page chap5 Chapter 5: File Server Interfaces
- There are several interfaces offered by the File Server, allowing it to export the files stored within the set of AFS volumes resident on its disks to the AFS community in a secure fashion and to perform self-administrative tasks. This chapter will cover the three File Server interfaces, summarized below. There is one File Server interface that will not be discussed in this document, namely that used by the Volume Server. It will be fully described in the companion AFS-3 Programmer's Reference:Volume Server/Volume Location Server Interface.
- RPC: This is the main File Server interface, supporting all of the Cache Manager's needs for providing its own clients with appropriate access to file system objects stored within AFS. It is closedly tied to the callback interface exported by the Cache Manager as described in Section 6.5, which has special implications for any application program making direct calls to this interface.
- Signals: Certain operations on a File Server must be performed by it sending unix signals on the machine on which it is executing. These operations include performing clean shutdowns and adjusting debugging output levels. Properly-authenticated administrative users do not have to be physically logged into a File Server machine to generate these signals. Rather, they may use the RPC interface exported by that machine's BOS Server process to generate them from any AFS-capable machine.
- Command Line: Many of the File Server's operating parameters may be set upon startup via its command line interface. Such choices as the number of data buffers and callback records to hold in memory may be made here, along with various other decisions such as lightweight thread stack size.
- The documentation for the AFS-3 File Server RPC interface commences with some basic definitions and data structures used in conjunction with the function calls. This is followed by an examination of the set of non-streamed RPC functions, namely those routines whose parameters are all fixed in size. Next, the streamed RPC functions, those with parameters that allow an arbitrary amount of data to be delivered, are described. A code fragment and accompanying description and analysis are offered as an example of how to use the streamed RPC calls. Finally, a description of the special requirements on any application program making direct calls to this File Server interface appears. The File Server assumes that any entity making calls to its RPC functionality is a bona fide and full-fledged Cache Manager. Thus, it expects this caller to export the Cache Manager's own RPC interface, even if the application simply uses File Server calls that don't transfer files and thus generate callbacks.
- Within those sections describing the RPC functions themselves, the purpose of each call is detailed, and the nature and use of its parameters is documented. Each of these RPC interface routines returns an integer error code, and a subset of the possible values are described. A complete and systematic list of potential error returns for each function is difficult to construct and unwieldy to examine. This is due to fact that error codes from many different packages and from many different levels may arise. Instead of attempting completeness, the error return descriptions discuss error codes generated within the functions themselves (or a very small number of code levels below them) within the File Server code itself, and not from such associated packages as the Rx, volume, and protection modules. Many of these error code are defined in the companion AFS-3 documents.
- By convention, a return value of zero reveals that the function call was successful and that all of its OUT parameters have been set by the File Server.
- The following constants and typedefs are required to properly use the File Server RPC interface, both to provide values and to interpret information returned by the calls. The constants appear first, followed by the list of typedefs, which sometimes depend on the constants above. Items are alphabetized within each group.
- All of the constants appearing below whose names contain the XSTAT string are used in conjuction with the extended data collection facility supported by the File Server. The File Server defines some number of data collections, each of which consists of an array of longword values computed by the File Server.
- There are currently two data collections defined for the File Server. The first is identified by the AFS XSTATSCOLL CALL INFO constant. This collection of longwords relates the number of times each internal function within the File Server code has been executed, thus providing profiling information. The second File Server data collection is identified by the AFS XSTATSCOLL PERF INFO constant. This set of longwords contains information related to the File Server's performance.
- Section 5.1.2.1.1 AFS DISKNAMESIZE [Value = 32]
- Specifies the maximum length for an AFS disk partition, used directly in the definition for the DiskName typedef. A DiskName appears as part of a struct ViceDisk, a group of which appear inside a struct ViceStatistics, used for carrying basic File Server statistics information.
- Section 5.1.2.1.2 AFS MAX XSTAT LONGS [Value = 1,024]
- Defines the maximum size for a File Server data collection, as exported via the RXAFS GetXStats() RPC call. It is used directly in the AFS CollData typedef.
- Section 5.1.2.1.3 AFS XSTATSCOLL CALL INFO [Value = 0]
- This constant identifies the File Server's data collection containing profiling information on the number of times each of its internal procedures has been called.
- Please note that this data collection is not supported by the File Server at this time. A request for this data collection will result the return of a zero-length array.
- Section 5.1.2.1.4 AFS XSTATSCOLL PERF INFO [Value = 1]
- This constant identifies the File Server's data collection containing performance-related information.
- Section 5.1.2.1.5 AFS CollData [typedef long AFS CollData;]
- This typedef is used by Rxgen to create a structure used to pass File Server data collections to the caller. It resolves into a C typedef statement defining a structure of the same name with the following fields:
Fields
- u int AFS CollData len - The number of longwords contained within the data pointed to by the next field.
- long *AFS CollData val - A pointer to a sequence of AFS CollData len long-words.
- Section 5.1.2.1.6 AFSBulkStats [typedef AFSFetchStatus
- AFSBulkStats<AFSCBMAX>;] This typedef is used by Rxgen to create a structure used to pass a set of statistics structures, as described in the RXAFS BulkStatus documentation in Section 5.1.3.21. It resolves into a C typedef statement defining a structure of the same name with the following fields:
Fields
- u int AFSBulkStats len - The number of struct AFSFetchStatus units contained within the data to which the next field points.
- AFSFetchStatus *AFSBulkStats val - This field houses pointer to a sequence of AFSBulkStats len units of type struct AFSFetchStatus.
- Section 5.1.2.1.7 DiskName [typedef opaque DiskName[AFS DISKNAMESIZE];]
- The name of an AFS disk partition. This object appears as a field within a struct ViceDisk,a group of which appear inside a struct ViceStatistics, used for carrying basic File Server statistics information. The term opaque appearing above inidcates that the object being defined will be treated as an undifferentiated string of bytes.
- Section 5.1.2.1.8 ViceLockType [typedef long ViceLockType;]
- This defines the format of a lock used internally by the Cache Manager. The content of these locks is accessible via the RXAFSCB GetLock() RPC function. An isomorphic and more refined version of the lock structure used by the Cache Manager, mapping directly to this definition, is struct AFSDBLockDesc, defined in Section 4.3.1.
- This structure conveys volume synchronization information across many of the File Server RPC interface calls, allowing something akin to a "whole-volume callback" on read-only volumes.
Fields
- unsigned long spare1 ... spare6 - The first longword, spare1, contains the volume's creation date. The rest are currently unused.
- This structure defines the information returned when a file system object is fetched from a File Server.
Fields
- unsigned long InterfaceVersion - RPC interface version, defined to be 1.
- unsigned long FileType - Distinguishes the object as either a file, directory, symlink, or invalid.
- unsigned long LinkCount - Number of links to this object.
- unsigned long Length - Length in bytes.
- unsigned long DataVersion - Object's data version number.
- unsigned long Author - Identity of the object's author.
- unsigned long Owner - Identity of the object's owner.
- unsigned long CallerAccess - The set of access rights computed for the caller on this object.
- unsigned long AnonymousAccess - The set of access rights computed for any completely unauthenticated principal.
- unsigned long UnixModeBits - Contents of associated unix mode bits.
- unsigned long ParentVnode - Vnode for the object's parent directory.
- unsigned long ParentUnique - Uniquifier field for the parent object.
- unsigned long SegSize - (Not implemented).
- unsigned long ClientModTime - Time when the caller last modified the data within the object.
- unsigned long ServerModTime - Time when the server last modified the data within the object.
- unsigned long Group - (Not implemented).
- unsigned long SyncCounter - (Not implemented).
- unsigned long spare1 ... spare4 - Spares.
- This structure is used to convey which of a file system object's status fields should be set, and their new values. Several File Server RPC calls, including RXAFS StoreStatus(), RXAFS CreateFile(), RXAFS SymLink(), RXAFS MakeDir(), and the streamed call to store file data onto the File Server.
Fields
- unsigned long Mask - Bit mask, specifying which of the following fields should be assigned into the File Server's status block on the object.
- unsigned long ClientModTime - The time of day that the object was last modified.
- unsigned long Owner - The principal identified as the owner of the file system object.
- unsigned long Group - (Not implemented).
- unsigned long UnixModeBits - The set of associated unix mode bits.
- unsigned long SegSize - (Not implemented).
- This structure occurs in struct ViceStatistics, and describes the characteristics and status of a disk partition used for AFS storage.
Fields
- long BlocksAvailable - Number of 1 Kbyte disk blocks still available on the partition.
- long TotalBlocks - Total number of disk blocks in the partition.
- DiskName Name - The human-readable character string name of the disk partition (e.g., /vicepa).
- This is the File Server statistics structure returned by the RXAFS GetStatistics() RPC call.
Fields
- unsigned long CurrentMsgNumber - Not used.
- unsigned long OldestMsgNumber - Not used.
- unsigned long CurrentTime - Time of day, as understood by the File Server.
- unsigned long BootTime - Kernel's boot time.
- unsigned long StartTime - Time when the File Server started up.
- long CurrentConnections - Number of connections to Cache Manager instances.
- unsigned long TotalViceCalls - Count of all calls made to the RPC interface.
- unsigned long TotalFetchs - Total number of fetch operations, either status or data, performed.
- unsigned long FetchDatas - Total number of data fetch operations exclusively.
- unsigned long FetchedBytes - Total number of bytes fetched from the File Server since it started up.
- long FetchDataRate - Result of dividing the FetchedBytes field by the number of seconds the File Server has been running.
- unsigned long TotalStores - Total number of store operations, either status or data, performed.
- unsigned long StoreDatas - Total number of data store operations exclusively.
- unsigned long StoredBytes - Total number of bytes stored to the File Server since it started up.
- long StoreDataRate - The result of dividing the StoredBytes field by the number of seconds the File Server has been running.
- unsigned long TotalRPCBytesSent - Outdated
- unsigned long TotalRPCBytesReceived - Outdated
- unsigned long TotalRPCPacketsSent - Outdated
- unsigned long TotalRPCPacketsReceived - Outdated
- unsigned long TotalRPCPacketsLost - Outdated
- unsigned long TotalRPCBogusPackets - Outdated
- long SystemCPU - Result of reading from the kernel the usage times attributed to system activities.
- long UserCPU - Result of reading from the kernel the usage times attributed to user-level activities.
- long NiceCPU - Result of reading from the kernel the usage times attributed to File Server activities that have been nice()d (i.e., run at a lower priority).
- long IdleCPU - Result of reading from the kernel the usage times attributed to idling activities.
- long TotalIO - Summary of the number of bytes read/written from the disk.
- long ActiveVM - Amount of virtual memory used by the File Server.
- long TotalVM - Total space available on disk for virtual memory activities.
- long EtherNetTotalErrors - Not used.
- long EtherNetTotalWrites - Not used.
- long EtherNetTotalInterupts - Not used.
- long EtherNetGoodReads - Not used.
- long EtherNetTotalBytesWritten - Not used.
- long EtherNetTotalBytesRead - Not used.
- long ProcessSize - The size of the File Server's data space in 1 Kbyte chunks.
- long WorkStations - The total number of client Cache Managers (workstations) for which information is held by the File Server.
- long ActiveWorkStations - The total number of client Cache Managers (workstations) that have recently interacted with the File Server. This number is strictly less than or equal to the WorkStations field.
- long Spare1 ... Spare8 - Not used.
- ViceDisk Disk1 ... Disk10 - Statistics concerning up to 10 disk partitions used by the File Server. These records keep information on all partitions, not just partitions reserved for AFS storage.
- This is the structure corresponding to the AFS XSTATSCOLL PERF INFO data collection that is defined by the File Server (see Section 5.1.2.1.4). It is accessible via the RXAFS GetXStats() interface routine, as defined in Section 5.1.3.26. The fields within this structure fall into the following classifications:
- Number of requests for the structure.
- Vnode cache information.
- Directory package numbers.
- Rx information.
- Host module fields
- Spares.
- Please note that the Rx fields represent the contents of the rx stats structure maintained by Rx RPC facility itself. Also, a full description of all the structure's fields is not possible here. For example, the reader is referred to the companion Rx document for further clarification on the Rx-related fields within afs PerfStats.
Fields
- long numPerfCalls - Number of performance collection calls received.
- long vcache L Entries - Number of entries in large (directory) vnode cache.
- long vcache L Allocs - Number of allocations for the large vnode cache.
- long vcache L Gets - Number of get operations for the large vnode cache.
- long vcache L Reads - Number of reads performed on the large vnode cache.
- long vcache L Writes - Number of writes executed on the large vnode.cache.
- long vcache S Entries - Number of entries in the small (file) vnode cache.
- long vcache S Allocs - Number of allocations for the small vnode cache.
- long vcache S Gets - Number of get operations for the small vnode cache.
- long vcache S Reads - Number of reads performed on the small vnode cache.
- long vcache S Writes - Number of writes executed on the small vnode cache.
- long vcache H Entries - Number of entries in the header of the vnode cache.
- long vcache H Gets - Number of get operations on the header of the vnode cache.
- long vcache H Replacements - Number of replacement operations on the header of the vnode cache.
- long dir Buffers - Number of directory package buffers in use.
- long dir Calls - Number of read calls made to the directory package.
- long dir IOs - Number of directory I/O operations performed.
- long rx packetRequests - Number of Rx packet allocation requests.
- long rx noPackets RcvClass - Number of failed packet reception requests.
- long rx noPackets SendClass - Number of failed packet transmission requests.
- long rx noPackets SpecialClass - Number of 'special' Rx packet rquests.
- long rx socketGreedy - Did setting the Rx socket to SO GREEDY succeed?
- long rx bogusPacketOnRead - Number of short packets received.
- long rx bogusHost - Latest host address from bogus packets.
- long rx noPacketOnRead - Number of attempts to read a packet when one was not physically available.
- long rx noPacketBuffersOnRead - Number of packets dropped due to buffer shortages.
- long rx selects - Number of selects performed, waiting for a packet arrival or a timeout.
