d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
7533652025
freebsd-dev/sys/kern/vfs_hash.c

156 lines
4.2 KiB
C
Raw Normal View History

Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
/*-
* Copyright (c) 2005 Poul-Henning Kamp
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
Add mnt_hashseed to struct mount and initialize it witn PRNG bits, use it to get better hashing in vfs_hash. In case of an insert collision in vfs_hash_insert(), put the loosing vnode on a special list so that vfs_hash_remove() can just assume that it is on a list. Drop the VI_HASHED flag.
2005-03-16 07:35:06 +00:00
#include <sys/mount.h>
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
#include <sys/vnode.h>
Normalize a significant number of kernel malloc type names: - Prefer '_' to ' ', as it results in more easily parsed results in memory monitoring tools such as vmstat. - Remove punctuation that is incompatible with using memory type names as file names, such as '/' characters. - Disambiguate some collisions by adding subsystem prefixes to some memory types. - Generally prefer lower case to upper case. - If the same type is defined in multiple architecture directories, attempt to use the same name in additional cases. Not all instances were caught in this change, so more work is required to finish this conversion. Similar changes are required for UMA zone names.
2005-10-31 15:41:29 +00:00
static MALLOC_DEFINE(M_VFS_HASH, "vfs_hash", "VFS hash table");
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
Add mnt_hashseed to struct mount and initialize it witn PRNG bits, use it to get better hashing in vfs_hash. In case of an insert collision in vfs_hash_insert(), put the loosing vnode on a special list so that vfs_hash_remove() can just assume that it is on a list. Drop the VI_HASHED flag.
2005-03-16 07:35:06 +00:00
static LIST_HEAD(vfs_hash_head, vnode) *vfs_hash_tbl;
static LIST_HEAD(,vnode) vfs_hash_side;
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
static u_long vfs_hash_mask;
static struct mtx vfs_hash_mtx;
static void
vfs_hashinit(void *dummy __unused)
{
vfs_hash_tbl = hashinit(desiredvnodes, M_VFS_HASH, &vfs_hash_mask);
mtx_init(&vfs_hash_mtx, "vfs hash", NULL, MTX_DEF);
Add mnt_hashseed to struct mount and initialize it witn PRNG bits, use it to get better hashing in vfs_hash. In case of an insert collision in vfs_hash_insert(), put the loosing vnode on a special list so that vfs_hash_remove() can just assume that it is on a list. Drop the VI_HASHED flag.
2005-03-16 07:35:06 +00:00
LIST_INIT(&vfs_hash_side);
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
}
/* Must be SI_ORDER_SECOND so desiredvnodes is available */
SYSINIT(vfs_hash, SI_SUB_VFS, SI_ORDER_SECOND, vfs_hashinit, NULL)
Add mnt_hashseed to struct mount and initialize it witn PRNG bits, use it to get better hashing in vfs_hash. In case of an insert collision in vfs_hash_insert(), put the loosing vnode on a special list so that vfs_hash_remove() can just assume that it is on a list. Drop the VI_HASHED flag.
2005-03-16 07:35:06 +00:00
static struct vfs_hash_head *
In vfs_hash_get(): mount point should never be changed so explicitly constify the mp parameter. Reviewed by: phk
2006-04-18 08:05:08 +00:00
vfs_hash_index(const struct mount *mp, u_int hash)
Add mnt_hashseed to struct mount and initialize it witn PRNG bits, use it to get better hashing in vfs_hash. In case of an insert collision in vfs_hash_insert(), put the loosing vnode on a special list so that vfs_hash_remove() can just assume that it is on a list. Drop the VI_HASHED flag.
2005-03-16 07:35:06 +00:00
{
return(&vfs_hash_tbl[(hash + mp->mnt_hashseed) & vfs_hash_mask]);
}
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
int
In vfs_hash_get(): mount point should never be changed so explicitly constify the mp parameter. Reviewed by: phk
2006-04-18 08:05:08 +00:00
vfs_hash_get(const struct mount *mp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg)
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
{
struct vnode *vp;
int error;
while (1) {
mtx_lock(&vfs_hash_mtx);
Add mnt_hashseed to struct mount and initialize it witn PRNG bits, use it to get better hashing in vfs_hash. In case of an insert collision in vfs_hash_insert(), put the loosing vnode on a special list so that vfs_hash_remove() can just assume that it is on a list. Drop the VI_HASHED flag.
2005-03-16 07:35:06 +00:00
LIST_FOREACH(vp, vfs_hash_index(mp, hash), v_hashlist) {
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
if (vp->v_hash != hash)
continue;
if (vp->v_mount != mp)
continue;
Add two arguments to the vfs_hash() KPI so that filesystems which do not have unique hashes (NFS) can also use it.
2005-03-16 11:20:51 +00:00
if (fn != NULL && fn(vp, arg))
continue;
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
VI_LOCK(vp);
mtx_unlock(&vfs_hash_mtx);
error = vget(vp, flags | LK_INTERLOCK, td);
Don't retry when vget() returns ENOENT in the nonblocking case due to the vnode being doomed. It causes a livelock.
