name to the link's object. The link's name (plus any necessary path) is passed as a CPTR (ARG2) which points to a C-style string, not a BSTR. ACTION_READ_LINK returns the path name in ARG3. The length of the target string is returned in RES1 (or a -1 indicating an error). ACTION_CHANGE_MODE 1028 ChangeMode(type,obj,mode) ARG1: LONG Type of object to change - either CHANGE_FH or CHANGE_LOCK ARG2: BPTR object to be changed ARG3: LONG New mode for object - see ACTION_FINDINPUT, and ACTION_LOCATE_OBJECT RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE This action requests that the handler change the mode of the given file handle or lock to the mode in ARG3. This request should fail if the handler can't change the mode as requested (for example an exclusive request for an object that has multiple users). ACTION_COPY_DIR_FH 1030 DupLockFromFH(fh) ARG1: LONG fh_Arg1 of file handle RES1: BPTR Lock associated with file handle or NULL RES2: CODE Failure code if RES1 = NULL This action requests that the handler return a lock associated with the currently opened file handle. The request may fail for any restriction imposed by the file system (for example when the file handle is not opened in a shared mode). The file handle is still usable after this call, unlike the lock in ACTION_FH_FROM_LOCK. ACTION_PARENT_FH 1031 ParentOfFH(fh) ARG1: LONG fh_Arg1 of File handle to get parent of RES1: BPTR Lock on parent of a file handle RES2: CODE Failure code if RES1 = NULL This action obtains a lock on the parent directory (or root of the volume if at the top level) for a currently opened file handle. The lock is returned as a shared lock and must be freed. Note that unlike ACTION_COPY_DIR_FH, the mode of the file handle is unimportant. For an open file, ACTION_PARENT_FH should return a lock under all circumstances. ACTION_EXAMINE_ALL 1033 ExAll(lock,buff,size,type,ctl) ARG1: BPTR Lock on directory to examine ARG2: APTR Buffer to store results ARG3: LONG Length (in bytes) of buffer (ARG2) ARG4: LONG Type of request - one of the following: ED_NAME Return only file names ED_TYPE Return above plus file type ED_SIZE Return above plus file size ED_PROTECTION Return above plus file protection ED_DATE Return above plus 3 longwords of date ED_COMMENT Return above plus comment or NULL ARG5: BPTR Control structure to store state information. The control structure must be allocated with AllocDosObject()! RES1: LONG Continuation flag - DOSFALSE indicates termination RES2: CODE Failure code if RES1 is DOSFALSE This action allows an application to obtain information on multiple directory entries. It is particularly useful for applications that need to obtain information on a large number of files and directories. This action fills the buffer (ARG2) with partial or whole ExAllData structures. The size of the ExAllData structure depends on the type of request. If the request type field (ARG4) is set to ED_NAME, only the ed_Name field is filled in. Instead of copying the unused fields of the ExAllData structure into the buffer, ACTION_EXAMINE_ALL truncates the unused fields. This effect is cumulative, so requests to fill in other fields in the ExAllData structure causes all fields that appear in the structure before the requested field will be filled in as well. Like the ED_NAME case mentioned above, any field that appears after the requested field will be truncated (see the ExAllData structure below). For example, if the request field is set to ED_COMMENT, ACTION_EXAMINE_ALL fills in all the fields of the ExAllData structure, because the ed_Comment field is last. This is the only case where the packet returns entire ExAllData structures. struct ExAllData { struct ExAllData *ed_Next; UBYTE *ed_Name; LONG ed_Type; ULONG ed_Size; ULONG ed_Prot; ULONG ed_Days; ULONG ed_Mins; ULONG ed_Ticks; UBYTE *ed_Comment; /* strings will be after last used field. Note: */ }; /* Bug in V37 FFS treats this as a BSTR. */ Each ExAllData structure entry has an ead_Next field which points to the next ExAllData structure. Using these links, a program can easily chain through the ExAllData structures without having to worry about how large the structure is. Do not examine the fields beyond those requested as they certainly will not be initialized (and will probably overlay the next entry). The most important part of this action is the ExAllControl structure. It must be allocated and freed through AllocDosObject()/FreeDosObject(). This allows the structure to grow if