Manual browser: cache_purgevfs(9)

Section:
Page:
NAMECACHE(9) Kernel Developer's Manual NAMECACHE(9)

NAME

namecache, cache_lookup, cache_revlookup, cache_enter, cache_purge, cache_purgevfs, namecache_printname lookup cache

SYNOPSIS

#include <sys/namei.h>
#include <sys/proc.h>
#include <sys/uio.h>
#include <sys/vnode.h>

int
cache_lookup(struct vnode *dvp, const char *name, size_t namelen, uint32_t nameiop, uint32_t nameiflags, int *iswhiteout, struct vnode **vpp);

int
cache_revlookup(struct vnode *vp, struct vnode *dvp, char **bpp, char *bufp);

void
cache_enter(struct vnode *dvp, struct vnode *vp, const char *name, size_t namelen, uint32_t nameiflags);

void
cache_purge(struct vnode *vp);

void
cache_purgevfs(struct mount *mp);

void
namecache_print(struct vnode *vp, void (*func)(const char *, ...));

DESCRIPTION

The name lookup cache is the mechanism to allow the file system type dependent algorithms to quickly resolve a file's vnode from its pathname. The name lookup cache is generally accessed through the higher-level namei(9) interface.

The name of the file is used to look up an entry associated with the file in the name lookup cache. If no entry is found, one is created for it. If an entry is found, the information obtained from the cache lookup contains information about the file which is useful to the file system type dependent functions.

The name lookup cache is managed by a least recently used (LRU) algorithm so frequently used names will hang around. The cache itself is a hash table called nchashtbl, containing namecache entries that are allocated dynamically from a kernel memory pool. Each entry has the following structure:

#define NCHNAMLEN	31	/* maximum name segment length */ 
struct  namecache { 
        LIST_ENTRY(namecache) nc_hash;  /* hash chain */ 
        TAILQ_ENTRY(namecache) nc_lru;  /* LRU chain */ 
        LIST_ENTRY(namecache) nc_vhash; /* directory hash chain */ 
        LIST_ENTRY(namecache) nc_dvlist; 
        struct  vnode *nc_dvp;          /* vnode of parent of name */ 
        LIST_ENTRY(namecache) nc_vlist; 
        struct  vnode *nc_vp;           /* vnode the name refers to */ 
        int     nc_flags;               /* copy of componentname's ISWHITEOUT */ 
        char    nc_nlen;                /* length of name */ 
        char    nc_name[NCHNAMLEN];     /* segment name */ 
};

For simplicity (and economy of storage), names longer than a maximum length of NCHNAMLEN are not cached; they occur infrequently in any case, and are almost never of interest.

Each namecache entry can appear on two hash chains in addition to nshashtbl: ncvhashtbl (the name cache directory hash chain), and nclruhead (the name cache LRU chain). The hash chains are indexed by a hash value obtained from the file's name and the address of its parent directory vnode.

Functions which access to the name cache pass arguments in the partially initialised componentname structure. See vnodeops(9) for details on this structure.

FUNCTIONS

cache_lookup(dvp, name, namelen, nameiop, nameiflags, iswhiteout, vpp)
Look for a name in the cache. cache_lookup() is called with dvp pointing to the vnode of the directory to search. The name and namelen arguments specify the name to look for. The nameiop and nameiflags should be taken from the cn_nameiop and cn_flags fields of struct componentname.

The lookup can produce either a cache miss or a cache hit, and a cache hit can either be a positive hit, where the name looked up refers to some existing object, or a negative hit, where the name looked up is known to refer to no existing object. (The lookup cannot fail, in the sense of generating an error condition that requires aborting the operation in progress.)

On a cache miss, cache_lookup() returns zero (false). On a positive hit, the unlocked vnode for the object found is stored in vpp, and a nonzero (true) value is returned. On a negative hit, vpp is set to contain a null pointer and a nonzero value is returned. Usually a negative hit will prompt the caller to fail with ENOENT.

The iswhiteout argument is a pointer to an integer result that will be set to nonzero if the entry represents a whiteout, and zero if it does not. This pointer may be NULL if the caller does not support whiteouts. According to the current scheme for handling whiteouts, if cache_lookup() sets iswhiteout the caller should add ISWHITEOUT to the cn_flags field of its struct componentname.

cache_revlookup(vp, dvp, bpp, bufp)
Scan cache looking for name of directory entry pointing at vp and fill in dvpp. If bufp is non-NULL, also place the name in the buffer which starts at bufp, immediately before bpp, and move bpp backwards to point at the start of it. If the lookup succeeds, the vnode is referenced. Returns 0 on success, -1 on cache miss, positive errno on failure.
cache_enter(dvp, vp, name, namelen, nameiflags)
Add an entry to the cache. The name and namelen arguments specify the name to add to the cache. The dvp argument specifies the directory the name appears in. The vp argument specifies the object to enter in the cache. The nameiflags argument should come from the cn_flags member of struct componentname.

If vp is NULL, a negative cache entry is created, specifying that the entry does not exist in the file system.

cache_purge(vp)
Flush the cache of a particular vnode vp. cache_purge() is called when a vnode is renamed to hide entries that would now be invalid.
cache_purgevfs(mp)
Flush cache of a whole file system mp. cache_purgevfs() is called when file system is unmounted to remove entries that would now be invalid.
namecache_print(vp, func)
Print all entries of the name cache. func is the printf(9) format. namecache_print() is only defined if the kernel option DDB is compiled into the kernel.

CODE REFERENCES

The name lookup cache is implemented within the file sys/kern/vfs_cache.c.

HISTORY

The name lookup cache first appeared in 4.2BSD.

AUTHORS

The original name lookup cache was written by Robert Elz.
February 7, 2014 NetBSD 7.0