On Sun, May 27, 2018 at 04:37:52PM +0200, Klemens Nanni wrote:
> Spotted and reported on IRC by Georg Bege <[email protected]>,
> vfs_cache(9) lacks way behind beck's "Namecache revamp" from 2009.
>
> This diff syncs the manual with sys/sys/namei.h and sys/kern/vfs_cache.c:
> I went through it and checked the APIs, this seems fine to me for now
> but since I'm not all too familiar with VFS yet, feedback is welcome.
>
> Maybe we want to sync descriptions in the manual a bit more with code
> comments to reduce differences in two places describing the same thing?
I have OKs from visa and jmc for the initial diff already; here's an
updated one that syncs other manuals as well.
Syncing descriptions and comments turned out to be much more churn than
useful changes. One thing I stripped though is the note about too long
names in vfs_lookup()'s comment; that bit is already mentioned around
NAMECACHE_MAXLEN's definition.
For struct vnode, I copied it as is to avoid picking around whitespace
changes so we end up with the exact same struct in both places.
There might be more.
Feedback? OK?
Index: sys/kern/vfs_cache.c
===================================================================
RCS file: /cvs/src/sys/kern/vfs_cache.c,v
retrieving revision 1.56
diff -u -p -r1.56 vfs_cache.c
--- sys/kern/vfs_cache.c 27 May 2018 06:02:14 -0000 1.56
+++ sys/kern/vfs_cache.c 1 Jun 2018 11:32:32 -0000
@@ -128,9 +128,7 @@ cache_zap(struct namecache *ncp)
}
/*
- * Look for a name in the cache. We don't do this if the segment name is
- * long, simply so the cache can avoid holding long names (which would
- * either waste space, or add greatly to the complexity).
+ * Look for a name in the cache.
* dvp points to the directory to search. The componentname cnp holds
* the information on the entry being sought, such as its length
* and its name. If the lookup succeeds, vpp is set to point to the vnode
Index: share/man/man9/VOP_LOOKUP.9
===================================================================
RCS file: /cvs/src/share/man/man9/VOP_LOOKUP.9,v
retrieving revision 1.41
diff -u -p -r1.41 VOP_LOOKUP.9
--- share/man/man9/VOP_LOOKUP.9 28 Apr 2018 03:13:04 -0000 1.41
+++ share/man/man9/VOP_LOOKUP.9 1 Jun 2018 12:04:24 -0000
@@ -327,7 +327,7 @@ struct vattr {
uid_t va_uid; /* owner user id */
gid_t va_gid; /* owner group id */
long va_fsid; /* file system id */
- long va_fileid; /* file id */
+ u_quad_t va_fileid; /* file id */
u_quad_t va_size; /* file size in bytes */
long va_blocksize; /* blocksize preferred for i/o */
struct timespec va_atime; /* time of last access */
Index: share/man/man9/vfs_cache.9
===================================================================
RCS file: /cvs/src/share/man/man9/vfs_cache.9,v
retrieving revision 1.3
diff -u -p -r1.3 vfs_cache.9
--- share/man/man9/vfs_cache.9 31 May 2007 19:20:01 -0000 1.3
+++ share/man/man9/vfs_cache.9 1 Jun 2018 11:32:31 -0000
@@ -37,15 +37,16 @@ recently looked-up file name translation
Entries in this cache have the following definition:
.Bd -literal
struct namecache {
- LIST_ENTRY(namecache) nc_hash; /* hash chain */
- LIST_ENTRY(namecache) nc_vhash; /* (reverse) dir hash chain */
- TAILQ_ENTRY(namecache) nc_lru; /* LRU chain */
+ TAILQ_ENTRY(namecache) nc_lru; /* Regular Entry LRU chain */
+ TAILQ_ENTRY(namecache) nc_neg; /* Negative Entry LRU chain */
+ RBT_ENTRY(namecache) n_rbcache; /* Namecache rb tree from vnode */
+ TAILQ_ENTRY(namecache) nc_me; /* ncp's referring to me */
struct vnode *nc_dvp; /* vnode of parent of name */
u_long nc_dvpid; /* capability number of nc_dvp */
struct vnode *nc_vp; /* vnode the name refers to */
u_long nc_vpid; /* capability number of nc_vp */
char nc_nlen; /* length of name */
- char nc_name[NCHNAMLEN]; /* segment name */
+ char nc_name[NAMECACHE_MAXLEN]; /* segment name */
};
.Ed
.Pp
@@ -55,7 +56,7 @@ Negative caching is also performed so th
names of files that do not exist do not result in expensive lookups.
.Pp
File names with length longer than
-.Dv NCHNAMLEN
+.Dv NAMECACHE_MAXLEN
are not cached to simplify lookups and to save space.
Such names are rare and are generally not worth caching.
