Module Name: src
Committed By: ad
Date: Sat Apr 4 20:49:31 UTC 2020
Modified Files:
src/sys/fs/cd9660: cd9660_lookup.c cd9660_vfsops.c
src/sys/fs/msdosfs: msdosfs_lookup.c msdosfs_vfsops.c
src/sys/fs/tmpfs: tmpfs_subr.c tmpfs_vfsops.c
src/sys/kern: vfs_cache.c vfs_getcwd.c vfs_lookup.c vfs_syscalls.c
vfs_vnode.c
src/sys/miscfs/genfs: layer_vnops.c
src/sys/miscfs/nullfs: null_vfsops.c
src/sys/miscfs/procfs: procfs_vfsops.c
src/sys/sys: fstypes.h namei.src
src/sys/ufs/chfs: chfs_vnops.c
src/sys/ufs/ext2fs: ext2fs_lookup.c ext2fs_vfsops.c
src/sys/ufs/ffs: ffs_vfsops.c
src/sys/ufs/lfs: lfs_vfsops.c ulfs_lookup.c
src/sys/ufs/ufs: ufs_lookup.c ufs_vnops.c
Log Message:
Merge the remaining changes from the ad-namecache branch, affecting namei()
and getcwd():
- push vnode locking back as far as possible.
- do most lookups directly in the namecache, avoiding vnode locks & refs.
- don't block new refs to vnodes across VOP_INACTIVE().
- get shared locks for VOP_LOOKUP() if the file system supports it.
- correct lock types for VOP_ACCESS() / VOP_GETATTR() in a few places.
Possible future enhancements:
- make the lookups lockless.
- support dotdot lookups by being lockless and inferring absence of chroot.
- maybe make it work for layered file systems.
- avoid vnode references at the root & cwd.
To generate a diff of this commit:
cvs rdiff -u -r1.30 -r1.31 src/sys/fs/cd9660/cd9660_lookup.c
cvs rdiff -u -r1.95 -r1.96 src/sys/fs/cd9660/cd9660_vfsops.c
cvs rdiff -u -r1.35 -r1.36 src/sys/fs/msdosfs/msdosfs_lookup.c
cvs rdiff -u -r1.133 -r1.134 src/sys/fs/msdosfs/msdosfs_vfsops.c
cvs rdiff -u -r1.107 -r1.108 src/sys/fs/tmpfs/tmpfs_subr.c
cvs rdiff -u -r1.76 -r1.77 src/sys/fs/tmpfs/tmpfs_vfsops.c
cvs rdiff -u -r1.136 -r1.137 src/sys/kern/vfs_cache.c
cvs rdiff -u -r1.56 -r1.57 src/sys/kern/vfs_getcwd.c
cvs rdiff -u -r1.214 -r1.215 src/sys/kern/vfs_lookup.c
cvs rdiff -u -r1.544 -r1.545 src/sys/kern/vfs_syscalls.c
cvs rdiff -u -r1.116 -r1.117 src/sys/kern/vfs_vnode.c
cvs rdiff -u -r1.68 -r1.69 src/sys/miscfs/genfs/layer_vnops.c
cvs rdiff -u -r1.97 -r1.98 src/sys/miscfs/nullfs/null_vfsops.c
cvs rdiff -u -r1.103 -r1.104 src/sys/miscfs/procfs/procfs_vfsops.c
cvs rdiff -u -r1.37 -r1.38 src/sys/sys/fstypes.h
cvs rdiff -u -r1.52 -r1.53 src/sys/sys/namei.src
cvs rdiff -u -r1.36 -r1.37 src/sys/ufs/chfs/chfs_vnops.c
cvs rdiff -u -r1.89 -r1.90 src/sys/ufs/ext2fs/ext2fs_lookup.c
cvs rdiff -u -r1.217 -r1.218 src/sys/ufs/ext2fs/ext2fs_vfsops.c
cvs rdiff -u -r1.366 -r1.367 src/sys/ufs/ffs/ffs_vfsops.c
cvs rdiff -u -r1.377 -r1.378 src/sys/ufs/lfs/lfs_vfsops.c
cvs rdiff -u -r1.42 -r1.43 src/sys/ufs/lfs/ulfs_lookup.c
cvs rdiff -u -r1.151 -r1.152 src/sys/ufs/ufs/ufs_lookup.c
cvs rdiff -u -r1.249 -r1.250 src/sys/ufs/ufs/ufs_vnops.c
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/sys/fs/cd9660/cd9660_lookup.c
diff -u src/sys/fs/cd9660/cd9660_lookup.c:1.30 src/sys/fs/cd9660/cd9660_lookup.c:1.31
--- src/sys/fs/cd9660/cd9660_lookup.c:1.30 Sat Mar 28 19:24:05 2015
+++ src/sys/fs/cd9660/cd9660_lookup.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: cd9660_lookup.c,v 1.30 2015/03/28 19:24:05 maxv Exp $ */
+/* $NetBSD: cd9660_lookup.c,v 1.31 2020/04/04 20:49:30 ad Exp $ */
/*-
* Copyright (c) 1989, 1993, 1994
@@ -39,7 +39,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: cd9660_lookup.c,v 1.30 2015/03/28 19:24:05 maxv Exp $");
+__KERNEL_RCSID(0, "$NetBSD: cd9660_lookup.c,v 1.31 2020/04/04 20:49:30 ad Exp $");
#include <sys/param.h>
#include <sys/namei.h>
@@ -152,6 +152,9 @@ cd9660_lookup(void *v)
cnp->cn_nameiop, cnp->cn_flags, NULL, vpp)) {
return *vpp == NULLVP ? ENOENT : 0;
}
+ /* May need to restart the lookup with an exclusive lock. */
+ if (VOP_ISLOCKED(vdp) != LK_EXCLUSIVE)
+ return ENOLCK;
len = cnp->cn_namelen;
name = cnp->cn_nameptr;
Index: src/sys/fs/cd9660/cd9660_vfsops.c
diff -u src/sys/fs/cd9660/cd9660_vfsops.c:1.95 src/sys/fs/cd9660/cd9660_vfsops.c:1.96
--- src/sys/fs/cd9660/cd9660_vfsops.c:1.95 Mon Mar 16 21:20:09 2020
+++ src/sys/fs/cd9660/cd9660_vfsops.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: cd9660_vfsops.c,v 1.95 2020/03/16 21:20:09 pgoyette Exp $ */
+/* $NetBSD: cd9660_vfsops.c,v 1.96 2020/04/04 20:49:30 ad Exp $ */
/*-
* Copyright (c) 1994
@@ -37,7 +37,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: cd9660_vfsops.c,v 1.95 2020/03/16 21:20:09 pgoyette Exp $");
+__KERNEL_RCSID(0, "$NetBSD: cd9660_vfsops.c,v 1.96 2020/04/04 20:49:30 ad Exp $");
#if defined(_KERNEL_OPT)
#include "opt_compat_netbsd.h"
@@ -446,7 +446,7 @@ iso_mountfs(struct vnode *devvp, struct
mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
mp->mnt_stat.f_namemax = ISO_MAXNAMLEN;
mp->mnt_flag |= MNT_LOCAL;
- mp->mnt_iflag |= IMNT_MPSAFE;
+ mp->mnt_iflag |= IMNT_MPSAFE | IMNT_SHRLOOKUP;
mp->mnt_dev_bshift = iso_bsize;
mp->mnt_fs_bshift = isomp->im_bshift;
isomp->im_mountp = mp;
Index: src/sys/fs/msdosfs/msdosfs_lookup.c
diff -u src/sys/fs/msdosfs/msdosfs_lookup.c:1.35 src/sys/fs/msdosfs/msdosfs_lookup.c:1.36
--- src/sys/fs/msdosfs/msdosfs_lookup.c:1.35 Sat Jan 30 09:59:27 2016
+++ src/sys/fs/msdosfs/msdosfs_lookup.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: msdosfs_lookup.c,v 1.35 2016/01/30 09:59:27 mlelstv Exp $ */