2005-09-12 01:48:57 +00:00
if (error == ENOENT && (flags & LK_NOWAIT) == 0)
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
break;
if (error)
return (error);
*vpp = vp;
return (0);
}
if (vp == NULL) {
mtx_unlock(&vfs_hash_mtx);
*vpp = NULL;
return (0);
}
}
}
void
vfs_hash_remove(struct vnode *vp)
{
mtx_lock(&vfs_hash_mtx);
LIST_REMOVE(vp, v_hashlist);
mtx_unlock(&vfs_hash_mtx);
}
int
Add two arguments to the vfs_hash() KPI so that filesystems which do not have unique hashes (NFS) can also use it.
2005-03-16 11:20:51 +00:00
vfs_hash_insert(struct vnode *vp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg)
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
{
struct vnode *vp2;
int error;
Simplify the vfs_hash calling convention.
2005-03-15 08:07:07 +00:00
*vpp = NULL;
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
while (1) {
mtx_lock(&vfs_hash_mtx);
LIST_FOREACH(vp2,
Add mnt_hashseed to struct mount and initialize it witn PRNG bits, use it to get better hashing in vfs_hash. In case of an insert collision in vfs_hash_insert(), put the loosing vnode on a special list so that vfs_hash_remove() can just assume that it is on a list. Drop the VI_HASHED flag.
2005-03-16 07:35:06 +00:00
vfs_hash_index(vp->v_mount, hash), v_hashlist) {
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
if (vp2->v_hash != hash)
continue;
if (vp2->v_mount != vp->v_mount)
continue;
Fix a bad copy&paste mistake I made. Spotted by: truckman
2005-03-18 06:01:21 +00:00
if (fn != NULL && fn(vp2, arg))
Add two arguments to the vfs_hash() KPI so that filesystems which do not have unique hashes (NFS) can also use it.
2005-03-16 11:20:51 +00:00
continue;
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
VI_LOCK(vp2);
mtx_unlock(&vfs_hash_mtx);
error = vget(vp2, flags | LK_INTERLOCK, td);
Don't retry when vget() returns ENOENT in the nonblocking case due to the vnode being doomed. It causes a livelock.
2005-09-12 01:48:57 +00:00
if (error == ENOENT && (flags & LK_NOWAIT) == 0)
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
break;
Add mnt_hashseed to struct mount and initialize it witn PRNG bits, use it to get better hashing in vfs_hash. In case of an insert collision in vfs_hash_insert(), put the loosing vnode on a special list so that vfs_hash_remove() can just assume that it is on a list. Drop the VI_HASHED flag.
2005-03-16 07:35:06 +00:00
mtx_lock(&vfs_hash_mtx);
LIST_INSERT_HEAD(&vfs_hash_side, vp, v_hashlist);
mtx_unlock(&vfs_hash_mtx);
Improve the vfs_hash() API: vput() the unneeded vnode centrally to avoid replicating the vput in all the filesystems.
2005-03-15 20:00:03 +00:00
vput(vp);
if (!error)
*vpp = vp2;
return (error);
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
}
if (vp2 == NULL)
break;
}
vp->v_hash = hash;
Add mnt_hashseed to struct mount and initialize it witn PRNG bits, use it to get better hashing in vfs_hash. In case of an insert collision in vfs_hash_insert(), put the loosing vnode on a special list so that vfs_hash_remove() can just assume that it is on a list. Drop the VI_HASHED flag.
2005-03-16 07:35:06 +00:00
LIST_INSERT_HEAD(vfs_hash_index(vp->v_mount, hash), vp, v_hashlist);
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
mtx_unlock(&vfs_hash_mtx);
return (0);
}
void
vfs_hash_rehash(struct vnode *vp, u_int hash)
{
mtx_lock(&vfs_hash_mtx);
LIST_REMOVE(vp, v_hashlist);
Fix bug in vfs_hash_rehash(): use correct bucket. This only affected msdosfs which is broken in other ways too.
2005-04-07 07:54:08 +00:00
LIST_INSERT_HEAD(vfs_hash_index(vp->v_mount, hash), vp, v_hashlist);
Currently (almost) all filesystems maintain a local inode hash table to get from (mount + inode) to vnode. These tables are mostly copy&pasted from UFS, sized based on desiredvnodes and therefore quite large (128K-512K). Several filesystems are buggy enough that they allocate the hash table even before they know if they will ever be used or not. Add "vfs_hash", a system wide hash table, which will replace all the per-filesystem hash-tables. The fields we add to struct vnode will more or less be saved in the respective filesystems inodes. Having one central implementation will save code and will allow us to justify the complexity of code to dynamically (re)size the hash at a later point.
2005-03-14 10:01:29 +00:00
vp->v_hash = hash;
mtx_unlock(&vfs_hash_mtx);
}
Reference in New Issue Copy Permalink