.Pp
@@ -169,7 +170,8 @@ API is implemented in the file
.Xr vmstat 8 ,
.Xr namei 9 ,
.Xr vfs 9 ,
-.Xr vnode 9
+.Xr vnode 9 ,
+.Xr VOP_LOOKUP 9
.Sh HISTORY
The
.Nm
Index: share/man/man9/vnode.9
===================================================================
RCS file: /cvs/src/share/man/man9/vnode.9,v
retrieving revision 1.28
diff -u -p -r1.28 vnode.9
--- share/man/man9/vnode.9 18 Jul 2011 12:03:45 -0000 1.28
+++ share/man/man9/vnode.9 1 Jun 2018 12:15:40 -0000
@@ -66,41 +66,45 @@ provided by the VFS to create and manage
The definition of a vnode is as follows:
.Bd -literal
struct vnode {
- struct uvm_vnode v_uvm; /* uvm(9) data */
- int (**v_op)(void *); /* vnode operations vector */
- enum vtype v_type; /* vnode type */
- u_int v_flag; /* vnode flags (see below) */
- u_int v_usecount; /* reference count of users */
- u_int v_writecount; /* reference count of writers */
+ struct uvm_vnode *v_uvm; /* uvm data */
+ struct vops *v_op; /* vnode operations vector */
+ enum vtype v_type; /* vnode type */
+ enum vtagtype v_tag; /* type of underlying data */
+ u_int v_flag; /* vnode flags (see below) */
+ u_int v_usecount; /* reference count of users */
+ /* reference count of writers */
+ u_int v_writecount;
/* Flags that can be read/written in interrupts */
- u_int v_bioflag; /* flags used by intr handlers */
- u_int v_holdcnt; /* buffer references */
- u_int v_id; /* capability identifier */
- struct mount *v_mount; /* ptr to vfs we are in */
- TAILQ_ENTRY(vnode) v_freelist; /* vnode freelist */
- LIST_ENTRY(vnode) v_mntvnodes; /* vnodes for mount point */
- struct buflists v_cleanblkhd; /* clean blocklist head */
- struct buflists v_dirtyblkhd; /* dirty blocklist head */
- u_int v_numoutput; /* num of writes in progress */
- LIST_ENTRY(vnode) v_synclist; /* vnode with dirty buffers */
+ u_int v_bioflag;
+ u_int v_holdcnt; /* buffer references */
+ u_int v_id; /* capability identifier */
+ u_int v_inflight;
+ struct mount *v_mount; /* ptr to vfs we are in */
+ TAILQ_ENTRY(vnode) v_freelist; /* vnode freelist */
+ LIST_ENTRY(vnode) v_mntvnodes; /* vnodes for mount point */
+ struct buf_rb_bufs v_bufs_tree; /* lookup of all bufs */
+ struct buflists v_cleanblkhd; /* clean blocklist head */
+ struct buflists v_dirtyblkhd; /* dirty blocklist head */
+ u_int v_numoutput; /* num of writes in progress */
+ LIST_ENTRY(vnode) v_synclist; /* vnode with dirty buffers */
union {
- struct mount *vu_mountedhere;/* ptr to mounted vfs (VDIR) */
- struct socket *vu_socket; /* UNIX IPC (VSOCK) */
- struct specinfo *vu_specinfo; /* device (VCHR, VBLK) */
- struct fifoinfo *vu_fifoinfo; /* fifo (VFIFO) */
+ struct mount *vu_mountedhere;/* ptr to mounted vfs (VDIR) */
+ struct socket *vu_socket; /* unix ipc (VSOCK) */
+ struct specinfo *vu_specinfo; /* device (VCHR, VBLK) */
+ struct fifoinfo *vu_fifoinfo; /* fifo (VFIFO) */
} v_un;
- enum vtagtype v_tag; /* type of underlying data */
- void *v_data; /* private data for fs */
- struct {
- struct simplelock vsi_lock; /* lock to protect below */
- struct selinfo vsi_selinfo; /* identity of poller(s) */
- } v_selectinfo;
+ /* VFS namecache */
+ struct namecache_rb_cache v_nc_tree;
+ TAILQ_HEAD(, namecache) v_cache_dst; /* cache entries to us */
+
+ void *v_data; /* private data for fs */
+ struct selinfo v_selectinfo; /* identity of poller(s) */
};
-#define v_mountedhere v_un.vu_mountedhere
-#define v_socket v_un.vu_socket
-#define v_specinfo v_un.vu_specinfo
-#define v_fifoinfo v_un.vu_fifoinfo
+#define v_mountedhere v_un.vu_mountedhere
+#define v_socket v_un.vu_socket
+#define v_specinfo v_un.vu_specinfo
+#define v_fifoinfo v_un.vu_fifoinfo
.Ed
.Ss Vnode life cycle
When a client of the VFS requests a new vnode, the vnode allocation