- long rx sendSelects - Number of selects forced upon a send.
- long rx packetsRead RcvClass - Number of packets read belonging to the 'Rcv' class.
- long rx packetsRead SendClass - Number of packets read that belong to the 'Send' class.
- long rx packetsRead SpecialClass - Number of packets read belonging to the 'Special' class.
- long rx dataPacketsRead - Number of unique data packets read off the wire.
- long rx ackPacketsRead - Number of acknowledgement packets read.
- long rx dupPacketsRead - Number of duplicate data packets read.
- long rx spuriousPacketsRead - Number of inappropriate packets read.
- long rx packetsSent RcvClass - Number of packets sent belonging to the 'Rcv' class.
- long rx packetsSent SendClass - Number of packets sent belonging to the 'Send' class.
- long rx packetsSent SpecialClass - Number of packets sent belonging to the 'Special' class.
- long rx ackPacketsSent - Number of acknowledgement packets sent.
- long rx pingPacketsSent - Number of ping packets sent.
- long rx abortPacketsSent - Number of abort packets sent.
- long rx busyPacketsSent - Number of busy packets sent.
- long rx dataPacketsSent - Number of unique data packets sent.
- long rx dataPacketsReSent - Number of retransmissions sent.
- long rx dataPacketsPushed - Number of retransmissions pushed by a NACK.
- long rx ignoreAckedPacket - Number of packets whose acked flag was set at rxi Start() time.
- long rx totalRtt Sec - Total round trip time in seconds.
- long rx totalRtt Usec - Microsecond portion of the total round trip time,
- long rx minRtt Sec - Minimum round trip time in seconds.
- long rx minRtt Usec - Microsecond portion of minimal round trip time.
- long rx maxRtt Sec - Maximum round trip time in seconds.
- long rx maxRtt Usec - Microsecond portion of maximum round trip time.
- long rx nRttSamples - Number of round trip samples.
- long rx nServerConns - Total number of server connections.
- long rx nClientConns - Total number of client connections.
- long rx nPeerStructs - Total number of peer structures.
- long rx nCallStructs - Total number of call structures.
- long rx nFreeCallStructs - Total number of call structures residing on the free list.
- long host NumHostEntries - Number of host entries.
- long host HostBlocks - Number of blocks in use for host entries.
- long host NonDeletedHosts - Number of non-deleted host entries.
- long host HostsInSameNetOrSubnet - Number of host entries in the same [sub]net as the File Server.
- long host HostsInDiffSubnet - Number of host entries in a different subnet as the File Server.
- long host HostsInDiffNetwork - Number of host entries in a different network entirely as the File Server.
- long host NumClients - Number of client entries.
- long host ClientBlocks - Number of blocks in use for client entries.
- long spare[32] - Spare fields, reserved for future use.
- The results of asking the File Server for status information concerning a particular volume it hosts.
Fields
- long Vid - Volume ID.
- long ParentId - Volume ID in which the given volume is 'primarily' mounted.
- This is used to properly resolve pwd operations, as a volume may be mounted simultaneously at multiple locations.
- char Online - Is the volume currently online and fully available?
- char InService - This field records whether the volume is currently in service. It is indistinguishable from the Blessed field,
- char Blessed - See the description of the InService field immediately above.
- char NeedsSalvage -Should this volume be salvaged (run through a consistency- checking procedure)?
- long Type - The classification of this volume, namely a read/write volume (RWVOL = 0), read-only volume (ROVOL = 1), or backup volume (BACKVOL = 2).
- long MinQuota - Minimum number of 1 Kbyte disk blocks to be set aside for this volume. Note: this field is not currently set or accessed by any AFS agents.
- long MaxQuota - Maximum number of 1 Kbyte disk blocks that may be occupied by this volume.
- long BlocksInUse - Number of 1 Kbyte disk blocks currently in use by this volume.
- long PartBlocksAvail - Number of available 1 Kbyte blocks currently unused in the volume's partition.
- long PartMaxBlocks - Total number of blocks, in use or not, for the volume's partition.
- This structure is used to convey which of a file system object's status fields should be set, and their new values. The RXAFS SetVolumeStatus() RPC call is the only user of this structure.
Fields
- long Mask - Bit mask to determine which of the following two fields should be stored in the centralized status for a given volume.
- long MinQuota - Minimum number of 1 Kbyte disk blocks to be set aside for this volume.
- long MaxQuota - Maximum number of 1 Kbyte disk blocks that may be occupied by this volume.
- This field conveys information regarding a particular volume through certain File Server RPC interface calls. For information regarding the different volume types that exist, please consult the companion document, AFS-3 Programmer's Reference:Volume Server/Volume Location Server Interface.
Fields
- unsigned long Vid - Volume ID.
- long Type - Volume type (see struct AFSFetchVolumeStatus in Section 5.1.2.8 above).
- unsigned long Type0 ... Type4 - The volume IDs for the possible volume types in existance for this volume.
- unsigned long ServerCount - The number of File Server machines on which an instance of this volume is located.
- unsigned long Server0 ... Server7 - Up to 8 IP addresses of File Server machines hosting an instance on this volume. The first ServerCount of these fields hold valid server addresses.
- unsigned short Port0 ... Port7 - Up to 8 UDP port numbers on which operations on this volume should be directed. The first ServerCount of these fields hold valid port identifiers.
- The following is a description of the File Server RPC interface routines that utilize only parameters with fixed maximum lengths. The majority of the File Server calls fall into this suite, with only a handful using streaming techniques to pass objects of unbounded size between a File Server and Cache Manager.
- Each function is labeled with an opcode number. This is the low-level numerical identifier for the function, and appears in the set of network packets constructed for the RPC call.
ACL associated with the given AFS file identifier
int RXAFS FetchACL(IN struct rx connection *a rxConnP,
IN AFSFid *a dirFidP,
OUT AFSOpaque *a ACLP,
OUT AFSFetchStatus *a dirNewStatP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 131] Fetch the ACL for the directory identified by a dirFidP, placing it in the space described by the opaque structure to which a ACLP points. Also returned is the given directory's status, written to a dirNewStatP. An ACL may thus take up at most AFSOPAQUEMAX (1,024) bytes, since this is the maximum size of an AFSOpaque.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller is not permitted to perform this operation. EINVAL An internal error in looking up the client record was encountered, or an invalid fid was provided. VICETOKENDEAD Caller's authentication token has expired.
the status information regarding a given file system object
int RXAFS FetchStatus(IN struct rx connection *a rxConnP,
IN AFSFid *a fidToStatP,
OUT AFSFetchStatus *a currStatP,
OUT AFSCallBack *a callBackP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 132] Fetch the current status information for the file or directory identified by a fidToStatP, placing it into the area to which a currStatP points. If the object resides in a read/write volume, then the related callback information is returned in a callBackP.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller is not permitted to perform this operation. EINVAL An internal error in looking up the client record was encountered, or an invalid fid was provided. VICETOKENDEAD Caller's authentication token has expired.
the given ACL with the named directory
int RXAFS StoreACL(IN struct rx connection *a rxConnP,
IN AFSOpaque *a ACLToStoreP,
IN AFSFid *a dirFidP,
OUT AFSFetchStatus *a dirNewStatP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 134] Store the ACL information to which a ACLToStoreP points to the File Server, associating it with the directory identified by a dirFidP. The resulting status information for the a dirFidP directory is returned in a dirNewStatP. Note that the ACL supplied via a ACLToStoreP may be at most AFSOPAQUEMAX (1,024) bytes long, since this is the maximum size accommodated by an AFSOpaque.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller is not permitted to perform this operation.
E2BIG The given ACL is too large.
EINVAL The given ACL could not translated to its on-disk format.
the given status information for the specified file
int RXAFS StoreStatus(IN struct rx connection *a rxConnP,
IN AFSFid *a fidP,
IN AFSStoreStatus *a currStatusP,
OUT AFSFetchStatus *a srvStatusP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 135] Store the status information to which a currStatusP points, associating it with the file identified by a fidP. All outstanding callbacks on this object are broken. The resulting status structure stored at the File Server is returned in a srvStatusP.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller is not permitted to perform this operation.
EINVAL An internal error in looking up the client record was encountered, or an invalid fid was provided, or an attempt was made to change the mode of a symbolic link.
VICETOKENDEAD Caller's authentication token has expired.
the given file
int RXAFS RemoveFile(IN struct rx connection *a rxConnP,
IN AFSFid *a dirFidP,
IN char *a name<AFSNAMEMAX>,
OUT AFSFetchStatus *a srvStatusP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 136] Destroy the file named a name within the directory identified by a dirFidP. All outstanding callbacks on this object are broken. The resulting status structure stored at the File Server is returned in a srvStatusP.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller is not permitted to perform this operation.
EINVAL An internal error in looking up the client record was encountered, or an invalid fid was provided, or an attempt was made to remove "." or "..".
EISDIR The target of the deletion was supposed to be a file, but it is really a directory.
ENOENT The named file was not found.
ENOTDIR The a dirFidP parameter references an object which is not a directory, or the deletion target is supposed to be a directory but is not.
ENOTEMPTY The target directory being deleted is not empty.
VICETOKENDEAD Caller's authentication token has expired.
the given file
int RXAFS CreateFile(IN struct rx connection *a rxConnP,
IN AFSFid *DirFid,
IN char *Name,
IN AFSStoreStatus *InStatus,
OUT AFSFid *OutFid,
OUT AFSFetchStatus *OutFidStatus,
OUT AFSFetchStatus *OutDirStatus,
OUT AFSCallBack *CallBack,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 137] This call is used to create a file, but not for creating a directory or a symbolic link. If this call succeeds, it is the Cache Manager's responsibility to either create an entry locally in the directory specified by DirFid or to invalidate this directory's cache entry.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller is not permitted to perform this operation.
EINVAL An internal error in looking up the client record was encountered, or an invalid fid or name was provided.
ENOTDIR The DirFid parameter references an object which is not a directory.
VICETOKENDEAD Caller's authentication token has expired.
specified file in the given directory
int RXAFS Rename(IN struct rx connection *a rxConnP,
IN AFSFid *a origDirFidP,
IN char *a origNameP,
IN AFSFid *a newDirFidP,
IN char *a newNameP,
OUT AFSFetchStatus *a origDirStatusP,
OUT AFSFetchStatus *a newDirStatusP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 138] Rename file a origNameP in the directory identified by a origDirFidP. Its new name is to be a newNameP, and it will reside in the directory identified by a newDirFidP. Each of these names must be no more than AFSNAMEMAX (256) characters long. The status of the original and new directories after the rename operation completes are deposited in a origDirStatusP and a newDirStatusP respectively. Existing callbacks are broken for all files and directories involved in the operation.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES New file exists but user doesn't have Delete rights in the directory.
EINVAL Name provided is invalid.
EISDIR Original object is a file and new object is a directory.
ENOENT The object to be renamed doesn't exist in the parent directory.
ENOTDIR Original object is a directory and new object is a file.
EXDEV Rename attempted across a volume boundary, or create a pathname loop, or hard links exist to the file.
symbolic link
int RXAFS Symlink(IN struct rx connection *a rxConnP,
IN AFSFid *a dirFidP,
IN char *a nameP,
IN char *a linkContentsP,
IN AFSStoreStatus *a origDirStatP,
OUT AFSFid *a newFidP,
OUT AFSFetchStatus *a newFidStatP,
OUT AFSFetchStatus *a newDirStatP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 139] Create a symbolic link named a nameP in the directory identified by a dirFidP. The text of the symbolic link is provided in a linkContentsP, and the desired status fields for the symbolic link given by a origDirStatP. The name offered in a nameP must be less than AFSNAMEMAX (256) characters long, and the text of the link to which a linkContentsP points must be less than AFSPATHMAX (1,024) characters long. Once the symbolic link has been successfully created, its file identifier is returned in a newFidP. Existing callbacks to the a dirFidP directory are broken before the symbolic link creation completes. The status fields for the symbolic link itself and its parent's directory are returned in a newFidStatP and a newDirStatP respectively.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller does not have the necessary access rights.
EINVAL Illegal symbolic link name provided.
link
int RXAFS Link(IN struct rx connection *a rxConnP,
IN AFSFid *a dirFidP,
IN char *a nameP,
IN AFSFid *a existingFidP,
OUT AFSFetchStatus *a newFidStatP,
OUT AFSFetchStatus *a newDirStatP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 140] Create a hard link named a nameP in the directory identified by a dirFidP. The file serving as the basis for the hard link is identified by existingFidP. The name offered in a nameP must be less than AFSNAMEMAX (256) characters long. Existing callbacks to the a dirFidP directory are broken before the hard link creation completes. The status fields for the file itself and its parent's directory are returned in a newFidStatP and a newDirStatP respectively. Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller does not have the necessary access rights.
EISDIR An attempt was made to create a hard link to a directory.
EXDEV Hard link attempted across directories.
directory
int RXAFS MakeDir(IN struct rx connection *a rxConnP,
IN AFSFid *a parentDirFid,P
IN char *a newDirNameP,
IN AFSStoreStatus *a currStatP,
OUT AFSFid *a newDirFidP,
OUT AFSFetchStatus *a dirFidStatP,
OUT AFSFetchStatus *a parentDirStatP,
OUT AFSCallBack *a newDirCallBackP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 141] Create a directory named a newDirNameP within the directory identified by a parentDirFidP. The initial status fields for the new directory are provided in a currStatP. The new directory's name must be less than AFSNAMEMAX (256) characters long. The new directory's ACL is inherited from its parent. Existing callbacks on the parent directory are broken before the creation completes. Upon successful directory creation, the new directory's file identifier is returned in a newDirFidP, and the resulting status information for the new and parent directories are stored in a dirFidStatP and a parentDirStatP respectively. In addition, a callback for the new directory is returned in a newDirCallBackP.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller does not have the necessary access rights.