+/* $NetBSD: msdosfs_lookup.c,v 1.36 2020/04/04 20:49:30 ad Exp $ */
/*-
* Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
@@ -52,7 +52,7 @@
#endif
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: msdosfs_lookup.c,v 1.35 2016/01/30 09:59:27 mlelstv Exp $");
+__KERNEL_RCSID(0, "$NetBSD: msdosfs_lookup.c,v 1.36 2020/04/04 20:49:30 ad Exp $");
#include <sys/param.h>
@@ -161,6 +161,10 @@ msdosfs_lookup(void *v)
return *vpp == NULLVP ? ENOENT: 0;
}
+ /* May need to restart the lookup with an exclusive lock. */
+ if (VOP_ISLOCKED(vdp) != LK_EXCLUSIVE)
+ return ENOLCK;
+
/*
* If they are going after the . or .. entry in the root directory,
* they won't find it. DOS filesystems don't have them in the root
Index: src/sys/fs/msdosfs/msdosfs_vfsops.c
diff -u src/sys/fs/msdosfs/msdosfs_vfsops.c:1.133 src/sys/fs/msdosfs/msdosfs_vfsops.c:1.134
--- src/sys/fs/msdosfs/msdosfs_vfsops.c:1.133 Mon Mar 16 21:20:10 2020
+++ src/sys/fs/msdosfs/msdosfs_vfsops.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: msdosfs_vfsops.c,v 1.133 2020/03/16 21:20:10 pgoyette Exp $ */
+/* $NetBSD: msdosfs_vfsops.c,v 1.134 2020/04/04 20:49:30 ad Exp $ */
/*-
* Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
@@ -48,7 +48,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: msdosfs_vfsops.c,v 1.133 2020/03/16 21:20:10 pgoyette Exp $");
+__KERNEL_RCSID(0, "$NetBSD: msdosfs_vfsops.c,v 1.134 2020/04/04 20:49:30 ad Exp $");
#if defined(_KERNEL_OPT)
#include "opt_compat_netbsd.h"
@@ -868,6 +868,7 @@ msdosfs_mountfs(struct vnode *devvp, str
mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
mp->mnt_stat.f_namemax = MSDOSFS_NAMEMAX(pmp);
mp->mnt_flag |= MNT_LOCAL;
+ mp->mnt_iflag |= IMNT_SHRLOOKUP;
mp->mnt_dev_bshift = pmp->pm_bnshift;
mp->mnt_fs_bshift = pmp->pm_cnshift;
Index: src/sys/fs/tmpfs/tmpfs_subr.c
diff -u src/sys/fs/tmpfs/tmpfs_subr.c:1.107 src/sys/fs/tmpfs/tmpfs_subr.c:1.108
--- src/sys/fs/tmpfs/tmpfs_subr.c:1.107 Sat Mar 14 13:37:49 2020
+++ src/sys/fs/tmpfs/tmpfs_subr.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: tmpfs_subr.c,v 1.107 2020/03/14 13:37:49 ad Exp $ */
+/* $NetBSD: tmpfs_subr.c,v 1.108 2020/04/04 20:49:30 ad Exp $ */
/*
* Copyright (c) 2005-2013 The NetBSD Foundation, Inc.
@@ -73,7 +73,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: tmpfs_subr.c,v 1.107 2020/03/14 13:37:49 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: tmpfs_subr.c,v 1.108 2020/04/04 20:49:30 ad Exp $");
#include <sys/param.h>
#include <sys/cprng.h>
@@ -147,6 +147,8 @@ tmpfs_init_vnode(struct vnode *vp, tmpfs
vp->v_data = node;
node->tn_vnode = vp;
uvm_vnp_setsize(vp, node->tn_size);
+ KASSERT(node->tn_mode != VNOVAL);
+ cache_enter_id(vp, node->tn_mode, node->tn_uid, node->tn_gid);
}
/*
@@ -1039,6 +1041,7 @@ tmpfs_chmod(vnode_t *vp, mode_t mode, ka
node->tn_mode = (mode & ALLPERMS);
tmpfs_update(vp, TMPFS_UPDATE_CTIME);
VN_KNOTE(vp, NOTE_ATTRIB);
+ cache_enter_id(vp, node->tn_mode, node->tn_uid, node->tn_gid);
return 0;
}
@@ -1083,6 +1086,7 @@ tmpfs_chown(vnode_t *vp, uid_t uid, gid_
node->tn_gid = gid;
tmpfs_update(vp, TMPFS_UPDATE_CTIME);
VN_KNOTE(vp, NOTE_ATTRIB);
+ cache_enter_id(vp, node->tn_mode, node->tn_uid, node->tn_gid);
return 0;
}
Index: src/sys/fs/tmpfs/tmpfs_vfsops.c
diff -u src/sys/fs/tmpfs/tmpfs_vfsops.c:1.76 src/sys/fs/tmpfs/tmpfs_vfsops.c:1.77
--- src/sys/fs/tmpfs/tmpfs_vfsops.c:1.76 Fri Jan 17 20:08:08 2020
+++ src/sys/fs/tmpfs/tmpfs_vfsops.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: tmpfs_vfsops.c,v 1.76 2020/01/17 20:08:08 ad Exp $ */
+/* $NetBSD: tmpfs_vfsops.c,v 1.77 2020/04/04 20:49:30 ad Exp $ */
/*
* Copyright (c) 2005, 2006, 2007 The NetBSD Foundation, Inc.
@@ -42,7 +42,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: tmpfs_vfsops.c,v 1.76 2020/01/17 20:08:08 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: tmpfs_vfsops.c,v 1.77 2020/04/04 20:49:30 ad Exp $");
#include <sys/param.h>
#include <sys/atomic.h>
@@ -182,7 +182,8 @@ tmpfs_mount(struct mount *mp, const char
mp->mnt_stat.f_namemax = TMPFS_MAXNAMLEN;
mp->mnt_fs_bshift = PAGE_SHIFT;
mp->mnt_dev_bshift = DEV_BSHIFT;
- mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO;
+ mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO | IMNT_SHRLOOKUP |
+ IMNT_NCLOOKUP;
vfs_getnewfsid(mp);
/* Allocate the tmpfs mount structure and fill it. */
Index: src/sys/kern/vfs_cache.c
diff -u src/sys/kern/vfs_cache.c:1.136 src/sys/kern/vfs_cache.c:1.137
--- src/sys/kern/vfs_cache.c:1.136 Mon Mar 30 19:15:28 2020
+++ src/sys/kern/vfs_cache.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: vfs_cache.c,v 1.136 2020/03/30 19:15:28 ad Exp $ */
+/* $NetBSD: vfs_cache.c,v 1.137 2020/04/04 20:49:30 ad Exp $ */
/*-
* Copyright (c) 2008, 2019, 2020 The NetBSD Foundation, Inc.
@@ -172,7 +172,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: vfs_cache.c,v 1.136 2020/03/30 19:15:28 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: vfs_cache.c,v 1.137 2020/04/04 20:49:30 ad Exp $");
#define __NAMECACHE_PRIVATE
#ifdef _KERNEL_OPT
@@ -624,7 +624,6 @@ cache_lookup_raw(struct vnode *dvp, cons
* names in the cache. The node locks are chained along the way: a parent's
* lock is not dropped until the child's is acquired.
