Linux Kernel Architecture

Chapter 8: The Virtual Filesystem

Implementation ofdo_lookup

do_lookupstarts from a path component and thenameidatainstance with the data of the initial directory

and returns the associated inode.

The kernel first attempts to find the inode in the dentry cache using the__d_lookupfunction described in

Section 8.3.5. Even if a matching element is found, this does not necessarily mean that it is current — the

d_revalidatefunction in thedentry_operationsof the underlying filesystem must be invoked to check

whether it is still valid. If so, it is returned as the result of the cache search. If not, a lookup operation

must be initiated in the low-level filesystem. The same operation is used when no entry is found in the

cache.

real_lookupperforms the filesystem-specific lookup action. Its work involves allocating data structures

in memory (to hold the lookup result) and, above all, invoking the filesystem-specificlookupfunction

made available by the inode operation structureinode_operations.

If the required directory exists, the result received by the kernel is a filleddentryinstance; otherwise, a

null pointer is returned. Note that Chapter 9 describesin greater detail how filesystems perform low-level

lookups.

do_lookupalso needs to take care of following mount points. If a validdentryis found in the cache,

__follow_mounttakes care of this. As discussed in Section 8.4.1, the kernel records the fact that a filesys-

tem is mounted by incrementing thed_mountedstructure element of the associateddentrys. To ensure

that mounting has the desired effect, the kernel musttake this fact into account when traversing the direc-

tory structure. This is done by invoking__follow_mount, whose implementation is surprisingly simple

(thepathstructure used as argument collects the required pointers to thevfsmountanddentryinstances

of the mount point).^22

fs/namei.c

static int __follow_mount(struct path *path)

{

int res = 0;

while (d_mountpoint(path->dentry)) {

struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);

if (!mounted)

break;

path->mnt = mounted;

path->dentry = mounted->mnt_root;

res = 1;

}

return res;

}

How does this loop work? A check is first made to ascertain whether the currentdentryinstance is a

mount point. In this context, thed_mountpointmacro need only check whether the value ofd_mounted

is greater than 0. Thelookup_mountfunction extracts thevfsmountinstance of the mounted filesystem

from themount_hashtablediscussed in Section 8.4.1. Themnt_rootfield of the mounted filesystem is

used as the new value for thedentrystructure; all this means is that the root directory of the mounted

filesystem is used as the mount point — and this is exactly what we want to achieve.

(^22) I have omitted the required locking and reference counting operations, which would make the code less readable.