EINVAL The directory name provided is unacceptable.
directory
int RXAFS RemoveDir(IN struct rx connection *a rxConnP,
IN AFSFid *a parentDirFidP,
IN char *a dirNameP,
OUT AFSFetchStatus *a newParentDirStatP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 142] Remove the directory named a dirNameP from within its parent directory, identified by a parentDirFid. The directory being removed must be empty, and its name must be less than AFSNAMEMAX (256) characters long. Existing callbacks to the directory being removed and its parent directory are broken before the deletion completes. Upon successful deletion, the status fields for the parent directory are returned in a newParentDirStatP.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller does not have the necessary access rights.
common File Server statistics
int RXAFS GetStatistics(IN struct rx connection *a rxConnP,
OUT ViceStatistics *a FSInfoP)
- Description
- [Opcode 146] Fetch the structure containing a set of common File Server statistics. These numbers represent accumulated readings since the time the File Server last restarted. For a full description of the individual fields contained in this structure, please see Section 5.1.2.6.
- Rx connection information for the related File Server is contained in a rxConnP.
- Error Codes
- ---No error codes generated.
Ask the File Server to break the given set of callbacks on the corresponding set of file identifiers
int RXAFS GiveUpCallBacks(IN struct rx connection *a rxConnP,
IN AFSCBFids *a fidArrayP,
IN AFSCBs *a callBackArrayP)
- Description
- [Opcode 147] Given an array of up to AFSCBMAX file identifiers in a fidArrayP and a corresponding number of callback structures in a callBackArrayP, ask the File Server to remove these callbacks from its register. Note that this routine only affects callbacks outstanding on the given set of files for the host issuing the RXAFS GiveUpCallBacks call. Callback promises made to other machines on any or all of these files are not affected.
- Rx connection information for the related File Server is contained in a rxConnP.
- Error Codes
- EINVAL More file identifiers were provided in the a fidArrayP than callbacks in the a callBackArray.
information about a volume given its name
int RXAFS GetVolumeInfo(IN struct rx connection *a rxConnP,
IN char *a volNameP,
OUT VolumeInfo *a volInfoP)
- Description
- [Opcode 148] Ask the given File Server for information regarding a volume whose name is a volNameP. The volume name must be less than AFSNAMEMAX characters long, and the volume itself must reside on the File Server being probed.
- Rx connection information for the related File Server is contained in a rxConnP. Please note that definitions for the error codes with VL prefixes may be found in the vlserver.h include file
- Error Codes
- Could not contact any of the corresponding Volume Location Servers. VL BADNAME An improperly-formatted volume name provided.
VL ENTDELETED An entry was found for the volume, reporting that the volume has been deleted.
VL NOENT The given volume was not found.
Get basic status information for the named volume
int RXAFS GetVolumeStatus(IN struct rx connection *a rxConnP,
IN long a volIDP,
OUT AFSFetchVolumeStatus *a volFetchStatP,
OUT char *a volNameP,
OUT char *a offLineMsgP,
OUT char *a motdP)
- Description
- [Opcode 149] Given the numeric volume identifier contained in a volIDP, fetch the basic status information corresponding to that volume. This status information is stored into a volFetchStatP. A full description of this status structure is found in Section 5.1.2.8. In addition, three other facts about the volume are returned. The volume's character string name is placed into a volNameP. This name is guaranteed to be less than AFSNAMEMAX characters long. The volume's offline message, namely the string recording why the volume is off-line (if it is), is stored in a offLineMsgP . Finally, the volume's "Message of the Day" is placed in a motdP. Each of the character strings deposited into a offLineMsgP and a motdP is guaranteed to be less than AFSOPAQUEMAX (1,024) characters long.
- Rx connection information for the related File Server is contained in a rxConnP.
- Error Codes
- EACCES The caller does not have the necessary access rights.
EINVAL A volume identifier of zero was specified.
Set the basic status information for the named volume
int RXAFS SetVolumeStatus(struct rx connection *a rxConnP,
long avolIDP,
AFSStoreVolumeStatus *a volStoreStatP,
char *a volNameP,
char *a offLineMsgP,
char *a motdP)
- Description
- [Opcode 150] Given the numeric volume identifier contained in a volIDP, set that volume's basic status information to the values contained in a volStoreStatP. A full description of the fields settable by this call, including the necessary masking, is found in Section 5.1.2.9. In addition, three other items relating to the volume may be set. Non-null character strings found in a volNameP, a offLineMsgP, and a motdP will be stored in the volume's printable name, off-line message, and "Message of the Day" fields respectively. The volume name provided must be less than AFSNAMEMAX (256) characters long, and the other two strings must be less than AFSOPAQUEMAX (1,024) characters long each.
- Rx connection information for the related File Server is contained in a rxConnP.
- Error Codes
- EACCES The caller does not have the necessary access rights.
EINVAL A volume identifier of zero was specified.
Return the name of the root volume for the file system
int RXAFS GetRootVolume(IN struct rx connection *a rxConnP,
OUT char *a rootVolNameP)
- Description
- [Opcode 151] Fetch the name of the volume which serves as the root of the AFS file system and place it into a rootVolNameP. This name will always be less than AFSNAMEMAX characters long. Any File Server will respond to this call, not just the one hosting the root volume. The queried File Server first tries to discover the name of the root volume by reading from the /usr/afs/etc/RootVolume file on its local disks. If that file doesn't exist, then it will return the default value, namely "root.afs".
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- ---No error codes generated.
Check that the given user identifier matches the one in the supplied authentication token
int RXAFS CheckToken(IN struct rx connection *a rxConnP,
IN long ViceId,
IN AFSOpaque *token)
- Description
- [Opcode 152] This function only works for the now-obsolete RPC facility used by AFS, R. For modern systems using the Rx RPC mechanism, we always get an error return from this routine.
- Rx connection information for the related File Server is contained in a rxConnP.
- Error Codes
- ECONNREFUSED Always returned on Rx connections.
File Server's time of day
int RXAFS GetTime(IN struct rx connection *a rxConnP,
OUT unsigned long *a secondsP,
OUT unsigned long *a uSecondsP)
- Description
- [Opcode 153] Get the current time of day from the File Server specified in the Rx connection information contained in a rxConnP. The time is returned in elapsed seconds (a secondsP) and microseconds (a uSecondsP) since that standard unix "start of the world".
- Error Codes
- ---No error codes generated.
information about a volume given its name
int RXAFS NGetVolumeInfo(IN struct rx connection *a rxConnP,
IN char *a volNameP,
OUT AFSVolumeInfo *a volInfoP)
- Description
- [Opcode 154] This function is identical to RXAFS GetVolumeInfo() (see Section 5.1.3.14), except that it returns a struct AFSVolumeInfo instead of a struct VolumeInfo. The basic difference is that struct AFSVolumeInfo also carries an accompanying UDP port value for each File Server listed in the record.
the status information regarding a set of given file system objects
int RXAFS BulkStatus(IN struct rx connection *a rxConnP,
IN AFSCBFids *a fidToStatArrayP,
OUT AFSBulkStats *a currStatArrayP,
OUT AFSCBs *a callBackArrayP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 155] This routine is identical to RXAFS FetchStatus() as described in Section 5.1.3.2, except for the fact that it allows the caller to ask for the current status fields for a set of up to AFSCBMAX (50) file identifiers at once.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller does not have the necessary access rights.
EINVAL The number of file descriptors for which status information was requested is illegal.
advisory lock on the given file identifier
int RXAFS SetLock(IN struct rx connection *a rxConnP,
IN AFSFid *a fidToLockP,
IN ViceLockType a lockType,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 156] Set an advisory lock on the file identified by a fidToLockP. There are two types of locks that may be specified via a lockType: LockRead and LockWrite. An advisory lock times out after AFS LOCKWAIT (5) minutes, and must be extended in order to stay in force (see RXAFS ExtendLock(), Section 5.1.3.23).
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller does not have the necessary access rights.
EINVAL An illegal lock type was specified.
EWOULDBLOCK The lock was already incompatibly granted to another party.
an advisory lock on a file
int RXAFS ExtendLock(IN struct rx connection *a rxConnP,
IN AFSFid *a fidToBeExtendedP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 157] Extend the advisory lock that has already been granted to the caller on the file identified by a fidToBeExtendedP.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EINVAL The caller does not already have the given file locked.
Release the advisory lock on a file
int RXAFS ReleaseLock(IN struct rx connection *a rxConnP,
IN AFSFid *a fidToUnlockP,
OUT AFSVolSync *a volSyncP)
- Description
- [Opcode 158] Release the advisory lock held on the file identified by a fidToUnlockP. If this was the last lock on this file, the File Server will break all existing callbacks to this file.
- Rx connection information for the related File Server is contained in a rxConnP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller does not have the necessary access rights.
the version number associated with the File Server's extended statistics structure
int RXAFS XStatsVersion(IN struct rx connection *a rxConnP,
OUT long *a versionNumberP)
- Description
- [Opcode 159] This call asks the File Server for the current version number of the extended statistics structures it exports (see RXAFS GetXStats(), Section 5.1.3.26). The version number is placed into a versionNumberP.
- Rx connection information for the related File Server is contained in a rxConnP.
- Error Codes
- ---No error codes generated.
current contents of the specified extended statistics structure
int RXAFS GetXStats(IN struct rx connection *a rxConnP,
IN long a clientVersionNumber,
IN long a collectionNumber,
OUT long *a srvVersionNumberP,
OUT long *a timeP,
OUT AFS CollData *a dataP)
- Description
- [Opcode 160] This function fetches the contents of the specified File Server extended statistics structure. The caller provides the version number of the data it expects to receive in a clientVersionNumber. Also provided in a collectionNumber is the numerical identifier for the desired data collection. There are currently two of these data collections defined: AFS XSTATSCOLL CALL INFO, which is the list of tallies of the number of invocations of internal File Server procedure calls, and AFS XSTATSCOLL PERF INFO, which is a list of performance-related numbers. The precise contents of these collections are described in Sections 5.1.2.7. The current version number of the File Server collections is returned in a srvVersionNumberP, and is always set upon return, even if the caller has asked for a difierent version. If the correct version number has been specified, and a supported collection number given, then the collection data is returned in a dataP. The time of collection is also returned, being placed in a timeP.
- Rx connection information for the related File Server is contained in a rxConnP.
- Error Codes
- ---No error codes are generated.
- There are two streamed functions in the File Server RPC interface, used to fetch and store arbitrary amounts of data from a file. While some non-streamed calls pass such variable-length objects as struct AFSCBFids, these objects have a pre-determined maximum size.
- The two streamed RPC functions are also distinctive in that their single Rxgen declarations generate not one but two client-side stub routines. The first is used to ship the IN parameters off to the designated File Server, and the second to gather the OUT parameters and the error code. If a streamed definition declares a routine named X YZ(), the two resulting stubs will be named StartX YZ() and EndX YZ(). It is the application programmer's job to first invoke StartX YZ(), then manage the unbounded data transfer, then finish up by calling EndX YZ(). The first longword in the unbounded data stream being fetched from a File Server contains the number of data bytes to follow. The application then reads the specified number of bytes from the stream.
- The following sections describe the four client-side functions resulting from the Fetch-Data() and StoreData() declarations in the Rxgen interface definition file. These are the actual routines the application programmer will include in the client code. For reference, here are the interface definitions that generate these functions. Note that the split keyword is what causes Rxgen to generate the separate start and end routines. In each case, the number after the equal sign specifies the function's identifying opcode number. The opcode is passed to the File Server by the StartRXAFS FetchData() and StartRXAFS StoreData() stub routines.
FetchData(IN AFSFid *a_fidToFetchP,
IN long a_offset,
IN long a_lenInBytes,
OUT AFSFetchStatus *a_fidStatP,
OUT AFSCallBack *a_callBackP,
OUT AFSVolSync *a_volSyncP) split = 130;
StoreData(IN AFSFid *Fid,
IN AFSStoreStatus *InStatus,
IN long Pos,
IN long Length,
IN long FileLength,
OUT AFSFetchStatus *OutStatus,
OUT AFSVolSync *a_volSyncP) split = 133;
a request to fetch file data
int StartRXAFS FetchData(IN struct rx call *a rxCallP,
IN AFSFid *a fidToFetchP,
IN long a offset,
IN long a lenInBytes)
- Description
- Begin a request for a lenInBytes bytes of data starting at byte offset a offset from the file identified by a fidToFetchP. After successful completion of this call, the data stream will make the desired bytes accessible. The first longword in the stream contains the number of bytes to actually follow.
- Rx call information to the related File Server is contained in a rxCallP.
- Error Codes
- ---No error codes generated.
Conclude a request to fetch file data
int EndRXAFS FetchData(IN struct rx call *a rxCallP,
OUT AFSFetchStatus *a fidStatP,
OUT AFSCallBack *a callBackP,
OUT AFSVolSync *a volSyncP)
- Description
- Conclude a request to fetch file data, as commenced by an StartRXAFS FetchData() invocation. By the time this routine has been called, all of the desired data has been read off the data stream. The status fields for the file from which the data was read are stored in a fidStatP. If the file was from a read/write volume, its callback information is placed in a callBackP.
- Rx call information to the related File Server is contained in a rxCallP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller does not have the necessary access rights. EIO Given file could not be opened or statted on the File Server, or there was an error reading the given data off the File Server's disk.
-31 An Rx write into the stream ended prematurely.
a request to store file data
int StartRXAFS StoreData(IN struct rx call *a rxCallP,
IN AFSFid *a fidToStoreP,
IN reStatus *a fidStatusP,
IN AFSStolong a offset,
IN long a lenInBytes,
IN long a fileLenInBytes)
- Description
- Begin a request to write a lenInBytes of data starting at byte offset a offset to the file identified by a fidToStoreP, causing that file's length to become a fileLenInBytes bytes. After successful completion of this call, the data stream will be ready to begin accepting the actual data being written.
- Rx call information to the related File Server is contained in a rxCallP.
- Error Codes
- ---No error codes generated.