*/
-#ifdef notyet
bool
cache_lookup_linked(struct vnode *dvp, const char *name, size_t namelen,
struct vnode **vn_ret, krwlock_t **plock,
@@ -721,7 +720,6 @@ cache_lookup_linked(struct vnode *dvp, c
*vn_ret = ncp->nc_vp;
return true;
}
-#endif /* notyet */
/*
* Scan cache looking for name of directory entry pointing at vp.
@@ -967,7 +965,6 @@ cache_enter_id(struct vnode *vp, mode_t
* information, missing some updates, so always check the mount flag
* instead of looking for !VNOVAL.
*/
-#ifdef notyet
bool
cache_have_id(struct vnode *vp)
{
@@ -982,7 +979,6 @@ cache_have_id(struct vnode *vp)
return false;
}
}
-#endif /* notyet */
/*
* Name cache initialization, from vfs_init() when the system is booting.
Index: src/sys/kern/vfs_getcwd.c
diff -u src/sys/kern/vfs_getcwd.c:1.56 src/sys/kern/vfs_getcwd.c:1.57
--- src/sys/kern/vfs_getcwd.c:1.56 Sun Mar 22 14:38:37 2020
+++ src/sys/kern/vfs_getcwd.c Sat Apr 4 20:49:30 2020
@@ -1,7 +1,7 @@
-/* $NetBSD: vfs_getcwd.c,v 1.56 2020/03/22 14:38:37 ad Exp $ */
+/* $NetBSD: vfs_getcwd.c,v 1.57 2020/04/04 20:49:30 ad Exp $ */
/*-
- * Copyright (c) 1999 The NetBSD Foundation, Inc.
+ * Copyright (c) 1999, 2020 The NetBSD Foundation, Inc.
* All rights reserved.
*
* This code is derived from software contributed to The NetBSD Foundation
@@ -30,7 +30,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: vfs_getcwd.c,v 1.56 2020/03/22 14:38:37 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: vfs_getcwd.c,v 1.57 2020/04/04 20:49:30 ad Exp $");
#include <sys/param.h>
#include <sys/systm.h>
@@ -87,7 +87,7 @@ __KERNEL_RCSID(0, "$NetBSD: vfs_getcwd.c
* On exit, *uvpp is either NULL or is a locked vnode reference.
*/
static int
-getcwd_scandir(struct vnode **lvpp, struct vnode **uvpp, char **bpp,
+getcwd_scandir(struct vnode *lvp, struct vnode **uvpp, char **bpp,
char *bufp, struct lwp *l)
{
int error = 0;
@@ -101,12 +101,14 @@ getcwd_scandir(struct vnode **lvpp, stru
ino_t fileno;
struct vattr va;
struct vnode *uvp = NULL;
- struct vnode *lvp = *lvpp;
kauth_cred_t cred = l->l_cred;
struct componentname cn;
int len, reclen;
tries = 0;
+ /* Need exclusive for UFS VOP_GETATTR (itimes) & VOP_LOOKUP. */
+ KASSERT(VOP_ISLOCKED(lvp) == LK_EXCLUSIVE);
+
/*
* If we want the filename, get some info we need while the
* current directory is still locked.
@@ -114,8 +116,7 @@ getcwd_scandir(struct vnode **lvpp, stru
if (bufp != NULL) {
error = VOP_GETATTR(lvp, &va, cred);
if (error) {
- vput(lvp);
- *lvpp = NULL;
+ VOP_UNLOCK(lvp);
*uvpp = NULL;
return error;
}
@@ -134,24 +135,14 @@ getcwd_scandir(struct vnode **lvpp, stru
/* At this point, lvp is locked */
error = VOP_LOOKUP(lvp, uvpp, &cn);
- vput(lvp);
+ VOP_UNLOCK(lvp);
if (error) {
- *lvpp = NULL;
*uvpp = NULL;
return error;
}
uvp = *uvpp;
- /* Now lvp is unlocked, try to lock uvp */
- error = vn_lock(uvp, LK_EXCLUSIVE);
- if (error) {
- *lvpp = NULL;
- *uvpp = NULL;
- return error;
- }
-
/* If we don't care about the pathname, we're done */
if (bufp == NULL) {
- *lvpp = NULL;
return 0;
}
@@ -163,6 +154,14 @@ getcwd_scandir(struct vnode **lvpp, stru
dirbuflen = va.va_blocksize;
dirbuf = kmem_alloc(dirbuflen, KM_SLEEP);
+ /* Now lvp is unlocked, try to lock uvp */
+ error = vn_lock(uvp, LK_SHARED);
+ if (error) {
+ vrele(uvp);
+ *uvpp = NULL;
+ return error;
+ }
+
#if 0
unionread:
#endif
@@ -254,73 +253,21 @@ unionread:
vput(tvp);
vref(uvp);
*uvpp = uvp;
- vn_lock(uvp, LK_EXCLUSIVE | LK_RETRY);
+ vn_lock(uvp, LK_SHARED | LK_RETRY);
goto unionread;
}
#endif
error = ENOENT;
out:
- *lvpp = NULL;
+ VOP_UNLOCK(uvp);
kmem_free(dirbuf, dirbuflen);
return error;
}
/*
- * Look in the vnode-to-name reverse cache to see if
- * we can find things the easy way.
- *
- * XXX vget failure path is untested.
- *
- * On entry, *lvpp is a locked vnode reference.
- * On exit, one of the following is the case:
- * 0) Both *lvpp and *uvpp are NULL and failure is returned.
- * 1) *uvpp is NULL, *lvpp remains locked and -1 is returned (cache miss)
- * 2) *uvpp is a locked vnode reference, *lvpp is vput and NULL'ed
- * and 0 is returned (cache hit)
- */
-
-static int
-getcwd_getcache(struct vnode **lvpp, struct vnode **uvpp, char **bpp,
- char *bufp)
-{
- struct vnode *lvp, *uvp = NULL;
- int error;
-
- lvp = *lvpp;
-
- /*
- * This returns 0 on a cache hit, -1 on a clean cache miss,
- * or an errno on other failure.
- */
- error = cache_revlookup(lvp, uvpp, bpp, bufp, 0, 0);
- if (error) {
- if (error != -1) {
- vput(lvp);
- *lvpp = NULL;
- *uvpp = NULL;
- }
- return error;
- }
- uvp = *uvpp;
-
- /*
- * Since we're going up, we have to release the current lock
- * before we take the parent lock.
- */
-
- VOP_UNLOCK(lvp);
- vn_lock(uvp, LK_EXCLUSIVE | LK_RETRY);
- vrele(lvp);
- *lvpp = NULL;
-
- return error;
-}
-
-/*
* common routine shared by sys___getcwd() and vn_isunder()
*/
-
int
getcwd_common(struct vnode *lvp, struct vnode *rvp, char **bpp, char *bufp,
int limit, int flags, struct lwp *l)
@@ -345,11 +292,10 @@ getcwd_common(struct vnode *lvp, struct
/*
* Error handling invariant:
* Before a `goto out':
- * lvp is either NULL, or locked and held.
- * uvp is either NULL, or locked and held.
+ * lvp is either NULL, or held.
+ * uvp is either NULL, or held.
*/
- vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
if (bufp)
bp = *bpp;
@@ -369,63 +315,93 @@ getcwd_common(struct vnode *lvp, struct
* access check here is optional, depending on
* whether or not caller cares.