necessary with future revisions of the operating and file systems. Currently, ExAllControl contains four fields: Entries - This field is maintained by the file system and indicates the actual number of entries present in the buffer after the action is complete. Note that a value of zero is possible here as no entries may match the match string. LastKey - This field must be initialized to 0 by the calling application before using this packet for the first time. This field is maintained by the file system as a state indicator of the current place in the list of entries to be examined. The file system may test this field to determine if this is the first or a subsequent call to this action. MatchString - This field points to a pattern matching string parsed by ParsePattern() or ParsePatternNoCase(). The string controls which directory entries are returned. If this field is NULL, then all entries are returned. Otherwise, this string is used to pattern match the names of all directory entries before putting them into the buffer. The default AmigaDOS pattern match routine is used unless MatchFunc is not NULL (see below). Note that it is not acceptable for the application to change this field between subsequent calls to this action for the same directory. MatchFunc - This field contains a pointer to an alternate pattern matching routine to validate entries. If it is NULL then the standard AmigaDOS wild card routines will be used. Otherwise, MatchFunc points to a hook function that is called in the following manner: BOOL = MatchFunc(hookptr, data,typeptr) A0 A1 A2 hookptr Pointer to hook being called data Pointer to (partially) filled in ExAllData for item being checked. typeptr Pointer to longword indicating the type of the ExAll request (ARG4). This function is expected to return DOSTRUE if the entry is accepted and DOSFALSE if it is to be discarded. ACTION_EXAMINE_FH 1034 ExamineFH(fh,fib) ARG1: BPTR File handle on open file ARG2: BPTR FileInfoBlock to fill in RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE This function examines a file handle and fills in the FileInfoBlock (found in ARG2) with information about the current state of the file. This routine is analogous to the ACTION_EXAMINE_OBJECT action for locks. Because it is not always possible to provide an accurate file size (for example when buffers have not been flushed or two processes are writing to a file), the fib_Size field (see dos/dos.h) may be inaccurate. ACTION_ADD_NOTIFY 4097 StartNotify(NotifyRequest) ARG1: BPTR NotifyRequest structure RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE This action asks a file system to notify the calling program if a particular file is altered. A file system notifies a program either by sending a message or by signaling a task. struct NotifyRequest { UBYTE *nr_Name; UBYTE *nr_FullName; /* set by dos - don't touch */ ULONG nr_UserData; /* for applications use */ ULONG nr_Flags; union { struct { struct MsgPort *nr_Port; /* for SEND_MESSAGE */ } nr_Msg; struct { struct Task *nr_Task; /* for SEND_SIGNAL */ UBYTE nr_SignalNum; /* for SEND_SIGNAL */ UBYTE nr_pad[3]; } nr_Signal; } nr_stuff; ULONG nr_Reserved[4]; /* leave 0 for now */ /* internal use by handlers */ ULONG nr_MsgCount; /* # of outstanding msgs */ struct MsgPort *nr_Handler; /* handler sent to (for EndNotify) */ }; To use this packet, an application needs to allocate and initialize a NotifyRequest structure (see above). As of this writing, NotifyRequest structures are not allocated by AllocDosObject(), but this may change in the future. The handler gets the watched file's name from the nr_FullName field. The current file system does not currently support wild cards in this field, although there is nothing to prevent other handlers from doing so. The string in nr_FullName must be an absolute path, including the name of the root volume (no assigns). The absolute path is necessary because the file or its parent directories do not have to exist when the notification is set up. This allows notification on files in directories that do not yet exist. Notification will not occur until the directories and file are created. An application that uses the StartNotify() DOS call does not fill in the NotifyRequest's nr_FullName field, but instead fills in the nr_Name field. StartNotify() takes the name from the nr_Name field and uses GetDeviceProc() and NameFromLock() to expand any assigns (such as ENV:), storing the result in nr_FullName. Any application utilizing