Conclude a request to store file data
int EndRXAFS StoreData(IN struct rx call *a rxCallP,
OUT AFSFetchStatus *a fidStatP,
OUT AFSCallBack *a callBackP,
OUT AFSVolSync *a volSyncP)
- Description
- Conclude a request to store file data, as commenced by a StartRXAFS StoreData() invocation. By the time this routine has been called, all of the file data has been inserted into the data stream. The status fields for the file to which the data was written are stored in a fidStatP. All existing callbacks to the given file are broken before the store concludes.
- Rx call information to the related File Server is contained in a rxCallP. Volume version information is returned for synchronization purposes in a volSyncP.
- Error Codes
- EACCES The caller does not have the necessary access rights.
EISDIR The file being written to is a symbolic link.
ENOSPEC A write to the File Server's file on local disk failed.
-32 A short read was encountered by the File Server on the data stream.
Usage
- The following code fragment is offered as an example of how to use the streamed File Server RPC calls. In this case, a client fetches some amount of data from the given File Server and writes it to a local file it uses to cache the information. For simplicity, many issues faced by a true application programmer are not addressed here. These issues include locking, managing file chunking, data version number mismatches, volume location, Rx connection management, defensive programming (e.g., checking parameters before using them), client-side cache management algorithms, callback management, and full error detection and recovery. Pseudocode is incorporated when appropriate to keep the level of detail reasonable. For further descriptions of some of these details and issues, the reader is referred to such companion documents as AFS-3 Programmer's Reference: Specification for the Rx Remote Procedure Call Facility, AFS-3 Programmer's Reference:Volume Server/Volume Location Server Interface, and AFS-3 Programmer's Reference: Architectural Overview.
- A discussion of the methods used within the example code fragment follows immediately afterwards in Section 5.1.5.3.
Fetch Operation
int code;
long bytesRead;
struct myConnInfo *connP;
struct rx_call *rxCallP;
struct AFSFid *afsFidP;
int lclFid;
long offsetBytes;
long bytesToFetch;
long bytesFromFS;
char *fetchBuffP;
int currReadBytes;
. . .
rxCallP = rx_NewCall(connP->rxConnP);
code = StartRXAFS_FetchData( rxCallP,
afsFidP,
offsetBytes,
bytesToFetch);
if (code == 0)
{
bytesRead = rx_Read(rxCallP, &bytesFromFS, sizeof(long));
if (bytesRead != sizeof(long)) ExitWithError(SHORT_RX_READ);
bytesFromFS = ntohl(bytesFromFS);
xmitBuffer = malloc(FETCH_BUFF_BYTES);
lclFid = open(CacheFileName, O_RDWR, mode);
pos = lseek(lclFid, offsetBytes, L_SET);
while (bytesToFetch > 0) {
currReadBytes = (bytesToFetch > FETCH_BUFF_BYTES) ?
FETCH_BUFF_BYTES : bytesToFetch;
bytesRead = rx_Read(rxCallP, fetchBuffP, currReadBytes);
if (bytesRead != currReadBytes) ExitWithError(SHORT_RX_READ);
code = write(lclFid, fetchBuffP, currReadBytes);
if (code) ExitWithError(LCL_WRITE_FAILED);
bytesToFetch -= bytesRead;
}
close(lclFid);
} else ExitWithError(code);
code = EndRXAFS_FetchData( rxCallP,
fidStatP,
fidCallBackP,
volSynchP);
code = rx_EndCall(rxCallP, code);
return(code); . . .
- The opening assumption in this discussion is that all the information required to do the fetch has already been set up. These mandatory variables are the client-side connection information for the File Server hosting the desired file, the corresponding AFS file identifier, the byte offset into the file, the number of bytes to fetch, and the identifier for the local file serving as a cached copy.
- Given the Rx connection information stored in the client's connP record, rx NewCall() is used to create a new Rx call to handle this fetch operation. The structure containing this call handle is placed into rxCallP. This call handle is used immediately in the invocation of StartRXAFS FetchData(). If this setup call fails, the fragment exits. Upon success, though, the File Server will commence writing the desired data into the Rx data stream. The File Server first writes a single longword onto the stream announcing to the client how many bytes of data will actually follow. The fragment reads this number with its first rx Read() call. Since all Rx stream data is written in network byte order, the fragment translates the byte count to its own host byte order first to properly interpret it. Once the number of bytes to appear on the stream is known, the client code proceeds to open the appropriate cache file on its own local disk and seeks to the appropriate spot within it. A buffer into which the stream data will be placed is also created at this time.
- The example code then falls into a loop where it reads all of the data from the File Server and stores it in the corresponding place in the local cache file. For each iteration, the code decides whether to read a full buffer's worth or the remaining number of bytes, whichever is smaller. After all the data is pulled off the Rx stream, the local cache file is closed. At this point, the example finishes off the RPC by calling EndRXAFS FetchData(). This gathers in the required set of OUT parameters, namely the status fields for the file just fetched, callback and volume synchronization information, and the overall error code for the streamed routine. The Rx call created to perform the fetch is then terminated and cleaned up by invoking rx EndCall().
- The AFS File Server RPC interface was originally designed to interact only with Cache Manager agents, and thus made some assumptions about its callers. In particular, the File Server expected that the agents calling it would potentially have stored callback state on file system objects, and would have to be periodically pinged in order to garbage-collect its records, removing information on dead client machines. Thus, any entity making direct calls to this interface must mimic certain Cache Manager actions, and respond to certain Cache Manager RPC interface calls.
- To be safe, any application calling the File Server RPC interface directly should export the entire Cache Manager RPC interface. Realistically, though, it will only need to provide stubs for the three calls from this interface that File Servers know how to make: RXAFSCB InitCallBackState(), RXAFSCB Probe() and RXAFSCB CallBack(). The very first File Server call made by this application will prompt the given File Server to call RXAFSCB InitCallBackState(). This informs the application that the File Server has no record of its existence and hence this "Cache Manager" should clear all callback information for that server. Once the application responds positively to the inital RXAFSCB InitCallBackState(), the File Server will treat it as a bona fide, fully-fledged Cache Manager, and probe it every so often with RXAFSCB Probe() calls to make sure it is still alive.
- While the majority of communication with AFS File Servers occurs over the RPC interface, some important operations are invoked by sending unix signals to the process. This section describes the set of signals recognized by the File Server and the actions they trigger upon receipt, as summarized below:
- SIGQUIT: Shut down a File Server.
- SIGTSTP: Upgrade debugging output level.
- SIGHUP: Reset debugging output level.
- SIGTERM: Generate debugging output specifically concerning open files within the File Server process.
- Upon receipt of this signal, the File Server shuts itself down in an orderly fashion. It first writes a message to the console and to its log file (/usr/afs/logs/FileLog) stating that a shutdown has commenced. The File Server then flushes all modified buffers and prints out a set of internal statistics, including cache and disk numbers. Finally, each attached volume is taken offline, which means the volume header is written to disk with the appropriate bits set.
- In typical usage, human operators do not send the SIGQUIT signal directly to the File Server in order to affect an orderly shutdown. Rather, the BOS Server managing the server processes on that machine issues the signal upon receipt of a properly-authorized shutdown RPC request.
- Arrival of a SIGTSTP signal results in an increase of the debugging level used by the File Server. The routines used for writing to log files are sensitive to this debugging level, as recorded in the global LogLevel variable. Specifically, these routines will only generate output if the value of LogLevel is greater than or equal to the value of its threshold parameter. By default, the File Server sets LogLevel to zero upon startup. If a SIGTSTP signal is received when the debugging level is zero, it will be bumped to 1. If the signal arrives when the debugging level is positive, its value will be multiplied by 5. Thus, as more SIGTSTPs are received, the set of debugging messages eligible to be delivered to log files grows.
- Since the SIGTSTP signal is not supported under IBM's AIX 2.2.1 operating system, this form of debugging output manipulation is not possible on those platforms.
- Receiving a SIGHUP signal causes a File Server to reset its debugging level to zero. This effectively reduces the set of debugging messages eligible for delivery to log files to a bare minimum. This signal is used in conjunction with SIGTSTP to manage the verbosity of log information.
- Since the SIGHUP signal is not supported under IBM's AIX 2.2.1 operating system, this form of debugging output manipulation is not possible on those platforms.
- Receipt of a SIGTERM signal triggers a routine which sweeps through the given File Server's unix file descriptors. For each possible unix fid slot, an fstat() is performed on that descriptor, and the particulars of each open file are printed out. This action is designed solely for debugging purposes.
- Another interface exported by the File Server is the set of command line switches it accepts. Using these switches, many server parameters and actions can be set. Under normal conditions, the File Server process is started up by the BOS Server on that machine, as described in AFS-3 Programmer's Reference: BOS Server Interface. So, in order to utilize any combination of these command-line options, the system administrator must define the File Server bnode in such a way that these parameters are properly included. Note that the switch names must be typed exactly as listed, and that abbreviations are not allowed. Thus, specifying -b 300 on the command line is unambiguous, directing that 300 buffers are to be allocated. It is not an abbreviation for the -banner switch, asking that a message is to be printed to the console periodically.
- A description of the set of currently-supported command line switches follows.
- -b <# buffers> Choose the number of 2,048-byte data buffers to allocate at system startup. If this switch is not provided, the File Server will operate with 70 such buffers by default.
- -banner This switch instructs the File Server to print messages to the console every 10 minutes to demonstrate it is still running correctly. The text of the printed message is: File Server is running at <time>.
- -cb <# callbacks stored> Specify the maximum number of callback records stored simultaneously by the File Server. The default pool size is 20,000 records.
- -d <debug level>=""> Set the debugging output level at which File Server runs to the value provided. Specifically, the LogLevel global variable is set to the given value (See Section 5.2.2). If this switch is not provided, the default initial File Server debugging level is set to zero, producing the minimal debugging output to the log files.
- -k <stack size>=""> Set the stack size to provide server LWPs upon creation, measured in 1,024-byte blocks. The default LWP stack size is 24 blocks, or 24,576 bytes.
- -l <large (directory) vnodes> Select the number of "large" vnodes the File Server will cache. These vnodes are suitable for recording information about AFS directories. The extra space in the vnode allows ACL information to be stored along with the directory. The default allocation value is 200 directory vnodes.
- -pctspare <percent overrun="" blocks="" past="" quota>=""> Similar to the -spare switch, except that the number of allowable overrun blocks is expressed as a percentage of the given volume's quota. Note: this switch cannot be used in combination with the -spare switch.
- -rxdbg Instruct the File Server to open a file named rx dbg in the current directory, into which the Rx package will write general debugging information. If the file is already open (due to the appearance of the -rxdbge switch earlier in the command line), this results in a no-op.
- -rxdbge Instruct the File Server to open a file named rx dbg in the current directory, into which the Rx package will write debugging information related to its event-scheduling activities. If the file is already open (due to the appearance of the -rxdbg switch earlier in the command line), this results in a no-op.
- -rxpck <# packets> Set the number of extra Rx packet buffers to hold in reserve. These pre-allocated buffers assist in responding to spikes in network traffic demands. By default, 100 such packet buffers are maintained.
- -s <small (file) vnodes> Select the number of "small" vnodes the File Server will cache. These vnodes are suitable for recording information about non-directory files. As with directory vnodes, the File Server will allocate 200 small vnodes by default.
- -spare <# overrun blocks to allow> Tell the File Server to allow users performing a store operation to overrun the host volume's disk quota by a certain number of (1,024-byte) blocks. In other words, the first store resulting in a quota overrun will be allowed to succeed if and only if it uses no more than these many blocks beyond the quota. Further store operations will be rejected until the volume's storage is once again reduced below quota. By default, overruns of 1,024 blocks of 1,024 bytes each (1 megabyte total) are tolerated. Note: this switch cannot be used in combination with the -pctspare switch.
- -w <callback wait="" interval="" in="" seconds>=""> This switch determines how often the File Server periodic daemon lightweight processes run. Among other things, these daemon LWPs check on the validity of callback records, keep disk usage statistics up to date, and check the health of the various client machines that have previously interacted with the File Server. For a full description of these daemon LWPs, consult Section 2.3. The associated argument specifies the number of seconds to sleep between daemon invocations. By default, these periodic daemons run every 300 seconds (5 minutes).
- There are several interfaces offered by the Cache Manager, allowing clients to access the files stored by the community of AFS File Servers, to configure the Cache Manager's behavior and resources, to store and retrieve authentication information, to specify the location of community Authentication Server and Volume Location Server services, and to observe and debug the Cache Manager's state and actions. This chapter will cover the following five interfaces to the Cache Manager:
- ioctl(): The standard unix ioctl() system call has been extended to include more operations, namely waiting until data stores to a File Server complete before returning to the caller (VIOCCLOSEWAIT) and getting the name of the cell in which an open file resides (VIOCIGETCELL).
- pioctl(): An additional system call is provided through which applications can access operations specific to AFS, which are often tied to a particular pathname. These operations include Access Control List (ACL) and mount point management, Kerberos ticket management, cache configuration, cell configuration, and status of File Servers.
- RPC: Interface by which outside servers and investigators can manipulate the Cache Manager. There are two main categories of routines: callback management, typically called by the File Server, and debugging/statistics, called by programs such as cmdebug and via the xstat user-level library for collection of extended statistics.
- Files: Much of the Cache Manager's configuration information, as well as its view of the AFS services available from the outside world, is obtained from parsing various files. One set of these files is typically located in /usr/vice/etc, and includes CellServDB, ThisCell, and cacheinfo. Another set is usually found in /usr/vice/cache, namely CacheItems, VolumeItems, and AFSLog.
- Mariner: This is the interface by which file transfer activity between the Cache Manager and File Servers may be monitored. Specifically, it is used to monitor the names of the files and directories being fetched and/or stored over the network.
- Another important component not described in this document is the afsd program. It is afsd's job to initialize the Cache Manager on a given machine and to start up its related daemon threads. It accepts a host of configuration decisions via its command-line interface. In addition, it parses some of the information kept in the configuration files mentioned above and passes that information to the Cache Manager. The reader may find a full description of afsd in the AFS 3.0 Command Reference Manual[2].