*/
- if (flags & GETCWD_CHECK_ACCESS) {
- error = VOP_ACCESS(lvp, perms, cred);
- if (error)
- goto out;
- perms = VEXEC|VREAD;
- }
+ int chkaccess = (flags & GETCWD_CHECK_ACCESS);
+ bool locked = false;
/*
* step up if we're a covered vnode..
+ * check access on the first vnode only.
*/
- while (lvp->v_vflag & VV_ROOT) {
- struct vnode *tvp;
+ if (lvp->v_vflag & VV_ROOT) {
+ vn_lock(lvp, LK_SHARED | LK_RETRY);
+ if (chkaccess) {
+ error = VOP_ACCESS(lvp, perms, cred);
+ if (error) {
+ VOP_UNLOCK(lvp);
+ goto out;
+ }
+ chkaccess = 0;
+ }
+ while (lvp->v_vflag & VV_ROOT) {
+ struct vnode *tvp;
- if (lvp == rvp)
- goto out;
+ if (lvp == rvp) {
+ VOP_UNLOCK(lvp);
+ goto out;
+ }
- tvp = lvp;
- lvp = lvp->v_mount->mnt_vnodecovered;
- vput(tvp);
- /*
- * hodie natus est radici frater
- */
- if (lvp == NULL) {
- error = ENOENT;
- goto out;
+ tvp = lvp->v_mount->mnt_vnodecovered;
+ /*
+ * hodie natus est radici frater
+ */
+ if (tvp == NULL) {
+ VOP_UNLOCK(lvp);
+ error = ENOENT;
+ goto out;
+ }
+ vref(tvp);
+ vput(lvp);
+ lvp = tvp;
+ if (lvp->v_vflag & VV_ROOT)
+ vn_lock(lvp, LK_SHARED | LK_RETRY);
}
- vref(lvp);
- error = vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
- if (error != 0) {
- vrele(lvp);
- lvp = NULL;
+ }
+
+ /* Do we need to check access to the directory? */
+ if (chkaccess && !cache_have_id(lvp)) {
+ /* Need exclusive for UFS VOP_GETATTR (itimes) & VOP_LOOKUP. */
+ vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
+ error = VOP_ACCESS(lvp, perms, cred);
+ if (error) {
+ VOP_UNLOCK(lvp);
goto out;
}
+ chkaccess = 0;
+ locked = true;
}
+
/*
* Look in the name cache; if that fails, look in the
* directory..
*/
- error = getcwd_getcache(&lvp, &uvp, &bp, bufp);
+ error = cache_revlookup(lvp, &uvp, &bp, bufp, chkaccess,
+ perms);
if (error == -1) {
+ if (!locked) {
+ locked = true;
+ vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
+ }
if (lvp->v_type != VDIR) {
+ VOP_UNLOCK(lvp);
error = ENOTDIR;
goto out;
}
- error = getcwd_scandir(&lvp, &uvp, &bp, bufp, l);
+ error = getcwd_scandir(lvp, &uvp, &bp, bufp, l);
+ /* lvp now unlocked */
+ } else if (locked) {
+ VOP_UNLOCK(lvp);
}
if (error)
goto out;
#if DIAGNOSTIC
- if (lvp != NULL)
- panic("getcwd: oops, forgot to null lvp");
if (bufp && (bp <= bufp)) {
panic("getcwd: oops, went back too far");
}
#endif
+ perms = VEXEC | VREAD;
if (bp)
*(--bp) = '/';
+ vrele(lvp);
lvp = uvp;
uvp = NULL;
limit--;
@@ -435,9 +411,9 @@ out:
if (bpp)
*bpp = bp;
if (uvp)
- vput(uvp);
+ vrele(uvp);
if (lvp)
- vput(lvp);
+ vrele(lvp);
vrele(rvp);
return error;
}
@@ -556,11 +532,7 @@ vnode_to_path(char *path, size_t len, st
bp = bend = &path[len];
*(--bp) = '\0';
- error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
- if (error != 0)
- return error;
- error = cache_revlookup(vp, &dvp, &bp, path, 0, 0);
- VOP_UNLOCK(vp);
+ error = cache_revlookup(vp, &dvp, &bp, path, false, 0);
if (error != 0)
return (error == -1 ? ENOENT : error);
Index: src/sys/kern/vfs_lookup.c
diff -u src/sys/kern/vfs_lookup.c:1.214 src/sys/kern/vfs_lookup.c:1.215
--- src/sys/kern/vfs_lookup.c:1.214 Sun Feb 23 22:14:03 2020
+++ src/sys/kern/vfs_lookup.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: vfs_lookup.c,v 1.214 2020/02/23 22:14:03 ad Exp $ */
+/* $NetBSD: vfs_lookup.c,v 1.215 2020/04/04 20:49:30 ad Exp $ */
/*
* Copyright (c) 1982, 1986, 1989, 1993
@@ -37,7 +37,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: vfs_lookup.c,v 1.214 2020/02/23 22:14:03 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: vfs_lookup.c,v 1.215 2020/04/04 20:49:30 ad Exp $");
#ifdef _KERNEL_OPT
#include "opt_magiclinks.h"
@@ -50,6 +50,7 @@ __KERNEL_RCSID(0, "$NetBSD: vfs_lookup.c
#include <sys/time.h>
#include <sys/namei.h>
#include <sys/vnode.h>
+#include <sys/vnode_impl.h>
#include <sys/mount.h>
#include <sys/errno.h>
#include <sys/filedesc.h>
@@ -709,8 +710,6 @@ namei_start(struct namei_state *state, i
return ENOTDIR;
}
- vn_lock(startdir, LK_EXCLUSIVE | LK_RETRY);
-
*startdir_ret = startdir;
return 0;
}
@@ -748,15 +747,17 @@ namei_follow(struct namei_state *state,
size_t linklen;
int error;
- KASSERT(VOP_ISLOCKED(searchdir) == LK_EXCLUSIVE);
- KASSERT(VOP_ISLOCKED(foundobj) == LK_EXCLUSIVE);
if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
return ELOOP;
}
+
+ vn_lock(foundobj, LK_EXCLUSIVE | LK_RETRY);
if (foundobj->v_mount->mnt_flag & MNT_SYMPERM) {
error = VOP_ACCESS(foundobj, VEXEC, cnp->cn_cred);
- if (error != 0)
+ if (error != 0) {
+ VOP_UNLOCK(foundobj);
return error;
+ }
}
/* FUTURE: fix this to not use a second buffer */
@@ -770,6 +771,7 @@ namei_follow(struct namei_state *state,
auio.uio_resid = MAXPATHLEN;
UIO_SETUP_SYSSPACE(&auio);
error = VOP_READLINK(foundobj, &auio, cnp->cn_cred);
+ VOP_UNLOCK(foundobj);
if (error) {
PNBUF_PUT(cp);
return error;
@@ -806,14 +808,11 @@ namei_follow(struct namei_state *state,
/* we're now starting from the beginning of the buffer again */
cnp->cn_nameptr = ndp->ni_pnbuf;
- /* must unlock this before relocking searchdir */
- VOP_UNLOCK(foundobj);
-
/*
* Check if root directory should replace current directory.