the packet level interface instead of StartNotify() must expand their own assigns. Handlers must not count on nr_Name being correct. The notification type depends on which bit is set in the NotifyRequest.nr_Flags field. If the NRF_SEND_MESSAGE bit is set, an application receives notification of changes to the file through a message (see NotifyMessage from dos/notify.h). In this case, the nr_Port field must point to the message port that will receive the notifying message . If the nr_Flags NRF_SEND_SIGNAL bit is set, the file system will signal a task instead of sending a message. In this case, nr_Task points to the task and nr_SignalNum is the signal number. Only one of these two bits should be set! When an application wants to limit the number of NotifyMessages an handler can send per NotifyRequest, the application sets the NRF_WAIT_REPLY bit in the nr_Flags field. This bit tells the handler not to send new NotifyMessages to a NotifyRequest's message port if the application has not returned a previous NotifyMessage. This pertains only to a specific NotifyRequest--if other NotifyRequests exist on the same file (or directory) the handler will still send NotifyMessages to the other NotifyRequest's message ports. The NRF_WAIT_REPLY bit only applies to message notification. If an application needs to know if a file or directory exists at the time the application sets up notification on that file or directory, the application can set the NRF_NOTIFY_INITIAL bit in the nr_Flags field. If the file or directory exists, the handler sends an initial message or gives an initial signal. Handlers should only perform a notification when the actual contents of the file have changed. This includes ACTION_WRITE, ACTION_SET_DATE, ACTION_DELETE, ACTION_RENAME_OBJECT, ACTION_FINDUPDATE, ACTION_FINDINPUT, and ACTION_FINDOUTPUT. It may also include other actions such as ACTION_SET_COMMENT or ACTION_SET_PROTECT, but this is not required (and may not be expected by the application as there is no need to reread the data). ACTION_REMOVE_NOTIFY 4098 EndNotify(NotifyRequest) ARG1: BPTR Pointer to previously added notify request RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE This action cancels a notification (see ACTION_ADD_NOTIFY) . ARG1 is the NotifyRequest structure used to initiate the notification. The handler should abandon any pending notification messages. Note that it is possible for a file system to receive a reply from a previously sent notification message even after the notification has been terminated. It should accept these messages silently and throw them away. @ENDNODE @NODE VolMan "Volume Manipulation/Information" The Volume Manipulation and Information actions are used to allow access to the underlying volume currently being manipulated by the file system. ACTION_CURRENT_VOLUME 7 RES1: BPTR Pointer to volume node of current volume This action returns a pointer to the volume node (from the DOS device list) associated with the file system. As the volume node may be removed from the device list when the file system mounts a different volume (such as when directed to by an ACTION_INHIBIT) there is no guarantee that this pointer will remain valid for any amount of time. This action is generally used by AmigaDOS to provide the volume line of a requester. ACTION_DISK_INFO 25 Info(...) ARG1: BPTR Pointer to an InfoData structure to fill in RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) ACTION_INFO 26 ARG1: LOCK Lock ARG2: BPTR Pointer to a InfoData Structure to fill in RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) These actions are used to get information about the device and status of the file handler. ACTION_DISK_INFO is used by the info command to report the status of the volume currently in the drive. It fills in an InfoData structure about the volume the file system currently controls. This structure should be longword aligned. ACTION_INFO fills in an InfoData structure for the volume the lock (ARG1) is on instead of the volume currently in the drive. These actions are generally expected to return DOSTRUE. The ACTION_DISK_INFO packet has a special meaning for console style handlers. When presented with this packet, a console style handler should return a pointer to the window associated with the open handle. ACTION_RENAME_DISK 9 Relabel(...) in 2.0 ARG1: BSTR New disk name RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) This action allows an application to change the name of the current volume. A