- This section defines data structures that are used by the pioctl() calls.
- The Cache Manager is the sole active AFS agent aware of the cellular architecture of the system. Since AFS file identifiers are not guaranteed to be unique across cell boundaries, it must further qualify them for its own internal bookkeeping. The struct VenusFid provides just such additional qualification, attaching the Cache Manager's internal cell identifier to the standard AFS fid.
Fields
- long Cell - The internal identifier for the cell in which the file resides.
- struct ViceFid Fid - The AFS file identifier within the above cell.
- This is the clear-text version of an AFS token of identity. Its fields are encrypted into the secret token format, and are made easily available to the Cache Manager in this structure.
Fields
- long AuthHandle - Key version number.
- char HandShakeKey[8] - Session key.
- long ViceId - Identifier for the AFS principal represented by this token.
- long BeginTimestamp - Timestamp of when this token was minted, and hence came into effect.
- long EndTimestamp - Timestamp of when this token is considered to be expired, and thus disregarded.
- The standard unix ioctl() system call performs operations on file system objects referenced with an open file descriptor. AFS has augmented this system call with two additional operations, one to perform "safe stores", and one to get the name of the cell in which a file resides. A third ioctl() extension is now obsolete, namely aborting a store operation currently in progress.
- [Opcode 1] Normally, a client performing a unix close() call on an AFS file resumes once the store operation on the given file data to the host File Server has commenced but before it has completed. Thus, it is possible that the store could actually fail (say, because of network partition or server crashes) without the client's knowledge. This new ioctl opcode specifies to the Cache Manager that all future close() operations will wait until the associated store operation to the File Server has completed fully before returning.
- [Opcode 2] This ioctl() extension is now obsolete. This call results in a noop. The original intention of this call was to allow a store operation currently in progress to a File Server on the named fid to be aborted.
- [Opcode 3] Get the name of the cell in which the given fid resides. If the file is not an AFS file, then ENOTTY is returned. The output buffer specified in the data area must be large enough to hold the null-terminated string representing the file's cell, otherwise EFAULT is returned. However, an out size value of zero specifies that the cell name is not to be copied into the output buffer. In this case, the caller is simply interested in whether the file is in AFS, and not its exact cell of residence.
- There is a new unix system call, pioctl(), which has been defined especially to support the AFS Cache Manager. Its functional definition is as follows:
int afs syscall pioctl(IN char *a pathP,
IN int a opcode,
IN struct ViceIoctl *a paramsP,
IN int a followSymLinks)
- This new call is much like the standard ioctl() call, but differs in that the affected file (when applicable) is specified by its path, not by a file descriptor. Another difference is the fourth parameter, a followSymLinks, determines which file should be used should a pathP be a symbolic link. If a followSymLinks be set to 1, then the symbolic link is followed to its target, and the pioctl() is applied to that resulting file. If a followSymLinks is set to 0, then the pioctl() applies to the symbolic link itself.
- Not all pioctl() calls affect files. In those cases, the a pathP parameter should be set to a null pointer. The second parameter to pioctl(), a opcode, specifies which operation is to be performed. The opcode for each of these operations is included in the text of the description. Note that not all pioctl() opcodes are in use. These unused values correspond to obsolete operations.
- The descriptions that follow identify some of the possible error codes for each pioctl() opcode, but do not offer a comprehensive lists. All pioctl() calls return 0 upon success.
- The rest of this section proceeds to describe the individual opcodes available. First, though, one asymmetry in this opcode set is pointed out, namely that while various operations are defined on AFS mount points, there is no direct way to create a mount point.
- This documentation partitions the pioctl() into several groups:
- Volume operations
- File Server operations
- Cell Operations
- Authentication Operations
- ACL Operations
- Cache operations
- Miscellaneous operations
- For all pioctl()s, the fields within the a paramsP parameter will be referred to directly. Thus, the values of in, in size, out, and out size are discussed, rather than the settings for a paramsP->in, a paramsP->in size, a paramsP->out, and a paramsP->out size.
- For convenience of reference, a list of the actively-supported pioctl()s, their opcodes, and brief description appears (in opcode order) below.
- [1] VIOCSETAL : Set the ACL on a directory
- [2] VIOCGETAL : Get the ACL for a directory
- [3] VIOCSETTOK : Set the caller's token for a cell
- [4] VIOCGETVOLSTAT : Get volume status
- [5] VIOCSETVOLSTAT : Set volume status
- [6] VIOCFLUSH : Flush an object from the cache
- [8] VIOCGETTOK : Get the caller's token for a cell
- [9] VIOCUNLOG : Discard authentication information
- [10] VIOCCKSERV : Check the status of one or more File Servers
- [11] VIOCCKBACK : Mark cached volume info as stale
- [12] VIOCCKCONN : Check caller's tokens/connections
- [14] VIOCWHEREIS : Find host(s) for a volume
- [20] VIOCACCESS : Check caller's access on object
- [21] VIOCUNPAG : See [9] VIOCUNLOG
- [22] VIOCGETFID : Get fid for named object
- [24] VIOCSETCACHESIZE : Set maximum cache size in blocks
- [25] VIOCFLUSHCB : Unilaterally drop a callback
- [26] VIOCNEWCELL : Set cell service information
- [27] VIOCGETCELL : Get cell configuration entry
- [28] VIOCAFS DELETE MT PT : Delete a mount point
- [29] VIOC AFS STAT MT PT : Get the contents of a mount point
- [30] VIOC FILE CELL NAME : Get cell hosting a given object
- [31] VIOC GET WS CELL : Get caller's home cell name
- [32] VIOC AFS MARINER HOST : Get/set file transfer monitoring output
- [33] VIOC GET PRIMARY CELL : Get the caller's primary cell
- [34] VIOC VENUSLOG : Enable/disable Cache Manager logging
- [35] VIOC GETCELLSTATUS : Get status info for a cell entry
- [36] VIOC SETCELLSTATUS : Set status info for a cell entry
- [37] VIOC FLUSHVOLUME : Flush cached data from a volume
- [38] VIOC AFS SYSNAME : Get/set the mapping
- [39] VIOC EXPORTAFS : Enable/disable NFS/AFS translation
- [40] VIOCGETCACHEPARAMS : Get current cache parameter values
- There is an irregularity which deserves to be mentioned regarding the pioctl() interface. There are pioctl() operations for getting information about a mount point (VIOC AFS STAT MT PT) and for deleting a mount point (VIOC AFS DELETE MT PT), but no operation for creating mount points. To create a mount point, a symbolic link obeying a particular format must be created. The first character must be either a "%" or a "#", depending on the type of mount point being created (see the discussion in Section 6.4.4.4). If the mount point carries the name of the cell explicitly, the full cell name will appear next, followed by a colon. In all cases, the next portion of the mount point is the volume name. By convention, the last character of a mount point must always be a period ("."). This trailing period is not visible in the output from fs lsmount.
- There are several pioctl() opcodes dealing with AFS volumes. It is possible to get and set volume information (VIOCGETVOLSTAT, VIOCSETVOLSTAT), discover which volume hosts a particular file system object (VIOCWHEREIS), remove all objects cached from a given volume (VIOC FLUSHVOLUME), and revalidate cached volume information (VIOCCKBACK).
status for pathname
- [Opcode 4] Fetch information concerning the volume that contains the file system object named by a pathP. There is no other input for this call, so in size should be set to zero. The status information is placed into the buffer named by out, if out size is set to a value of sizeof(struct VolumeStatus) or larger. Included in the volume information are the volume's ID, quota, and number of blocks used in the volume as well as the disk partition on which it resides. Internally, the Cache Manager calls the RXAFS GetVolumeInfo() RPC (See Section 5.1.3.14) to fetch the volume status.
- Among the possible error returns, EINVAL indicates that the object named by a pathP could not be found.
status for pathname
- [Opcode 5] Set the status fields for the volume hosting the file system object named by a pathP. The first object placed into the input buffer in is the new status image. Only those fields that may change, namely MinQuota and MaxQuota fields, are interpreted upon receipt by the File Server, and are set to the desired values. Immediately after the struct VolumeStatus image, the caller must place the null-terminated string name of the volume involved in the input buffer. New settings for the offline message and MOTD (Message of the Day) strings may appear after the volume name. If there are no changes in the offline and/or MOTD messages, a null string must appear for that item. The in size parameter must be set to the total number of bytes so inserted, including the nulls after each string. Internally, the Cache Manager calls the RXAFS SetVolumeStatus() RPC (See Section 5.1.3.16) to store the new volume status.
- Among the possible error returns, EINVAL indicates that the object named by a pathP could not be found.
server(s) hosting the pathname's volume
- [Opcode 14] Find the set of machines that host the volume in which the file system object named by a pathP resides. The input buffer in is not used by this call, so in size should be set to zero. The output buffer indicated by out is filled with up to 8 IP addresses, one for each File Server hosting the indicated volume. Thus, out size should be set to at least (8*sizeof(long)). This group of hosts is terminated by the first zeroed IP address that appears in the list, but under no circumstances are more than 8 host IP addresses returned.
- Among the possible error returns is EINVAL, indicating that the pathname is not in AFS, hence is not contained within a volume. If ENODEV is returned, the associated volume information could not be obtained.
data cached from the pathname's volume
- [Opcode 37] Determine the volume in which the file system object named by a pathP resides, and then throw away all currently cached copies of files that the Cache Manager has obtained from that volume. This call is typically used should a user suspect there is some cache corruption associated with the files from a given volume.
of all cached volume information
- [Opcode 11] Ask the Cache Manager to check the validity of all cached volume information. None of the call's parameters are referenced in this call, so a pathP and in should be set to the null pointer, and in size and out size should be set to zero.
- This operation is performed in two steps:
- 1 The Cache Manager first refreshes its knowledge of the root volume, usually named root.afs. On success, it wakes up any of its own threads waiting on the arrival of this information, should it have been previously unreachable. This typically happens should the Cache Manager discover in its startup sequence that information on the root volume is unavailable. Lacking this knowledge at startup time, the Cache Manager settles into a semi-quiescent state, checking every so often to see if volume service is available and thus may complete its own initialization.
- 2 Each cached volume record is flagged as being stale. Any future attempt to access information from these volumes will result in the volume record's data first being refreshed from the Volume Location Server.
- One group of pioctl() opcodes is aimed at performing operations against one or more File Servers directly. Specifically, a caller may translate a pathname into the corresponding AFS fid (VIOCGETFID), unilaterally discard a set of callback promises (VIOCFLUSHCB), get status on mount points (VIOC AFS STAT MT PT), delete unwanted mount points (VIOC AFS DELETE MT PT), and check the health of a group of File Servers(VIOCCKSERV).
fid for named file system object
- [Opcode 22] Return the augmented file identifier for the file system object named by a pathP. The desired struct VenusFid is placed in the output buffer specified by out. The output buffer size, as indicated by the out size parameter, must be set to the value of sizeof(struct VenusFid) or greater. The input buffer is not referenced in this call, so in should be set to the null pointer and in size set to zero.
- Among the possible error returns, EINVAL indicates that the object named by a pathP was not found.
drop a callback
- [Opcode 25] Remove any callback information kept by the Cache Manager on the file system object named by a pathP. Internally, the Cache Manager executes a call to the RXAFS GiveUpCallBacks() RPC (See Section 5.1.3.13) to inform the appropriate File Server that it is being released from its particular callback promise. Note that if the named file resides on a read-only volume, then the above call is not made, and success is returned immediately. This optimization is possible because AFS File Servers do not grant callbacks on files from read-only volumes.
- Among the possible error returns is EINVAL, which indicates that the object named by a pathP was not found.
Delete a mount point
- [Opcode 28] Remove an AFS mount point. The name of the directory in which the mount point exists is specified by a pathP, and the string name of the mount point within this directory is provided through the in parameter. The input buffer length, in size, is set to the length of the mount point name itself, including the trailing null. The output buffer is not accessed by this call, so out should be set to the null pointer and out size to zero.
- One important note is that the a followSymLinks argument must be set to zero for correct operation. This is counter-intuitive, since at first glance it seems that a symbolic link that resolves to a directory should be a valid pathname parameter. However, recall that mount points are implemented as symbolic links that do not actually point to another file system object, but rather simply contain cell and volume information (see the description in Section 6.4.2). This "special" symbolic link must not be resolved by the pioctl(), but rather presented as-is to the Cache Manager, which then properly interprets it and generates a reference to the given volume's root directory. As an unfortunate side-effect, a perfectly valid symbolic link referring to a directory will be rejected out of hand by this operation as a value for the a pathP parameter.
- Among the possible error returns, EINVAL reports that the named directory was not found, and ENOTDIR indicates that the pathname contained within a pathP is not a directory.
contents of a mount point
- [Opcode 29] Return the contents of the given mount point. The directory in which the mount point in question resides is provided via the a pathP argument, and the in buffer contains the name of the mount point object within this directory. As usual, in size is set to the length of the input buffer, including the trailing null. If the given object is truly a mount point and the out buffer is large enough (its length appears in out size), the mount point's contents are stored into out.
- The mount point string returned obeys a stylized format, as fully described in Section 5.6.2 of the AFS 3.0 System Administrator's Guide[1]. Briefly, a leading pound sign ("#") indicates a standard mount point, inheriting the read-only or read-write preferences of the mount point's containing volume. On the other hand, a leading percent sign ("%") advises the Cache Manager to cross into the read-write version of the volume, regardless of the existence of read-only clones. If a colon (":") separator occurs, the portion up to the colon itself denotes the fully-qualified cell name hosting the volume. The rest of the string is the volume name itself.
- Among the possible error codes is EINVAL, indicating that the named object is not an AFS mount point. Should the name passed in a pathP be something other than a directory, then ENOTDIR is returned.
of one or more File Servers
- [Opcode 10] Check the status of the File Servers that have been contacted over the lifetime of the Cache Manager. The a pathP parameter is ignored by this call, so it should be set to the null pointer. The input parameters as specified by in are completely optional. If something is placed in the input buffer, namely in size is not zero, then the first item stored there is a longword used as a bit array of flags. These flags carry instructions as to the domain and the "thoroughness" of this check.
