Linux Kernel Architecture

Chapter 10: Filesystems without Persistent Storage

proc_inode VFS-Inode

PROC_I

Figure 10-1: Connection betweenstruct proc_inode

andstruct inode.

The remaining elements of the structure are only used if the inode represents a process-specific entry

(which is therefore located in theproc/piddirectory). Their meanings are as follows:

❑ pidis a pointer to thepidinstance of a process. Because it is possible to access a large amount of

process-specific information this way, it is clear why a process-specific inode should be directly

associated with this data.

❑ proc_get_linkandproc_read(which are collected in a union because only one at a time makes

sense) are used to get process-specific information or to generate links to process-specific data in

the Virtual Filesystem.

❑ fdholds the filedescriptor for which a file in/proc/<pid>/fd/presents information. With the

help offd, all files in this directory can use the samefile_operations.

The meanings and use of these elements are discussed in detail in Section 10.1.7.

10.1.3 Initialization

Before theprocfilesystem can be used, it must be mounted withmount, and the kernel must set up and

initialize several data structures to describe the filesystem structure in kernel memory. Unfortunately, the

appearance and contents of/procdiffer substantially from platform to platform and from architecture

to architecture, and the code is crammed with#ifdefpre-processor statements that select code sections

according to the particular situation. Although this practice is frowned upon, it simply cannot be avoided.

Because initialization differences relate primarily to creation of the subdirectories that subse-

quently appear in/proc, they are not evident in Figure 10-2, which shows a code flow diagram of

proc_root_initinfs/proc/root.c.

proc_root_init

proc_init_inodecache

register_filesystem

kern_mount_data

proc_misc_init

proc_net_init

Create directories with proc_mkdir

Figure 10-2: Code flow diagram for

proc_root_init.