Linux Kernel Architecture

Chapter 8: The Virtual Filesystem

fd_install

sys_open

force_o_largefile

do_sys_open

get_unused_fd_flags

do_flip_open

open_namei

nameidata_to_filp

Figure 8-11: Code flow diagram forsys_open.

do_filp_openfinds the file inode with the support of two helper functions.

1. open_nameiinvokes thepath_lookupfunction to find the inode and performs several
   additional checks (e.g., to ascertain whether the application is trying to open a directory
   as if it were a regular file). If a new filesystem entry needs to be created, the function also
   applies the current default settings for the permission bits as stored in the process’s umask
   (current->fs->umask).


2. nameidata_to_filpinitializes the readahead structure, places the newly generatedfile
   instance on thes_fileslist of the superblock (see Section 8.4.1), and invokes theopenfunc-
   tion in thefile_operationsstructure of the underlying filesystem.

fd_installmust then install thefileinstance infiles->fdfrom the task structure of the process before

control is transferred back to the user process to which the file descriptor is returned.

Reading andWriting

Once a file has been successfully opened, a process either reads or modifies the data it contains using the

readandwritesystem calls provided by the kernel. As usual, the entry routines are calledsys_readand

sys_write, and both are implemented infs/read_write.c.

Read

Thereadfunction requires three parameters — the file descriptor, a buffer to hold data, and a length

argument to specify the number of characters to be read. The parameters are passed directly to the kernel.

For the VFS layer, reading from a file is not particularly difficult, as Figure 8-12 illustrates.

By reference to the file descriptor number, the kernel (using thefget_lightfunction from

fs/file_table.c)isabletofindthefileinstance associated with the task structure of the process.

After finding the current position in the file withfile_pos_read(the routine just needs to return the

value offile->f_pos), the read operation itself is delegated tovfs_read. This routine invokes either