*/
if (ndp->ni_pnbuf[0] == '/') {
- vput(searchdir);
+ vrele(searchdir);
/* Keep absolute symbolic links inside emulation root */
searchdir = ndp->ni_erootdir;
if (searchdir == NULL ||
@@ -824,7 +823,6 @@ namei_follow(struct namei_state *state,
searchdir = ndp->ni_rootdir;
}
vref(searchdir);
- vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
while (cnp->cn_nameptr[0] == '/') {
cnp->cn_nameptr++;
ndp->ni_pathlen--;
@@ -832,7 +830,6 @@ namei_follow(struct namei_state *state,
}
*newsearchdir_ret = searchdir;
- KASSERT(VOP_ISLOCKED(searchdir) == LK_EXCLUSIVE);
return 0;
}
@@ -860,7 +857,7 @@ lookup_parsepath(struct namei_state *sta
* responsibility for freeing the pathname buffer.
*
* At this point, our only vnode state is that the search dir
- * is held and locked.
+ * is held.
*/
cnp->cn_consume = 0;
cnp->cn_namelen = namei_getcomponent(cnp->cn_nameptr);
@@ -917,6 +914,111 @@ lookup_parsepath(struct namei_state *sta
}
/*
+ * Take care of crossing a mounted-on vnode. On error, foundobj_ret will be
+ * vrele'd, but searchdir is left alone.
+ */
+static int
+lookup_crossmount(struct namei_state *state,
+ struct vnode **searchdir_ret,
+ struct vnode **foundobj_ret,
+ bool *searchdir_locked)
+{
+ struct componentname *cnp = state->cnp;
+ struct vnode *foundobj;
+ struct vnode *searchdir;
+ struct mount *mp;
+ int error, lktype;
+
+ searchdir = *searchdir_ret;
+ foundobj = *foundobj_ret;
+ error = 0;
+
+ KASSERT((cnp->cn_flags & NOCROSSMOUNT) == 0);
+ KASSERT(searchdir != NULL);
+
+ /* First, unlock searchdir (oof). */
+ if (*searchdir_locked) {
+ lktype = VOP_ISLOCKED(searchdir);
+ VOP_UNLOCK(searchdir);
+ *searchdir_locked = false;
+ } else {
+ lktype = LK_NONE;
+ }
+
+ /*
+ * Do an unlocked check to see if the vnode has been mounted on; if
+ * so find the root of the mounted file system.
+ */
+ while (foundobj->v_type == VDIR &&
+ (mp = foundobj->v_mountedhere) != NULL &&
+ (cnp->cn_flags & NOCROSSMOUNT) == 0) {
+ KASSERTMSG(searchdir != foundobj, "same vn %p", searchdir);
+ /*
+ * First get the vnode stable. LK_SHARED works brilliantly
+ * here because almost nothing else wants to lock the
+ * covered vnode.
+ */
+ error = vn_lock(foundobj, LK_SHARED);
+ if (error != 0) {
+ vrele(foundobj);
+ *foundobj_ret = NULL;
+ break;
+ }
+
+ /* Then check to see if something is still mounted on it. */
+ if ((mp = foundobj->v_mountedhere) == NULL) {
+ VOP_UNLOCK(foundobj);
+ break;
+ }
+
+ /* Get a reference to the mountpoint, and ditch foundobj. */
+ error = vfs_busy(mp);
+ vput(foundobj);
+ if (error != 0) {
+ *foundobj_ret = NULL;
+ break;
+ }
+
+ /* Now get a reference on the root vnode, and drop mount. */
+ error = VFS_ROOT(mp, LK_NONE, &foundobj);
+ vfs_unbusy(mp);
+ if (error) {
+ *foundobj_ret = NULL;
+ break;
+ }
+
+ /*
+ * Avoid locking vnodes from two filesystems because
+ * it's prone to deadlock, e.g. when using puffs.
+ * Also, it isn't a good idea to propagate slowness of
+ * a filesystem up to the root directory. For now,
+ * only handle the common case, where foundobj is
+ * VDIR.
+ *
+ * In this case set searchdir to null to avoid using
+ * it again. It is not correct to set searchdir ==
+ * foundobj here as that will confuse the caller.
+ * (See PR 40740.)
+ */
+ if (searchdir == NULL) {
+ /* already been here once; do nothing further */
+ } else if (foundobj->v_type == VDIR) {
+ vrele(searchdir);
+ *searchdir_ret = searchdir = NULL;
+ *foundobj_ret = foundobj;
+ lktype = LK_NONE;
+ }
+ }
+
+ /* If searchdir is still around, re-lock it. */
+ if (error == 0 && lktype != LK_NONE) {
+ vn_lock(searchdir, lktype | LK_RETRY);
+ *searchdir_locked = true;
+ }
+ return error;
+}
+
+/*
* Call VOP_LOOKUP for a single lookup; return a new search directory
* (used when crossing mountpoints up or searching union mounts down) and
* the found object, which for create operations may be NULL on success.
@@ -932,19 +1034,19 @@ static int
lookup_once(struct namei_state *state,
struct vnode *searchdir,
struct vnode **newsearchdir_ret,
- struct vnode **foundobj_ret)
+ struct vnode **foundobj_ret,
+ bool *newsearchdir_locked_ret)
{
struct vnode *tmpvn; /* scratch vnode */
struct vnode *foundobj; /* result */
- struct mount *mp; /* mount table entry */
struct lwp *l = curlwp;
- int error;
+ bool searchdir_locked = false;
+ int error, lktype;
struct componentname *cnp = state->cnp;
struct nameidata *ndp = state->ndp;
KASSERT(cnp == &ndp->ni_cnd);
- KASSERT(VOP_ISLOCKED(searchdir) == LK_EXCLUSIVE);
*newsearchdir_ret = searchdir;
/*
@@ -976,9 +1078,7 @@ lookup_once(struct namei_state *state,
if (ndp->ni_rootdir != rootvnode) {
int retval;
- VOP_UNLOCK(searchdir);
retval = vn_isunder(searchdir, ndp->ni_rootdir, l);
- vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
if (!retval) {
/* Oops! We got out of jail! */
log(LOG_WARNING,
@@ -987,12 +1087,11 @@ lookup_once(struct namei_state *state,
p->p_pid, kauth_cred_geteuid(l->l_cred),
p->p_comm);
/* Put us at the jail root. */
- vput(searchdir);
+ vrele(searchdir);
searchdir = NULL;
foundobj = ndp->ni_rootdir;
vref(foundobj);
vref(foundobj);
- vn_lock(foundobj, LK_EXCLUSIVE | LK_RETRY);
*newsearchdir_ret = foundobj;
*foundobj_ret = foundobj;
error = 0;
@@ -1005,18 +1104,35 @@ lookup_once(struct namei_state *state,
tmpvn = searchdir;
searchdir = searchdir->v_mount->mnt_vnodecovered;
vref(searchdir);
- vput(tmpvn);
- vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
+ vrele(tmpvn);
*newsearchdir_ret = searchdir;
}
}
/*
+ * If the file system supports VOP_LOOKUP() with a shared lock, and
+ * we are not making any modifications (nameiop LOOKUP) or this is
+ * not the last component then get a shared lock. Where we can't do
+ * fast-forwarded lookups (for example with layered file systems)
+ * then this is the fallback for reducing lock contention.
+ */
+ if ((searchdir->v_mount->mnt_iflag & IMNT_SHRLOOKUP) != 0 &&
+ (cnp->cn_nameiop == LOOKUP || (cnp->cn_flags & ISLASTCN) == 0)) {
+ lktype = LK_SHARED;
+ } else {
+ lktype = LK_EXCLUSIVE;
+ }
+
+ /*
* We now have a segment name to search for, and a directory to search.
- * Our vnode state here is that "searchdir" is held and locked.
+ * Our vnode state here is that "searchdir" is held.