- Only the settings of the least-significant two bits are recognized. Enabling the lowest bit tells the Cache Manager not to ping its list of servers, but simply report their status as contained in the internal server records. Enabling the next-higher bit limits the search to only those File Servers in a given cell. If in size is greater than sizeof(long),a null-terminated cell name string follows the initial flag array, specifying the cell to check. If this search bit is set but no cell name string follows the longword of flags, then the search is restricted to those servers contacted from the same cell as the caller.
- This call returns at least one longword into the output buffer out, specifying the number of hosts it discovered to be down. If this number is not zero, then the longword IP address for each dead (or unreachable) host follows in the output buffer. At most 16 server addresses will be returned, as this is the maximum number of servers for which the Cache Manager keeps information.
- Among the possible error returns is ENOENT, indicating that the optional cell name string input value is not known to the Cache Manager.
- The Cache Manager is the only active AFS agent that understands the system's cellular architecture. Thus, it keeps important information concerning the identities of the cells in the community, which cell is in direct administrative control of the machine upon which it is running, status and configuration of its own cell, and what cell-specific operations may be legally executed. The following pioctl()s allow client processes to access and update this cellular information. Supported operations include adding or updating knowledge of a cell, including the cell overseeing the caller's machine (VIOCNEWCELL), fetching the contents of a cell configuration entry (VIOCGETCELL), finding out which cell hosts a given file system object (VIOC FILE CELL NAME), discovering the cell to which the machine belongs (VIOC GET WS CELL), finding out the caller's "primary" cell (VIOC GET PRIMARY CELL), and getting/setting certain other per-cell system parameters (VIOC GETCELLSTATUS, VIOC SETCELLSTATUS).
service information
- [Opcode 26] Give the Cache Manager all the information it needs to access an AFS cell. Exactly eight longwords are placed at the beginning of the in input buffer. These specify the IP addresses for the machine providing AFS authentication and volume location authentication services. The first such longword set to zero will signal the end of the list of server IP addresses. After these addresses, the input buffer hosts the null-terminated name of the cell to which the above servers belong. The a pathP parameter is not used, and so should be set to the null pointer.
- Among the possible error returns is EACCES, indicating that the caller does not have the necessary rights to perform the operation. Only root is allowed to set cell server information. If either the IP address array or the server name is unacceptable, EINVAL will be returned.
configuration entry
- [Opcode 27] Get the i'th cell configuration entry known to the Cache Manager. The index of the desired entry is placed into the in input buffer as a longword, with the first legal value being zero. If there is a cell associated with the given index, the output buffer will be filled with an array of 8 longwords, followed by a null-terminated string.
- The longwords correspond to the list of IP addresses of the machines providing AFS authentication and volume location services. The string reflects the name of the cell for which the given machines are operating. There is no explicit count returned of the number of valid IP addresses in the longword array. Rather, the list is terminated by the first zero value encountered, or when the eighth slot is filled.
- This routine is intended to be called repeatedly, with the index starting at zero and increasing each time. The array of cell information records is kept compactly, without holes. A return value of EDOM indicates that the given index does not map to a valid entry, and thus may be used as the terminating condition for the iteration.
cell hosting a given object
- [Opcode 30] Ask the Cache Manager to return the name of the cell in which the file system object named by a pathP resides. The input arguments are not used, so in should be set to the null pointer and in size should be set to zero. The null-terminated cell name string is returned in the out output buffer.
- Among the possible error values, EINVAL indicates that the pathname provided in a pathP is illegal. If there is no cell information associated with the given object, ESRCH is returned.
caller's home cell name
- [Opcode 31] Return the name of the cell to which the caller's machine belongs. This cell name is returned as a null-terminated string in the output buffer. The input arguments are not used, so in should be set to the null pointer and in size should be set to zero.
- Among the possible error returns is ESRCH, stating that the caller's home cell information was not available.
the caller's primary cell
- [Opcode 33] Ask the Cache Manager to return the name of the caller's primary cell. Internally, the Cache Manager scans its user records, and the cell information referenced by that record is used to extract the cell's string name. The input arguments are not used, so in should be set to the null pointer and in size should be set to zero. The a pathP pathname argument is not used either, and should similarly be set to the null pointer. The null-terminated cell name string is placed into the output buffer pointed to by out if it has suffcient room.
- Among the possible error returns is ESRCH, stating that the caller's primary cell information was not available.
status info for a cell entry
- [Opcode 35] Given a cell name, return a single longword of status flags from the Cache Manager's entry for that cell. The null-terminated cell name string is expected to be in the in parameter, with in size set to its length plus one for the trailing null. The status flags are returned in the out buffer, which must have out size set to sizeof(long) or larger.
- The Cache Manager defines the following output flag values for this operation:
- 0x1 This entry is considered the caller's primary cell.
- 0x2 The unix setuid() operation is not honored.
- 0x4 An obsolete version of the Volume Location Server's database is being used. While defined, this flag should no longer be set in modern systems.
- Among the possible error returns is ENOENT, informing the caller that the Cache Manager has no knowledge of the given cell name.
status info for a cell entry
- [Opcode 36] Given a cell name and an image of the cell status bits that should be set, record the association in the Cache Manager. The input buffer in must be set up as follows. The first entry is the longword containing the cell status bits to be set (see the VIOC GETCELLSTATUS description above for valid flag definitions). The next entry is another longword, ignored by the Cache Manager. The third and final entry in the input buffer is a null-terminated string containing the name of the cell for which the status flags are to be applied.
- Among the possible error returns is ENOENT, reflecting the Cache Manager's inability to locate its record for the given cell. Only root is allowed to execute this operation, and an EACCES return indicates the caller was not effectively root when the call took place.
- The Cache Manager serves as the repository for authentication information for AFS clients. Each client process belongs to a single Process Authentication Group (PAG). Each process in a given PAG shares authentication information with the other members, and thus has the identical rights with respect to AFS Access Control Lists (ACLs) as all other processes in the PAG. As the Cache Manager interacts with File Servers as a client process' agent, it automatically and transparently presents the appropriate authentication information as required in order to gain the access to which the caller is entitled. Each PAG can host exactly one token per cell. These tokens are objects that unequivocally codify the principal's identity, and are encrypted for security. Token operations between a Cache Manager and File Server are also encrypted, as are the interchanges between clients and the Authentication Servers that generate these tokens.
- There are actually two different flavors of tokens, namely clear and secret. The data structure representing clear tokens is described in Section 6.2.2, and the secret token appears as an undifferentiated byte stream.
- This section describes the operations involving these tokens, namely getting and setting the caller's token for a particular cell (VIOCGETTOK, VIOCSETTOK), checking a caller's access on a specified file system object (VIOCACCESS), checking the status of caller's tokens associated with the set of File Server connections maintained on its behalf (VIOCCKCONN), and discarding tokens entirely (VIOCUNLOG, VIOCUNPAG). These abilities are used by such programs as login, klog, unlog, and tokens, which must generate, manipulate, and/or destroy AFS tokens.
token for a cell
- [Opcode 3] Store the caller's secret and clear tokens within the Cache Manager. The input buffer is used to hold the following quantities, laid out end to end. The first item placed in the buffer is a longword, specifying the length in bytes of the secret token, followed by the body of the secret token itself. The next field is another longword, this time describing the length in bytes of the struct ClearToken, followed by the structure. These are all required fields. The caller may optionally include two additional fields, following directly after the required ones. The first optional field is a longword which is set to a non-zero value if the cell in which these tokens were generated is to be marked as the caller's primary cell. The second optional argument is a null-terminated string specifying the cell in which these tokens apply. If these two optional arguments do not appear, the Cache Manager will default to using its home cell and marking the entry as non-primary. The a pathP pathname parameter is not used, and thus should be set to the null pointer.
- If the caller does not have any tokens registered for the cell, the Cache Manager will store them. If the caller already has tokens for the cell, the new values will overwrite their old values. Because these are stored per PAG, the new tokens will thus determine the access rights of all other processes belonging to the PAG.
- Among the possible error returns is ESRCH, indicating the named cell is not recognized, and EIO, if information on the local cell is not available.
token for a cell
- [Opcode 8] Get the specified authentication tokens associated with the caller. The a pathP parameter is not used, so it should be set to the null pointer. Should the input parameter in be set to a null pointer, then this call will place the user's tokens for the machine's home cell in the out output buffer, if such tokens exist. In this case, the following objects are placed in the output buffer. First, a longword specifying the number of bytes in the body of the secret token is delivered, followed immediately by the secret token itself. Next is a longword indicating the length in bytes of the clear token, followed by the clear token. The input parameter may also consist of a single longword, indicating the index of the token desired. Since the Cache Manager is capable of storing multiple tokens per principal, this allows the caller to iteratively extract the full set of tokens stored for the PAG. The first valid index value is zero. The list of tokens is kept compactly, without holes. A return value of EDOM indicates that the given index does not map to a valid token entry, and thus may be used as the terminating condition for the iteration.
- Other than EDOM, another possible error return is ENOTCONN, specifying that the caller does not have any AFS tokens whatsoever.
access on object
- [Opcode 20] This operation is used to determine whether the caller has specific access rights on a particular file system object. A single longword is placed into the input buffer, in, representing the set of rights in question. The acceptable values for these access rights are listen in Section 6.4.5. The object to check is named by the a pathP parameter. The output parameters are not accessed, so out should be set to the null pointer, and out size set to zero. If the call returns successfully, the caller has at least the set of rights denoted by the bits set in the input buffer. Otherwise, EACCESS is returned.
caller's tokens/connections
- [Opcode 12] Check whether the suite of File Server connections maintained on behalf of the caller by the Cache Manager has valid authentication tokens. This function always returns successfully, communicating the health of said connections by writing a single longword value to the specified output buffer in out. If zero is returned to the output buffer, then two things are true. First, the caller has tokens for at least one cell. Second, all tokens encountered upon a review of the caller's connections have been properly minted (i.e., have not been generated fraudulently), and, in addition, have not yet expired. If these conditions do not currently hold for the caller, then the output buffer value will be set to EACCES. Neither the a pathP nor input parameters are used by this call.
authentication information
- [Opcode 9] Discard all authentication information held in trust for the caller. The Cache Manager sweeps through its user records, destroying all of the caller's associated token information. This results in reducing the rights of all processes within the caller's PAG to the level of file system access granted to the special system:anyuser group.
- This operation always returns successfully. None of the parameters are referenced, so they should all be set to null pointers and zeroes as appropriate.
authentication information
- [Opcode 21] This call is essentially identical to the VIOCUNLOG operation, and is in fact implemented internally by the same code for VIOCUNLOG.
- This set of opcodes allows manipulation of AFS Access Control Lists (ACLs). Callers are allowed to fetch the ACL on a given directory, or to set the ACL on a directory. In AFS-3, ACLs are only maintained on directories, not on individual files. Thus, a directory ACL determines the allowable accesses on all objects within that directory in conjunction with their normal unix mode (owner) bits. Should the a pathP parameter specify a file instead of a directory, the ACL operation will be performed on the directory in which the given file resides.
- These pioctl() opcodes deal only in external formats for ACLs, namely the actual text stored in an AFS ACL container. This external format is a character string, composed of a descriptive header followed by some number of individual principal-rights pairs. AFS ACLs actually specify two sublists, namely the positive and negative rights lists. The positive list catalogues the set of rights that certain principals (individual users or groups of users) have, while the negative list contains the set of rights specifically denied to the named parties.
- These external ACL representations differ from the internal format generated by the Cache Manager after a parsing pass. The external format may be easily generated from the internal format as follows. The header format is expressed with the following printf() statement:
printf("%d\n%d\n", NumPositiveEntries, NumNegativeEntries);
- The header first specifies the number of entries on the positive rights list, which appear first in the ACL body. The number of entries on the negative list is the second item in the header. The negative entries appear after the last positive entry.
- Each entry in the ACL proper obeys the format imposed by the following printf() statement:
printf("%s\t%d\n", UserOrGroupName, RightsMask);
- Note that the string name for the user or group is stored in an externalized ACL entry. The Protection Server stores the mappings between the numerical identifiers for AFS principals and their character string representations. There are cases where there is no mapping from the numerical identifier to a string name. For example, a user or group may have been deleted sometime after they were added to the ACL and before the Cache Manager externalized the ACL for storage. In this case, the Cache Manager sets UserOrGroupName to the string version of the principal's integer identifier. Should the erz principal be deleted from the Protection Server's database in the above scenario, then the string '1019' will be stored, since it corresponded to erz's former numerical identifier.
- The RightsMask parameter to the above call represents the set of rights the named principal may exercise on the objects covered by the ACL. The following flags may be OR'ed together to construct the desired access rights placed in RightsMask:
#define PRSFS_READ 1
#define PRSFS_WRITE 2
#define PRSFS_INSERT 4
#define PRSFS_LOOKUP 8
#define PRSFS_DELETE 16
#define PRSFS_LOCK 32
#define PRSFS_ADMINISTER 64
directory
- [Opcode 1] Set the contents of the ACL associated with the file system object named by a pathP. Should this pathname indicate a file and not a directory, the Cache Manager will apply this operation to the file's parent directory. The new ACL contents, expressed in their externalized form, are made available in in, with in size set to its length in characters, including the trailing null. There is no output from this call, so out size should be set to zero. Internally, the Cache Manager will call the RXAFS StoreACL() RPC (see Section 5.1.3.3 to store the new ACL on the proper File Server.
- Possible error codes include EINVAL, indicating that one of three things may be true: the named path is not in AFS, there are too many entries in the specified ACL, or a non-existent user or group appears on the ACL.
directory
- [Opcode 2] Get the contents of the ACL associated with the file system object named by a pathP. Should this pathname indicate a file and not a directory, the Cache Manager will apply this operation to the file's parent directory. The ACL contents, expressed in their externalized form, are delivered into the out buffer if out size has been set to a value which indicates that there is enough room for the specified ACL. This ACL string will be null-terminated. There is no input to this call, so in size should be set to zero. Internally, the Cache Manager will call the RXAFS FetchACL() RPC (see Section 5.1.3.1) to fetch the ACL from the proper File Server.
