Manual browser: puffs_flush(3)

PUFFS_FLUSH(3) Library Functions Manual PUFFS_FLUSH(3)


puffs_flushpuffs kernel cache flushing and invalidation routines


puffs Convenience Library (libpuffs, -lpuffs)


#include <puffs.h>

puffs_inval_namecache_dir(struct puffs_usermount *pu, puffs_cookie_t cookie);

puffs_inval_namecache_all(struct puffs_usermount *pu);

puffs_inval_pagecache_node(struct puffs_usermount *pu, puffs_cookie_t cookie);

puffs_inval_pagecache_node_range(struct puffs_usermount *pu, puffs_cookie_t cookie, off_t start, off_t end);

puffs_flush_pagecache_node(struct puffs_usermount *pu, puffs_cookie_t cookie);

puffs_flush_pagecache_node_range(struct puffs_usermount *pu, puffs_cookie_t cookie, off_t start, off_t end);


These routines are used inform the kernel that any information it might have cached is no longer valid. puffs_inval_namecache_dir() invalidates the name cache for a given directory. The argument cookie should describe an existing and valid directory cookie for the file system. Similarly, puffs_inval_namecache_all() invalidates the name cache for the entire file system (this routine might go away).

The cached pages (file contents) for a regular file described by cookie are invalidated using puffs_inval_pagecache_node(). A specific range can be invalidated using puffs_inval_pagecache_node_range() for a platform specific page level granularity. The offset start will be truncated to a page boundary while end will be rounded up to the next page boundary. As a special case, specifying 0 as end will invalidate all contents from start to the end of the file.

It is especially important to note that these routines will not only invalidate data in the "read cache", but also data in the "write back" cache (conceptually speaking; in reality they are the same cache), which has not yet been flushed to the file server. Therefore any unflushed data will be lost.

The counterparts of the invalidation routines are the flushing routines puffs_flush_pagecache_node() and puffs_flush_pagecache_node_range(), which force unwritten data from the kernel page cache to be written. For the flush range version, the same range rules as with the invalidation routine apply. The data is flushed asynchronously, i.e. if the routine returns successfully, all the caller knows is that the data has been queued for writing.


April 7, 2007 NetBSD 7.0