*/
unionlookup:
foundobj = NULL;
+ if (!searchdir_locked) {
+ vn_lock(searchdir, lktype | LK_RETRY);
+ searchdir_locked = true;
+ }
error = VOP_LOOKUP(searchdir, &foundobj, cnp);
if (error != 0) {
@@ -1026,6 +1142,23 @@ unionlookup:
#ifdef NAMEI_DIAGNOSTIC
printf("not found\n");
#endif /* NAMEI_DIAGNOSTIC */
+
+ /*
+ * If ENOLCK, the file system needs us to retry the lookup
+ * with an exclusive lock. It's likely nothing was found in
+ * cache and/or modifications need to be made.
+ */
+ if (error == ENOLCK) {
+ KASSERT(VOP_ISLOCKED(searchdir) == LK_SHARED);
+ KASSERT(searchdir_locked);
+ if (vn_lock(searchdir, LK_UPGRADE | LK_NOWAIT)) {
+ VOP_UNLOCK(searchdir);
+ searchdir_locked = false;
+ }
+ lktype = LK_EXCLUSIVE;
+ goto unionlookup;
+ }
+
if ((error == ENOENT) &&
(searchdir->v_vflag & VV_ROOT) &&
(searchdir->v_mount->mnt_flag & MNT_UNION)) {
@@ -1033,7 +1166,7 @@ unionlookup:
searchdir = searchdir->v_mount->mnt_vnodecovered;
vref(searchdir);
vput(tmpvn);
- vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
+ searchdir_locked = false;
*newsearchdir_ret = searchdir;
goto unionlookup;
}
@@ -1087,85 +1220,187 @@ unionlookup:
cnp->cn_flags |= ISLASTCN;
}
- /*
- * "searchdir" is locked and held, "foundobj" is held,
- * they may be the same vnode.
- */
- if (searchdir != foundobj) {
- if (cnp->cn_flags & ISDOTDOT)
- VOP_UNLOCK(searchdir);
- error = vn_lock(foundobj, LK_EXCLUSIVE);
- if (cnp->cn_flags & ISDOTDOT)
- vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
- if (error != 0) {
- vrele(foundobj);
- goto done;
+ /* Unlock, unless the caller needs the parent locked. */
+ if (searchdir != NULL) {
+ KASSERT(searchdir_locked);
+ if ((cnp->cn_flags & (ISLASTCN | LOCKPARENT)) !=
+ (ISLASTCN | LOCKPARENT)) {
+ VOP_UNLOCK(searchdir);
+ searchdir_locked = false;
}
+ } else {
+ KASSERT(!searchdir_locked);
}
- /*
- * Check to see if the vnode has been mounted on;
- * if so find the root of the mounted file system.
- */
- KASSERT(searchdir != NULL);
- while (foundobj->v_type == VDIR &&
- (mp = foundobj->v_mountedhere) != NULL &&
- (cnp->cn_flags & NOCROSSMOUNT) == 0) {
+ *foundobj_ret = foundobj;
+ error = 0;
+done:
+ *newsearchdir_locked_ret = searchdir_locked;
+ return error;
+}
- KASSERT(searchdir != foundobj);
+/*
+ * Parse out the first path name component that we need to to consider.
+ *
+ * While doing this, attempt to use the name cache to fast-forward through
+ * as many "easy" to find components of the path as possible.
+ *
+ * We use the namecache's node locks to form a chain, and avoid as many
+ * vnode references and locks as possible. In the ideal case, only the
+ * final vnode will have its reference count adjusted and lock taken.
+ */
+static int
+lookup_fastforward(struct namei_state *state, struct vnode **searchdir_ret,
+ struct vnode **foundobj_ret)
+{
+ struct componentname *cnp = state->cnp;
+ struct nameidata *ndp = state->ndp;
+ krwlock_t *plock;
+ struct vnode *foundobj, *searchdir;
+ int error, error2;
+ size_t oldpathlen;
+ const char *oldnameptr;
- error = vfs_busy(mp);
- if (error != 0) {
- vput(foundobj);
- goto done;
+ /*
+ * Eat as many path name components as possible before giving up and
+ * letting lookup_once() handle it. Remember the starting point in
+ * case we can't get vnode references and need to roll back.
+ */
+ plock = NULL;
+ searchdir = *searchdir_ret;
+ oldnameptr = cnp->cn_nameptr;
+ oldpathlen = ndp->ni_pathlen;
+ for (;;) {
+ foundobj = NULL;
+
+ /*
+ * Get the next component name. There should be no slashes
+ * here, and we shouldn't have looped around if we were
+ * done.
+ */
+ KASSERT(cnp->cn_nameptr[0] != '/');
+ KASSERT(cnp->cn_nameptr[0] != '\0');
+ if ((error = lookup_parsepath(state)) != 0) {
+ break;
}
- if (searchdir != NULL) {
- VOP_UNLOCK(searchdir);
+
+ /*
+ * Can't deal with dotdot lookups, because it means lock
+ * order reversal, and there are checks in lookup_once()
+ * that need to be made. Also check for missing mountpoints.
+ */
+ if ((cnp->cn_flags & ISDOTDOT) != 0 ||
+ searchdir->v_mount == NULL) {
+ error = EOPNOTSUPP;
+ break;
}
- vput(foundobj);
- error = VFS_ROOT(mp, LK_EXCLUSIVE, &foundobj);
- vfs_unbusy(mp);
- if (error) {
- if (searchdir != NULL) {
- vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
+
+ /*
+ * Can't deal with last component when modifying; this needs
+ * searchdir locked and VOP_LOOKUP() called (which can and
+ * does modify state, despite the name).
+ */
+ if ((cnp->cn_flags & ISLASTCN) != 0) {
+ if (cnp->cn_nameiop != LOOKUP ||
+ (cnp->cn_flags & LOCKPARENT) != 0) {
+ error = EOPNOTSUPP;
+ break;
}
- goto done;
}
+
+ /* Can't deal with -o union lookups. */
+ if ((searchdir->v_vflag & VV_ROOT) != 0 &&
+ (searchdir->v_mount->mnt_flag & MNT_UNION) != 0) {
+ error = EOPNOTSUPP;
+ break;
+ }
+
/*
- * Avoid locking vnodes from two filesystems because
- * it's prone to deadlock, e.g. when using puffs.
- * Also, it isn't a good idea to propagate slowness of
- * a filesystem up to the root directory. For now,
- * only handle the common case, where foundobj is
- * VDIR.
+ * Good, now look for it in cache. cache_lookup_linked()
+ * will fail if there's nothing there, or if there's no
+ * ownership info for the directory, or if the user doesn't
+ * have permission to look up files in this directory.
+ */
+ if (!cache_lookup_linked(searchdir, cnp->cn_nameptr,
+ cnp->cn_namelen, &foundobj, &plock, cnp->cn_cred)) {
+ error = EOPNOTSUPP;
+ break;
+ }
+ KASSERT(plock != NULL && rw_lock_held(plock));
+
+ /* Scored a hit. Negative is good too (ENOENT). */
+ if (foundobj == NULL) {
+ error = ENOENT;
+ break;
+ }
+
+ /*
+ * Stop and get a hold on the vnode if there's something
+ * that can't be handled here:
*
- * In this case set searchdir to null to avoid using
- * it again. It is not correct to set searchdir ==
- * foundobj here as that will confuse the caller.
- * (See PR 40740.)
+ * - we've reached the last component.
+ * - or encountered a mount point that needs to be crossed.
+ * - or encountered something other than a directory.