file system implementing this function must also change the name stored in the volume node of the DOS device list. ACTION_FORMAT 1020 Format(fs,vol,type) ARG1: BSTR Name for volume (if supported) ARG2: LONG Type of format (file system specific) RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE This packet tells a file system to perform any device or file system specific formatting on any newly initialized media. Upon receiving this action, a file system can assume that the media has already been low level formatted and should proceed to write out any high level disk structure necessary to create an empty volume. @ENDNODE @NODE HandMain "Handler Maintenance and Control" A number of packets are defined to give an application some control over a file system: ACTION_DIE 5 RES1: BOOL DOSTRUE As its name implies, the ACTION_DIE packet tells a handler to quit. All new handlers are expected to implement this packet. Because of outstanding locks and the fact that the handler address is returned by the DeviceProc() routine, it is unlikely that the handler can disappear completely, but instead will have to release as many resources as possible and simply return an error on all packets sent to it. In the future, the system may be able to determine if there are any outstanding DeviceProc() references to a handler, and therefore make it is safe to shut down completely. ACTION_FLUSH 27 RES1: BOOL DOSTRUE This action causes the file system to flush out all buffers to disk before returning this packet. If any writes are pending, they must be processed before responding to this packet. This packet allows an application to make sure that the data that is supposed to be on the disk is actually written to the disk instead of waiting in a buffer. ACTION_MORE_CACHE 18 AddBuffers(...) in 2.0 ARG1: LONG Number of buffers to add RES1: BOOL DOSTRUE (-1L) RES2: LONG New total number of buffers This action allows an application to change the number of internal buffers used by the file system for caching. Note that a positive number increases the number of buffers while a negative number decreases the number of buffers. In all cases, the number of current buffers should be returned in RES2. This allows an application to inquire the number of buffers by sending in a value of 0 (resulting in no change). Note that the OFS and FFS in 1.3 do not accept a negative number of buffers. Note that there is a bug in the ROM file system in both Release 2.04 and Release 3.0 that jumbles its return values for this packet. The file system erroneously returns the new number of buffers in RES1 instead of RES2 (it returns a failure code in RES2). To work around this bug when using this packet, test RES1 to see if it is DOSTRUE (-1L). If it is, look at RES2 for the number of buffers, otherwise RES1 should contain the new total number of buffers. ACTION_INHIBIT 31 Inhibit(...) in 2.0 ARG1: BOOL DOSTRUE = inhibit, DOSFALSE = uninhibit RES1: BOOL Success/failure (DOSTRUE/DOSFALSE) This action is probably one of the most dangerous that a file system has to handle. When inhibited (ARG1 = DOSTRUE), the file system must not access any underlying media and return an error code on all attempts to access the device. Once uninhibited (ARG1 = DOSFALSE), the file system must assume that the medium has been changed. The file system must flush the buffers before the ACTION_INHIBIT , popping up a requester demanding that the user put back the current disk, if necessary. The handler may choose to reject an inhibit request if any objects are open for writing. Although it's not required, a handler should nest inhibits. Prior to 2.0, the system handlers did not keep a nesting count and were subject to some obscure race conditions. The 2.0 ROM filing system introduced a nesting count. ACTION_WRITE_PROTECT 1023 ARG1: BOOL DOSTRUE/DOSFALSE (write protect/un-write protect) ARG2: LONG 32 Bit pass key RES1: BOOL DOSTRUE/DOSFALSE This is a new packet defined for the Fast File System. This packet allows an application to change the write protect flag of a disk (if possible - applications cannot write to floppies that have their write-protect tabs set). This packet is primarily intended to allow write-protecting non-removable media such as hard disks. The value in ARG1 toggles the write status. The 32-bit passkey allows a program to prevent other programs from unwrite-protecting a disk. To unlock a disk, ARG2 must match the passkey of the packet that locked the disk, unless the disk