- Possible error codes include EINVAL, indicating that the named path is not in AFS.
- It is possible to inquire about and affect various aspects of the cache maintained locally by the Cache Manager through the group of pioctl()s described below. Specifically, one may force certain file system objects to be removed from the cache (VIOCFLUSH), set the maximum number of blocks usable by the cache (VIOCSETCACHESIZE), and ask for information about the cache's current state (VIOCGETCACHEPARAMS).
from the cache
- [Opcode 6] Flush the file system object specified by a pathP out of the local cache. The other parameters are not referenced, so they should be set to the proper combination of null pointers and zeroes.
- Among the possible error returns is EINVAL, indicating that the value supplied in the a pathP parameter is not acceptable.
maximum cache size in blocks
- [Opcode 24] Instructs the Cache Manager to set a new maximum size (in 1 Kbyte blocks) for its local cache. The input buffer located at in contains the new maximum block count. If zero is supplied for this value, the Cache Manager will revert its cache limit to its value at startup time. Neither the a pathP nor output buffer parameters is referenced by this operation. The Cache Manager recomputes its other cache parameters based on this new value, including the number of cache files allowed to be dirty at once and the total amount of space filled with dirty chunks. Should the new setting be smaller than the number of blocks currently being used, the Cache Manager will throw things out of the cache until it obeys the new limit.
- The caller is required to be effectively running as root, or this call will fail, returning EACCES. If the Cache Manager is configured to run with a memory cache instead of a disk cache, this operation will also fail, returning EROF.
current cache parameter values
- [Opcode 40] Fetch the current values being used for the cache parameters. The output buffer is filled with MAXGCSTATS (16) longwords, describing these parameters. Only the first two longwords in this array are currently set. The first contains the value of afs cacheBlocks, or the maximum number of 1 Kbyte blocks which may be used in the cache (see Section 6.4.8.2 for how this value may be set). The second longword contains the value of the Cache Manager's internal afs blocksUsed variable, or the number of these cache blocks currently in use. All other longwords in the array are set to zero. Neither the a pathP nor input buffer arguments are referenced by this call.
- This routine always returns successfully.
- There are several other AFS-specific operations accessible via the pioctl() interface that don't fit cleanly into the above categories. They are described in this section, and include manipulation of the socket-based Mariner file trace interface (VIOC AFS MARINER HOST), enabling and disabling of the file-based AFSLog output interface for debugging (VIOC VENUSLOG), getting and setting the value of the special pathname component mapping (VIOC AFS SYSNAME), and turning the NFS-AFS translator service on and off (VIOC EXPORTAFS).
Get/set file transfer monitoring output
- [Opcode 32] This operation is used to get or set the IP address of the host destined to receive Mariner output. A detailed description of the Cache Manager Mariner interface may be found in Section 6.7.
- The input buffer located at in is used to pass a single longword containing the IP address of the machine to receive output regarding file transfers between the Cache Manager and any File Server. If the chosen host IP address is 0xffffffff, the Cache Manager is prompted to turn off generation of Mariner output entirely. If the chosen host IP address is zero, then the Cache Manager will not set the Mariner host, but rather return the current Mariner host as a single longword written to the out output buffer. Any other value chosen for the host IP address enables Mariner output (if it was not already enabled) and causes all further traffic to be directed to the given machine.
- This function always returns successfully.
Enable/disable Cache Manager logging
- [Opcode 34] Tell the Cache Manager whether to generate debugging information, and what kind of debugging output to enable. The input buffer located at in is used to transmit a single longword to the Cache Manager, expressing the caller's wishes. Of the four bytes making up the longword, the highest byte indicates the desired value for the internal afsDebug variable, enabling or disabling general trace output. The next highest byte indicates the desired value for the internal netDebug variable, enabling or disabling network-level debugging traces. The third byte is unused, and the low-order byte represents an overall on/off value for the functionality. There is a special value for the low-order byte, 99, which instructs the Cache Manager to return the current debugging setting as a single longword placed into the output buffer pointed to by out. The a pathP parameter is not referenced by this routine.
- Trace output is delivered to the AFSLog file, typically located in the /usr/vice/etc directory. When this form of debugging output is enabled, the existing AFSLog file is truncated, and its file descriptor is stored for future use. When this debugging is disabled, a close() is done on the file, forcing all its data to disk. For additional information on the AFSLog file for collecting Cache Manager traces, please see the description in Section 6.6.2.1.
- This call will only succeed if the caller is effectively running as root. If this is not the case, an error code of EACCES is returned.
the mapping
- [Opcode 38] Get or set the value of the special pathname component understood by the Cache Manager. The input buffer pointed to by in is used to house a longword whose value determines whether the value is being set (1) or whether the current value is being fetched (0). If it is being set, then a null-terminated string is expected to follow in the input buffer, specifying the new value of . Otherwise, if we are asking the Cache Manager for the current setting, a null-terminated string bearing that value will be placed in the out output buffer. The a pathP parameter is not used by this call, and thus should be set to a null pointer.
- There are no special privileges required of the caller to fetch the value of the current mapping. However, a native caller must be running effectively as root in order to successfully alter the mapping. An unauthorized attempt to change the setting will be ignored, and cause this routine to return EACCES. This requirement is relaxed for VIOC AFS SYSNAME pioctl() calls emanating from foreign file systems such as NFS and accessing AFS files through the NFS-AFS translator. Each such remote caller may set its own notion of what the mapping is without affecting native AFS clients. Since the uid values received in calls from NFS machines are inherently insecure, it is impossible to enforce the fact that the caller is truly root on the NFS machine. This, while any principal running on an NFS machine may change that foreign machine's perception of , it does not impact native AFS users in any way.
Enable/disable NFS/AFS translation
- [Opcode 39] Enable or disable the ability of an AFS-capable machine to export AFS access to NFS clients. Actually, this is a general facility allowing exportation of AFS service to any number of other file systems, but the only support currently in place is for NFS client machines. A single longword is expected in the input buffer in. This input longword is partitioned into individual bytes, organized as follows. The high-order byte communicates the type of foreign client to receive AFS file services. There are currently two legal values for this field, namely 0 for the null foreign file system and 1 for NFS. The next byte determines whether the Cache Manager is being asked to get or set this information. A non-zero value here is interpreted as a command to set the export information according to what's in the input longword, and a zero-valued byte in this position instructs the Cache Manager to place a longword in the output buffer out, which contains the current export settings for the foreign system type specified in the high-order byte. The third input byte is not used, and the lowest-order input buffer byte determines whether export services for the specified system are being enabled or disabled. A non-zero value will turn on the services, and a zero value will shut them down. The a pathP pathname parameter is not used by this call, and the routine generates output only if the export information is being requested instead of being set.
- The caller must be effectively running as root in order for this operation to succeed. The call returns EACCES if the caller is not so authorized. If the caller specifies an illegal foreign system type in the high-order byte of the input longword, then ENODEV is returned. Again, NFS is the only foreign file system currently supported.
- Practically speaking, the machine providing NFS-AFS translation services must enable this service with this pioctl() before any NFS client machines may begin accessing AFS files. Conversely, if an administrator turns off this export facility, the export code on the translator machine will immediately stop responding to traffic from its active NFS clients.
- This section covers the structure and workings of the Cache Manager's RPC interface. Typically, these calls are made by File Server processes. However, some of the calls are designed specifically for debugging programs (e.g., the cmdebug facility) and for collection of statistical and performance information from the Cache Manager. Any client application that makes direct calls on the File Server RPC interface must be prepared to export a subset of the Cache Manager RPC interface, as discussed in Section 5.1.6.
- This section will first examine the Cache Manager's use of locks, whose settings may be observed via one of the RPC interface calls. Next, it will present some definitions and data structures used in the RPC interface, and finally document the individual calls available through this interface.
- The Cache Manager makes use of locking to insure its internal integrity in the face of its multi-threaded design. A total of 11 locks are maintained for this purpose, one of which is now obsolete and no longer used (see below). These locks are strictly internal, and the Cache Manager itself is the only one able to manipulate them. The current settings for these system locks are externally accessible for debugging purposes via the AFSRXCB GetLock() RPC interface call, as described in Section 6.5.5.4. For each lock, its index in the locking table is given in the following text.
- afs xvcache [Index 0]: This lock controls access to the status cache entries maintained by the Cache Manager. This stat cache keeps stat()-related information for AFS files it has dealt with. The stat information is kept separate from actual data contents of the related file, since this information may change independently (say, as a result of a unix chown() call.
- afs xdcache [Index 1]: This lock moderates access to the Cache Manager's data cache, namely the contents of the file system objects it has cached locally. As stated above, this data cache is separate from the associated stat() information.
- afs xserver [Index 2]: This lock controls access to the File Server machine description table, which keeps tabs on all File Servers contacted in recent history. This lock thus indirectly controls access to the set of per-server RPC connection descriptors the File Server table makes visible.
- afs xvcb [Index 3]: This lock supervises access to the volume callback information kept by the Cache Manager. This table is referenced, for example, when a client decides to remove one or more callbacks on files from a given volume (see the RXAFS GiveUpCallBacks() description on Section 5.1.3.13).
- afs xbrs [Index 4]: This lock serializes the actions of the Cache Manager's background daemons, which perform prefetching and background file storage duties.
- afs xcell [Index 5]: This lock controls the addition, deletion, and update of items on the linked list housing information on cells known to the Cache Manager.
- afs xconn [Index 6]: This lock supervises operations concerning the set of RPC connection structures kept by the system. This lock is used in combination with the
- afs xserver lock described above. In some internal Cache Manager code paths, the File Server description records are first locked, and then the afs xconn lock is used to access the associated Rx connection records. afs xuser [Index 7]: This lock serializes access to the per-user structures maintained by the Cache Manager.
- afs xvolume [Index 8]: This lock is used to control access to the Cache Manager's volume information cache, namely the set of entries currently in memory, a subset of those stably housed in the VolumeItems disk file (see Section 6.6.2.3).
- afs puttofileLock [Index 9]: This lock is obsolete, and while still defined by the system is no longer used. It formerly serialized writes to a debugging output interface buffer, but the internal mechanism has since been updated and improved.
- afs ftf [Index 10]: This lock is used when flushing cache text pages from the machine's virtual memory tables. For each specific machine architecture on which the Cache Manager runs, there is a set of virtual memory operations which must be invoked to perform this operation. The result of such activities is to make sure that the latest contents of new incarnations of binaries are used, instead of outdated copies of previous versions still resident in the virtual memory system.
- This section documents some macro definitions and typedefs referenced by the Cache Manager's RPC interface. Specifically, these definitions and typedefs are used in the RXAFSCB GetXStats() and RXAFSCB XStatsVersion calls as described in Sections 6.5.5.6 and 6.5.5.7.
const AFSCB_XSTAT_VERSION = 1;
const AFSCB_MAX_XSTAT_LONGS = 2048;
typedef long AFSCB_CollData<AFSCB_MAX_XSTAT_LONGS>;
const AFSCB_XSTATSCOLL_CALL_INFO = 0;
const AFSCB_XSTATSCOLL_PERF_INFO = 1;
- This section documents some structures used in the Cache Manager RPC interface. As with the constants and typedefs in the previous section, these items are used in the RXAFSCB GetXStats() and RXAFSCB XStatsVersion calls as described in Sections 6.5.5.6 and 6.5.5.7.
- This structure may be used to collect a running average figure. It is included in some of the statistics structures described below.
Fields
- long average - The computed average.
- long elements - The number of elements sampled for the above aveage.
- This structure maintains profiling information, communicating the number of times internal Cache Manager functions are invoked. Each field name has a "C " prefix, followed by the name of the function being watched. As this structure has entries for over 500 functions, it will not be described further here. Those readers who wish to see the full layout of this structure are referred to Appendix A.
- The AFSCB XSTATSCOLL CALL INFO data collection includes the information in this structure.
- This is the other part of the information (along with the struct afs CMCallStats construct described above) returned by the AFSCB XSTATSCOLL CALL INFO data collection defined by the Cache Manager (see Section 6.5.3). It is accessible via the RXAFSCB GetXStats() interface routine, as defined in Section 6.5.5.7.
- This structure represents the beginning of work to compute average values for some of the extended statistics collected by the Cache Manager.
Fields
- struct afs MeanStats something - Intended to collect averages for some of the Cache Manager extended statistics; not yet implemented.
- This structure defines the information returned by the AFSCB XSTATSCOLL CALL INFO data collection defined by the Cache Manager (see Section 6.5.3). It is accessible via the RXAFSCB GetXStats() interface routine, as defined in Section 6.5.5.7.
Fields
- struct afs CallStats callInfo - Contains the counts on the number of times each internal Cache Manager function has been called.
- struct afs MeanStats something - Intended to collect averages for some of the Cache Manager extended statistics; not yet implemented.
- This is the information returned by the AFSCB XSTATSCOLL PERF INFO data collection defined by the Cache Manager (see Section 6.5.3). It is accessible via the RXAFSCB GetXStats() interface routine, as defined in Section 6.5.5.7.
Fields
- long numPerfCalls - Number of performance calls received.
- long epoch - Cache Manager epoch time.
- long numCellsContacted - Number of cells contacted.
- long dlocalAccesses - Number of data accesses to files within the local cell.
- long vlocalAccesses - Number of stat accesses to files within the local cell.
- long dremoteAccesses - Number of data accesses to files outside of the local cell.
- long vremoteAccesses - Number of stat accesses to files outside of the local cell.
- long cacheNumEntries - Number of cache entries.
- long cacheBlocksTotal - Number of (1K) blocks configured for the AFS cache.
- long cacheBlocksInUse - Number of cache blocks actively in use.
- long cacheBlocksOrig - Number of cache blocks configured at bootup.