*/
- if (searchdir == NULL) {
- /* already been here once; do nothing further */
- } else if (foundobj->v_type == VDIR) {
- vrele(searchdir);
- *newsearchdir_ret = searchdir = NULL;
+ if ((cnp->cn_flags & ISLASTCN) != 0 ||
+ foundobj->v_type != VDIR ||
+ (foundobj->v_type == VDIR &&
+ foundobj->v_mountedhere != NULL)) {
+ mutex_enter(foundobj->v_interlock);
+ error = vcache_tryvget(foundobj);
+ /* v_interlock now unheld */
+ if (error != 0) {
+ foundobj = NULL;
+ }
+ break;
+ }
+
+ /*
+ * Otherwise, we're still in business. Set the found VDIR
+ * vnode as the search dir for the next component and
+ * continue on to it.
+ */
+ cnp->cn_nameptr = ndp->ni_next;
+ searchdir = foundobj;
+ }
+
+ /*
+ * If we ended up with a new search dir, ref it before dropping the
+ * namecache's lock. The lock prevents both searchdir and foundobj
+ * from disappearing. If we can't ref the new searchdir, we have a
+ * bit of a problem. Roll back the fastforward to the beginning and
+ * let lookup_once() take care of it.
+ */
+ if (searchdir != *searchdir_ret) {
+ mutex_enter(searchdir->v_interlock);
+ error2 = vcache_tryvget(searchdir);
+ /* v_interlock now unheld */
+ KASSERT(plock != NULL);
+ rw_exit(plock);
+ if (__predict_true(error2 == 0)) {
+ /* Returning new searchdir, and maybe new foundobj. */
+ vrele(*searchdir_ret);
+ *searchdir_ret = searchdir;
} else {
- VOP_UNLOCK(foundobj);
- vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
- vn_lock(foundobj, LK_EXCLUSIVE | LK_RETRY);
+ /* Returning nothing. */
+ if (foundobj != NULL) {
+ vrele(foundobj);
+ foundobj = NULL;
+ }
+ cnp->cn_nameptr = oldnameptr;
+ ndp->ni_pathlen = oldpathlen;
+ error = lookup_parsepath(state);
+ if (error == 0) {
+ error = EOPNOTSUPP;
+ }
}
+ } else if (plock != NULL) {
+ /* Drop any namecache lock still held. */
+ rw_exit(plock);
}
+ KASSERT(error == 0 ? foundobj != NULL : foundobj == NULL);
*foundobj_ret = foundobj;
- error = 0;
-done:
- KASSERT(*newsearchdir_ret == NULL ||
- VOP_ISLOCKED(*newsearchdir_ret) == LK_EXCLUSIVE);
- /*
- * *foundobj_ret is valid only if error == 0.
- */
- KASSERT(error != 0 || *foundobj_ret == NULL ||
- VOP_ISLOCKED(*foundobj_ret) == LK_EXCLUSIVE);
return error;
}
@@ -1182,6 +1417,7 @@ namei_oneroot(struct namei_state *state,
struct nameidata *ndp = state->ndp;
struct componentname *cnp = state->cnp;
struct vnode *searchdir, *foundobj;
+ bool searchdir_locked = false;
int error;
error = namei_start(state, isnfsd, &searchdir);
@@ -1222,44 +1458,47 @@ namei_oneroot(struct namei_state *state,
for (;;) {
KASSERT(searchdir != NULL);
- KASSERT(VOP_ISLOCKED(searchdir) == LK_EXCLUSIVE);
+ KASSERT(!searchdir_locked);
/*
- * If the directory we're on is unmounted, bail out.
- * XXX: should this also check if it's unlinked?
- * XXX: yes it should... but how?
+ * Parse out the first path name component that we need to
+ * to consider. While doing this, attempt to use the name
+ * cache to fast-forward through as many "easy" to find
+ * components of the path as possible.
*/
- if (searchdir->v_mount == NULL) {
- vput(searchdir);
- ndp->ni_dvp = NULL;
- ndp->ni_vp = NULL;
- return (ENOENT);
- }
+ error = lookup_fastforward(state, &searchdir, &foundobj);
/*
- * Look up the next path component.
- * (currently, this may consume more than one)
+ * If we didn't get a good answer from the namecache, then
+ * go directly to the file system.
*/
+ if (error != 0 && error != ENOENT) {
+ error = lookup_once(state, searchdir, &searchdir,
+ &foundobj, &searchdir_locked);
+ }
- /* There should be no slashes here. */
- KASSERT(cnp->cn_nameptr[0] != '/');
-
- /* and we shouldn't have looped around if we were done */
- KASSERT(cnp->cn_nameptr[0] != '\0');
-
- error = lookup_parsepath(state);
- if (error) {
- vput(searchdir);
- ndp->ni_dvp = NULL;
- ndp->ni_vp = NULL;
- state->attempt_retry = 1;
- return (error);
+ /*
+ * If the vnode we found is mounted on, then cross the mount
+ * and get the root vnode in foundobj. If this encounters
+ * an error, it will dispose of foundobj, but searchdir is
+ * untouched.
+ */
+ if (error == 0 && foundobj != NULL &&
+ foundobj->v_type == VDIR &&
+ foundobj->v_mountedhere != NULL &&
+ (cnp->cn_flags & NOCROSSMOUNT) == 0) {
+ error = lookup_crossmount(state, &searchdir,
+ &foundobj, &searchdir_locked);
}
- error = lookup_once(state, searchdir, &searchdir, &foundobj);
if (error) {
if (searchdir != NULL) {
- vput(searchdir);
+ if (searchdir_locked) {
+ searchdir_locked = false;
+ vput(searchdir);
+ } else {
+ vrele(searchdir);
+ }
}
ndp->ni_dvp = NULL;
ndp->ni_vp = NULL;
@@ -1296,6 +1535,11 @@ namei_oneroot(struct namei_state *state,
* them again.
*/
if (namei_atsymlink(state, foundobj)) {
+ /* Don't need searchdir locked any more. */
+ if (searchdir_locked) {
+ searchdir_locked = false;
+ VOP_UNLOCK(searchdir);
+ }
ndp->ni_pathlen += state->slashes;
ndp->ni_next -= state->slashes;
if (neverfollow) {
@@ -1337,14 +1581,13 @@ namei_oneroot(struct namei_state *state,
if (error) {
KASSERT(searchdir != foundobj);
if (searchdir != NULL) {
- vput(searchdir);
+ vrele(searchdir);
}
- vput(foundobj);
+ vrele(foundobj);
ndp->ni_dvp = NULL;
ndp->ni_vp = NULL;
return error;
}
- /* namei_follow unlocks it (ugh) so rele, not put */
vrele(foundobj);
foundobj = NULL;
@@ -1375,9 +1618,16 @@ namei_oneroot(struct namei_state *state,
(cnp->cn_flags & REQUIREDIR)) {
KASSERT(foundobj != searchdir);
if (searchdir) {
- vput(searchdir);
+ if (searchdir_locked) {
+ searchdir_locked = false;
+ vput(searchdir);
+ } else {
+ vrele(searchdir);
+ }
+ } else {
+ KASSERT(!searchdir_locked);
}
- vput(foundobj);
+ vrele(foundobj);
ndp->ni_dvp = NULL;
ndp->ni_vp = NULL;
state->attempt_retry = 1;
@@ -1395,15 +1645,21 @@ namei_oneroot(struct namei_state *state,
* Continue with the next component.