was locked with a passkey of 0. In this case, no passkey is necessary to unlock the disk. ACTION_IS_FILESYSTEM 1027 IsFileSystem(devname) RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE Through this function, a handler can indicates whether or not it is a file system (whether or not it can support separate files for storing information). Programs will assume a handler can create multiple, distinct files through calls to Open() if the handler returns this packet with a DOSTRUE value. A handler does not need to support directories and subdirectories in order to qualify as a file system. It does have to support the Examine()/ExNext() calls. Note that the AmigaDOS routine IsFileSystem() will attempt to use Lock(":",SHARED_ACCESS) if this packet returns ERROR_ACTION_NOT_KNOWN. @ENDNODE @NODE HandInt "Handler Internal" There are several actions that are generally used by handlers to allow messages returning from requested services (typically an Exec device) to look like incoming request packets. This allows the handler to request an asynchronous operation but be notified of the completion. For example, a handler sends the serial.device a request for a read, but instead of sending a plain IO request, it sends a DOS packet disguised as an IO request. The serial.device treats the packet like a normal IO request, returning it to the handler when it is finished. When the handler gets back its disguised DOS packet, it knows that the read has completed. ACTION_NIL 0 Although not specifically an action, many returns look like this value because the action field has not been filled in. ACTION_READ_RETURN 1001 Generally used to indicate the completion of an asynchronous read request. ACTION_WRITE_RETURN 1002 Generally used to indicate the completion of an asynchronous write request. ACTION_TIMER 30 Used to indicate the passage of a time interval. Many handlers have a steady stream of ACTION_TIMER packets so that they can schedule house keeping and flush buffers when no activity has occurred for a given time interval. @ENDNODE @NODE Obs "Obsolete Packets" There are several packet types that are documented within the system include files that are obsolete. A file system is not expected to handle these packets and any program which sends these packets can not expect them to work: ACTION_DISK_CHANGE 33 ACTION_DISK_TYPE 32 ACTION_EVENT 6 ACTION_GET_BLOCK 2 ACTION_SET_MAP 4 Of particular note here is ACTION_DISK_CHANGE. The DiskChange command uses the ACTION_INHIBIT packet to accomplish its task. @ENDNODE @NODE ConOnly "Console Only Packets" The remaining packets are only used for console handlers and do not need to be implemented by a file system. ACTION_SCREEN_MODE 994 SetMode() in 2.0 ARG1: LONG Mode (zero or one) RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE Switch the console to and from RAW mode. An ARG1 of one indicates the unprocessed, raw mode while an ARG1 of zero indicates the processed, ``cooked'' mode. ACTION_CHANGE_SIGNAL 995 ARG1: LONG The fh_Arg1 of the console file handle ARG2: APTR MsgPort of the process to signal ARG3: LONG Reserved, currently this must be zero RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE This packet redirects what process the console handler signals when the user hits Control-C, Control-D, Control-E, or Control-F. Normally the process that opened the file handle receives the break signal. ACTION_WAIT_CHAR 20 WaitForChar() ARG1: ULONG Timeout in microseconds RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) RES2: CODE Failure code if RES1 is DOSFALSE Performs a timed read of a character. The WaitForChar() function uses this packet. ACTION_DISK_INFO 25 ARG1: BPTR Pointer to an InfoData structure to fill in RES1: BOOL Success/Failure (DOSTRUE/DOSFALSE) The ACTION_DISK_INFO packet has a special meaning for console style handlers. When presented with this packet, a console style handler should return a pointer to the window associated with the open handle in the InfoData structure's id_VolumeNode field (the InfoData structure is defined in ). Note that some consoles can return a NULL Window pointer (for example, an AUTO CON: or a AUX: console). The Amiga's standard console handler, CON:, also returns a pointer to the console handler's IO request in the id_InUse field. In some cases, the IO request's io_Unit field (which normally point to a ConUnit structure) will be NULL. See also the ACTION_DISK_INFO packet in the ``Volume Manipulation/Information'' section. @ENDNODE