Readingand Writing
The SMB protocol uses the READ and WRITE message types to perform I/O operations on a file for the client. Using the READ request, a client can request that the server return information from the file by specifying a number of bytes and an offset into the file. The server returns the data, indicating the actual number of bytes returned, which can be less than requested if the user tries to read past the end of a file.
The WRITE command updates a file in a similar manner. The client sends in the data that will be written, indicating the number of bytes to write and an offset into the file where the write operation will begin. If the request causes a write past the end of the file, the file is extended to make it larger. The server sends a response telling the client the number of bytes that were written. If the number is less than the requested value, an error has occurred.
To increase read/write performance, the READ_RAW and WRITE_RAW message types can be used to exchange much larger blocks of information between the client and the server. When these are used, the client must have only one request issued to the server. In one send, the server will respond with data that can be as many as 65,535 bytes in length. The WRITE command works in the opposite direction, allowing the client to send a large buffer of raw data to the server for a write operation.
Locking Mechanisms
Locking allows a particular client exclusive access to a file or a part of a file when it is shared on th network. In SMB, the capability to create a lock is called an opportunistic lock, or oplock for short. This is better explained by looking at the way in which it works. A client can create a lock on a resource using three kinds of locks. The first is an exclusive lock, in which the client has exclusive access to the data held by the lock. A batch oplock is one that is kept open by the server when the client process has already closed the file. A Level II oplock is one in which there can be multiple readers of the same file.
The locking process consists of the client requesting the type of lock it wants when it opens the file. The server replies to the client with the type of lock that was granted when it responds to the open request.
A lock gives the client the capability to efficiently manage buffer space it uses when accessing a file over the network. For example, if a client has exclusive access to a file and is performing writes to it, it can buffer a lot of the newly written information before having to send it to the server to update the file. This can provide a reduced number of network packets when updating a file. A client that has an exclusive lock on a file can also buffer read-ahead data to make reading a file much faster.
These locks are called opportunistic locks for a reason. A client can be granted exclusive access to a file if no other client has it open at the time of the request. What happens when another client needs to read the file? The server notifies the first client that it needs to break the exclusive lock. The client then flushes its buffers so that any data that has not been written to the file is processed. The client then sends an acknowledgment to the server that it recognizes that the exclusive lock has been broken.
Batch oplocks are used to reduce the amount of traffic on the network when some programs require continual reopening of a file to obtain commands, as when a batch command procedure is executed.
For example, a batch procedure executed by the command processor usually opens a file, locates the next line to be executed, reads that line, closes the file, and then executes the command. The problem with this is that these steps are taken for each command line in the procedure, resulting in multiple file open/closes that are not really necessary.
This procedure for reading individual lines from a file is done by using a batch oplock whereby the client can read the data from its local read-ahead cache instead of reopening the file on the remote server to get each line.
Level II oplocks were new with the NT changes to SMB. This kind of lock allows more than one client to have a file opened for reading. When a client must read from a file that is opened by another exclusively, the server informs the current client that its exclusive lock has been broken and is now a Level II oplock. No client that has a Level II oplock will buffer data to or from the file. Thus, after the lock has changed to a Level II oplock (and the first client has flushed any data in its buffers), both clients can continue reading the file.