- long cacheMaxDirtyChunks - Maximum number of dirty cache chunks tolerated.
- long cacheCurrDirtyChunks - Current count of dirty cache chunks.
- long dcacheHits - Number of data file requests satisfied by the local cache.
- long vcacheHits - Number of stat entry requests satisfied by the local cache.
- long dcacheMisses - Number of data file requests not satisfied by the local cache.
- long vcacheMisses - Number of stat entry requests not satisfied by the local cache.
- long cacheFlushes - Number of files flushed from the cache.
- long cacheFilesReused - Number of cache files reused.
- long numServerRecords - Number of records used for storing information concerning File Servers.
- long ProtServerAddr - IP addres of the Protection Server used (not implemented).
- long spare[32] - A set of longword spares reserved for future use.
- This section discusses the Cache Manager interface calls. No special permissions are required of the caller for any of these operations. A summary of the calls making up the interface appears below:
- RXAFSCB Probe() "Are-you-alive" call.
- RXAFSCB CallBack() Report callbacks dropped by a File Server.
- RXAFSCB InitCallBackState() Purge callback state from a File Server.
- RXAFSCB GetLock() Get contents of Cache Manager lock table.
- RXAFSCB GetCE() Get cache file description.
- RXAFSCB XStatsVersion() Get version of extended statistics package.
- RXAFSCB GetXStats() Get contents of extended statistics data collection.
that the underlying callback service is still operational
int RXAFSCB Probe(IN struct rx call *a rxCallP)
- Description
- [Opcode 206] This call simply implements an "are-you-alive" operation, used to determine if the given Cache Manager is still running. Any File Server will probe each of the Cache Managers with which it has interacted on a regular basis, keeping track of their health. This information serves an important purpose for a File Server. In particular, it is used to trigger purging of deceased Cache Managers from the File Server's callback records, and also to instruct a new or "resurrected" Cache Manager to purge its own callback state for the invoking File Server.
- Rx call information for the related Cache Manager is contained in a rxCallP.
- Error Codes
- ---No error codes are generated.
callbacks dropped by a File Server
int RXAFSCB CallBack(IN struct rx call *a rxCallP,
IN AFSCBFids *a fidArrayP,
IN AFSCBs *a callBackArrayP)
- Description
- [Opcode 204] Provide information on dropped callbacks to the Cache Manager for the calling File Server. The number of fids involved appears in a fidArrayP->AFSCBFids len, with the fids themselves located at a fidArrayP->AFSCBFids val. Similarly, the number of associated callbacks is placed in a callBackArrayP->AFSCBs len, with the callbacks themselves located at a callBackArrayP->AFSCBs val.
- Rx call information for the related Cache Manager is contained in a rxCallP.
- Error Codes
- ---No error codes are generated.
Purge callback state from a File Server
int RXAFSCB InitCallBackState(IN struct rx call *a rxCallP)
- Description
- [Opcode 205] This routine instructs the Cache Manager to purge its callback state for all files and directories that live on the calling host. This function is typically called by a File Server when it gets a request from a Cache Manager that does not appear in its internal records. This handles situations where Cache Managers survive a File Server, or get separated from it via a temporary network partition. This also happens upon bootup, or whenever the File Server must throw away its record of a Cache Manager because its tables have been filled.
- Rx call information for the related Cache Manager is contained in a rxCallP.
- Error Codes
- ---No error codes are generated.
contents of Cache Manager lock table
int RXAFSCB GetLock(IN struct rx call *a rxCall,
IN long a index,
OUT AFSDBLock *a lockP)
- Description
- [Opcode 207] Fetch the contents of entry a index in the Cache Manager lock table. There are 11 locks in the table, as described in Section 6.5.2. The contents of the desired lock, including a string name representing the lock, are returned in a lockP.
- This call is not used by File Servers, but rather by debugging tools such as cmdebug.
- Rx call information for the related Cache Manager is contained in a rxCallP.
- Error Codes
- The index value supplied in a index is out of range; it must be between 0 and 10.
file description
int RXAFSCB GetCE(IN struct rx call *a rxCall,
IN long a index,
OUT AFSDBCacheEntry *a ceP)
- Description
- [Opcode 208] Fetch the description for entry a index in the Cache Manager file cache, storing it into the buffer to which a ceP points. The structure returned into this pointer variable is described in Section 4.3.2.
- This call is not used by File Servers, but rather by debugging tools such as cmdebug.
- Rx call information for the related Cache Manager is contained in a rxCallP.
- Error Codes
- The index value supplied in a index is out of range.
version of extended statistics package
int RXAFSCB XStatsVersion(IN struct rx call *a rxCall,
OUT long *a versionNumberP)
- Description
- [Opcode 209] This call asks the Cache Manager for the current version number of the extended statistics structures it exports (see RXAFSCB GetXStats(), Section 6.5.5.7). The version number is placed in a versionNumberP.
- Rx call information for the related Cache Manager is contained in a rxCallP.
- Error Codes
- ---No error codes are generated.
contents of extended statistics data collection
int RXAFSCB GetXStats(IN struct rx call *a rxCall,
IN long a clientVersionNumber,
IN long a collectionNumber,
OUT long *a srvVersionNumberP,
OUT long *a timeP,
OUT AFSCB CollData *a dataP)
- Description
- [Opcode 210] This function fetches the contents of the specified Cache Manager extended statistics structure. The caller provides the version number of the data it expects to receive in a clientVersionNumber. Also provided in a collectionNumber is the numerical identifier for the desired data collection. There are currently two of these data collections defined: AFSCB XSTATSCOLL CALL INFO, which is the list of tallies of the number of invocations of internal Cache Manager procedure calls, and AFSCB XSTATSCOLL PERF INFO, which is a list of performance-related numbers. The precise contents of these collections are described in Section 6.5.4. The current version number of the Cache Manager collections is returned in a srvVersionNumberP, and is always set upon return, even if the caller has asked for a different version. If the correct version number has been specified, and a supported collection number given, then the collection data is returned in a dataP. The time of collection is also returned, being placed in a timeP.
- Rx call information for the related Cache Manager is contained in a rxCallP.
- Error Codes
- The collection number supplied in a collectionNumber is out of range.
- The Cache Manager gets some of its start-up configuration information from files located on the client machine's hard disk. Each client is required to supply a /usr/vice/etc directory in which this configuration data is kept. Section 6.6.1 describes the format and purpose of the three files contributing this setup information: ThisCell, CellServDB, and cacheinfo.
- The Cache Manager, along with various applications, needs to be able to determine the cell to which its client machine belongs. This information is provided by the ThisCell file. It contains a single line stating the machine's fully-qualified cell name.
- As with the CellServDB configuration file, the Cache Manager reads the contents of ThisCell exactly once, at start-up time. Thus, an incarnation of the Cache Manager will maintain precisely one notion of its home cell for its entire lifetime. Thus, changes to the text of the ThisCell file will be invisible to the running Cache Manager. However, these changes will affect such application programs as klog, which allows a user to generate new authentication tickets. In this example, klog reads ThisCell every time it is invoked, and then interacts with the set of Authentication Servers running in the given home cell, unless the caller specifies the desired cell on the command line.
- The ThisCell file is not expected to be changed on a regular basis. Client machines are not imagined to be frequently traded between different administrative organizations. The Unix mode bits are set to specify that while everyone is allowed to read the file, only root is allowed to modify it.
- To conduct business with a given AFS cell, a Cache Manager must be informed of the cell's name and the set of machines running AFS database servers within that cell. Such servers include the Volume Location Server, Authentication Server, and Protection Server. This particular cell information is obtained upon startup by reading the CellServDB file. Thus, when the Cache Manager initialization is complete, it will be able to communicate with the cells covered by CellServDB.
- The following is an excerpt from a valid CellServDB file, demonstrating the format used.
...
>transarc.com #Transarc Corporation
192.55.207.7 #henson.transarc.com
192.55.207.13 #bigbird.transarc.com
192.55.207.22 #ernie.transarc.com
>andrew.cmu.edu #Carnegie Mellon University
128.2.10.2 #vice2.fs.andrew.cmu.edu
128.2.10.7 #vice7.fs.andrew.cmu.edu
128.2.10.10 #vice10.fs.andrew.cmu.edu
...
- There are four rules describing the legal CellServDB file format:
- 1. Each cell has a separate entry. The entries may appear in any order. It may be convenient, however, to have the workstation's local cell be the first to appear.
- 2. No blank lines should appear in the file, even at the end of the last entry.
- 3. The first line of each cell's entry begins with the '>' character, and specifies the cell's human-readable, Internet Domain-style name. Optionally, some white space and a comment (preceded by a '#') may follow, briefly describing the specified cell.
- 4. Each subsequent line in a cell's entry names one of the cell's database server machines. The following must appear on the line, in the order given:
- The Internet address of the server, in the standard 4-component dot notation.
- Some amount of whitespace.
- A '#', followed by the machine's complete Internet host name. In this instance, the '#' sign and the text beyond it specifying the machine name are NOT treated as a comment. This is required information.
- The Cache Manager will use the given host name to determine its current address via an Internet Domain lookup. If and only if this lookup fails does the Cache Manager fall back to using the dotted Internet address on the first part of the line. This dotted address thus appears simply as a hint in case of Domain database downtime.
- The CellServDB file is only parsed once, when the Cache Manager first starts. It is possible, however, to amend existing cell information records or add completely new ones at any time after Cache Manager initialization completes. This is accomplished via the VIOCNEWCELL pioctl() (see Section 6.4.5.1.
- This one-line file contains three fields separated by colons:
- AFS Root Directory: This is the directory where the Cache Manager mounts the AFS root volume. Typically, this is specified to be /afs.
- Cache Directory: This field names the directory where the Cache Manager is to create its local cache files. This is typically set to /usr/vice/cache.
- Cache Blocks: The final field states the upper limit on the number of 1,024-byte blocks that the Cache Manager is allowed to use in the partition hosting the named cache directory.
- Thus, the following cacheinfo file would instruct the Cache Manager to mount the AFS filespace at /afs, and inform it that it may expect to be able to use up to 25,000 blocks for the files in its cache directory, /usr/vice/cache.
/afs:/usr/vice/cache:25000
- This is the AFS log file used to hold Cache Manager debugging output. The file is set up when the Cache Manager first starts. If it already exists, it is truncated. If it doesn't, it is created. Output to this file is enabled and disabled via the the VIOC VENUSLOG pioctl() (see Section 6.4.9.2). Normal text messages are written to this file by the Cache Manager when output is enabled. Each time logging to this file is enabled, the AFSLog file is truncated. Only root can read and write this file.
- The Cache Manager only keeps a subset of its data cache entry descriptors in memory at once. The number of these in-memory descriptors is determined by afsd. All of the data cache entry descriptors are kept on disk, in the CacheItems file. The file begins with a header region, taking up four longwords:
struct fheader { long magic AFS_FHMAGIC 0x7635fab8 long firstCSize: First chunk
size long otherCSize: Next chunk sizes long spare }
- The header is followed by one entry for each cache file. Each is:
struct fcache {
short hvNextp;
short hcNextp;
short chunkNextp;
struct VenusFid fid;
long modTime;
long versionNo;
long chunk;
long inode;
long chunkBytes;
char states;
};
- The Cache Manager only keeps at most MAXVOLS (50) in-memory volume descriptions. However, it records all volume information it has obtained in the VolumeItems file in the chosen AFS cache directory. This file is truncated when the Cache Manager starts. Each volume record placed into this file has the following struct fvolume layout:
struct fvolume {
long cell;
long volume;
long next;
struct VenusFid dotdot;
struct VenusFid mtpoint;
};
- The Cache Manager Mariner interface allows interested parties to be advised in real time as to which files and/or directories are being actively transferred between the client machine and one or more File Servers. If enabled, this service delivers messages of two different types, as exemplified below:
Fetching myDataDirectory
Fetching myDataFile.c
Storing myDataObj.o
- In the first message, the myDataDirectory directory is shown to have just been fetched from a File Server. Similarly, the second message indicates that the C program myDataFile.c had just been fetched from its File Server of residence. Finally, the third message reveals that the myDataObj.o object file has just been written out over the network to its respective server.
- In actuality, the text of the messages carries a string prefix to indicate whether a Fetch or Store operation had been performed. So, the full contents of the above messages are as follows:
fetch$Fetching myDataDirectory
fetch$Fetching myDataFile.c
store$Storing myDataObj.o
The Mariner service may be enabled or disabled for a particular machine by using the VIOC AFS MARINER HOST pioctl() (see Section 6.4.9.1). This operation allows any host to be specified as the recipient of these messages. A potential recipient must have its host be declared the target of such messages, then listen to a socket on port 2106.
- Internally, the Cache Manager maintains a cache of NMAR (10) vnode structure pointers and the string name (up to 19 characters) of the associated file or directory. This cache is implemented as an array serving as a circular buffer. Each time a file is involved in a create or lookup operation on a File Server, the current slot in this circular buffer is filled with the relevant vnode and string name information, and the current position is advanced. If Mariner output is enabled, then an actual network fetch or store operation will trigger messages of the kind shown above. Since a fetch or store operation normally occurs shortly after the create or lookup, the mapping of vnode to name is likely to still be in the Mariner cache when it comes time to generate the appropriate message. However, since an unbounded number of other lookups or creates could have been performed in the interim, there is no guarantee that the mapping entry will not have been overrun. In these instances, the Mariner message will be a bit vaguer. Going back to our original example,
Fetching myDataDirectory
Fetching a file
Storing myDataObj.o
In this case, the cached association between the vnode containing myDataFile.c and its string name was thrown out of the Mariner cache before the network fetch operation could be performed. Unable to find the mapping, the generic phrase "a file" was used to identify the object involved.
- Mariner messages only get generated when RPC traffic for fetching or storing a file system object occurs between the Cache Manager and a File Server. Thus, file accesses that are handled by the Cache Manager's on-board data cache do not trigger such announcements.
Generated on Mon Aug 3 00:31:55 2009 by
1.5.1