*/
cnp->cn_nameptr = ndp->ni_next;
- if (searchdir == foundobj) {
- vrele(searchdir);
- } else if (searchdir != NULL) {
- vput(searchdir);
+ if (searchdir != NULL) {
+ if (searchdir_locked) {
+ searchdir_locked = false;
+ vput(searchdir);
+ } else {
+ vrele(searchdir);
+ }
}
searchdir = foundobj;
foundobj = NULL;
}
+ KASSERT((cnp->cn_flags & LOCKPARENT) == 0 || searchdir == NULL ||
+ VOP_ISLOCKED(searchdir) == LK_EXCLUSIVE);
+
skiploop:
if (foundobj != NULL) {
@@ -1416,16 +1672,17 @@ namei_oneroot(struct namei_state *state,
* forever. So convert it to the real root.
*/
if (searchdir != NULL) {
- if (searchdir == foundobj)
- vrele(searchdir);
- else
+ if (searchdir_locked) {
vput(searchdir);
+ searchdir_locked = false;
+ } else {
+ vrele(searchdir);
+ }
searchdir = NULL;
}
- vput(foundobj);
+ vrele(foundobj);
foundobj = ndp->ni_rootdir;
vref(foundobj);
- vn_lock(foundobj, LK_EXCLUSIVE | LK_RETRY);
}
/*
@@ -1438,9 +1695,15 @@ namei_oneroot(struct namei_state *state,
(searchdir == NULL ||
searchdir->v_mount != foundobj->v_mount)) {
if (searchdir) {
- vput(searchdir);
+ if (searchdir_locked) {
+ vput(searchdir);
+ searchdir_locked = false;
+ } else {
+ vrele(searchdir);
+ }
+ searchdir = NULL;
}
- vput(foundobj);
+ vrele(foundobj);
foundobj = NULL;
ndp->ni_dvp = NULL;
ndp->ni_vp = NULL;
@@ -1465,21 +1728,25 @@ namei_oneroot(struct namei_state *state,
if (state->rdonly &&
(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
if (searchdir) {
- if (foundobj != searchdir) {
+ if (searchdir_locked) {
vput(searchdir);
+ searchdir_locked = false;
} else {
vrele(searchdir);
}
searchdir = NULL;
}
- vput(foundobj);
+ vrele(foundobj);
foundobj = NULL;
ndp->ni_dvp = NULL;
ndp->ni_vp = NULL;
state->attempt_retry = 1;
return EROFS;
}
- if ((cnp->cn_flags & LOCKLEAF) == 0) {
+
+ /* Lock the leaf node if requested. */
+ if ((cnp->cn_flags & (LOCKLEAF | LOCKPARENT)) == LOCKPARENT &&
+ searchdir == foundobj) {
/*
* Note: if LOCKPARENT but not LOCKLEAF is
* set, and searchdir == foundobj, this code
@@ -1491,7 +1758,15 @@ namei_oneroot(struct namei_state *state,
* that uses this combination "knows" this, so
* it can't be safely changed. Feh. XXX
*/
- VOP_UNLOCK(foundobj);
+ KASSERT(searchdir_locked);
+ VOP_UNLOCK(searchdir);
+ searchdir_locked = false;
+ } else if ((cnp->cn_flags & LOCKLEAF) != 0 &&
+ (searchdir != foundobj ||
+ (cnp->cn_flags & LOCKPARENT) == 0)) {
+ const int lktype = (cnp->cn_flags & LOCKSHARED) != 0 ?
+ LK_SHARED : LK_EXCLUSIVE;
+ vn_lock(foundobj, lktype | LK_RETRY);
}
}
@@ -1503,11 +1778,7 @@ namei_oneroot(struct namei_state *state,
* If LOCKPARENT is not set, the parent directory isn't returned.
*/
if ((cnp->cn_flags & LOCKPARENT) == 0 && searchdir != NULL) {
- if (searchdir == foundobj) {
- vrele(searchdir);
- } else {
- vput(searchdir);
- }
+ vrele(searchdir);
searchdir = NULL;
}
@@ -1649,6 +1920,7 @@ do_lookup_for_nfsd_index(struct namei_st
struct nameidata *ndp = state->ndp;
struct vnode *startdir;
struct vnode *foundobj;
+ bool startdir_locked;
const char *cp; /* pointer into pathname argument */
KASSERT(cnp == &ndp->ni_cnd);
@@ -1681,30 +1953,37 @@ do_lookup_for_nfsd_index(struct namei_st
* own reference to it to avoid consuming the caller's.
*/
vref(startdir);
- vn_lock(startdir, LK_EXCLUSIVE | LK_RETRY);
- error = lookup_once(state, startdir, &startdir, &foundobj);
- if (error == 0 && startdir == foundobj) {
- vrele(startdir);
- } else if (startdir != NULL) {
- vput(startdir);
- }
- if (error) {
- goto bad;
- }
- ndp->ni_vp = foundobj;
+ error = lookup_once(state, startdir, &startdir, &foundobj,
+ &startdir_locked);
- if (foundobj == NULL) {
- return 0;
+ KASSERT((cnp->cn_flags & LOCKPARENT) == 0);
+ if (startdir_locked) {
+ VOP_UNLOCK(startdir);
+ startdir_locked = false;
}
- KASSERT((cnp->cn_flags & LOCKPARENT) == 0);
- if ((cnp->cn_flags & LOCKLEAF) == 0) {
- VOP_UNLOCK(foundobj);
+ /*
+ * If the vnode we found is mounted on, then cross the mount and get
+ * the root vnode in foundobj. If this encounters an error, it will
+ * dispose of foundobj, but searchdir is untouched.
+ */
+ if (error == 0 && foundobj != NULL &&
+ foundobj->v_type == VDIR &&
+ foundobj->v_mountedhere != NULL &&
+ (cnp->cn_flags & NOCROSSMOUNT) == 0) {
+ error = lookup_crossmount(state, &startdir, &foundobj,
+ &startdir_locked);
}
- return (0);
-bad:
- ndp->ni_vp = NULL;
+ /* Now toss startdir and see if we have an error. */
+ if (startdir != NULL)
+ vrele(startdir);
+ if (error)
+ foundobj = NULL;
+ else if (foundobj != NULL && (cnp->cn_flags & LOCKLEAF) != 0)
+ vn_lock(foundobj, LK_EXCLUSIVE | LK_RETRY);
+
+ ndp->ni_vp = foundobj;
return (error);
}
Index: src/sys/kern/vfs_syscalls.c
diff -u src/sys/kern/vfs_syscalls.c:1.544 src/sys/kern/vfs_syscalls.c:1.545
--- src/sys/kern/vfs_syscalls.c:1.544 Wed Mar 25 18:08:34 2020
+++ src/sys/kern/vfs_syscalls.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: vfs_syscalls.c,v 1.544 2020/03/25 18:08:34 gdt Exp $ */
+/* $NetBSD: vfs_syscalls.c,v 1.545 2020/04/04 20:49:30 ad Exp $ */
/*-
* Copyright (c) 2008, 2009, 2019, 2020 The NetBSD Foundation, Inc.
@@ -70,7 +70,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: vfs_syscalls.c,v 1.544 2020/03/25 18:08:34 gdt Exp $");
+__KERNEL_RCSID(0, "$NetBSD: vfs_syscalls.c,v 1.545 2020/04/04 20:49:30 ad Exp $");
#ifdef _KERNEL_OPT
#include "opt_fileassoc.h"
@@ -1529,7 +1529,7 @@ chdir_lookup(const char *path, int where
if (error) {
return error;
}
- NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
+ NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | LOCKSHARED | TRYEMULROOT, pb);
if ((error = namei(&nd)) != 0) {
pathbuf_destroy(pb);
return error;
@@ -2995,7 +2995,7 @@ do_sys_accessat(struct lwp *l, int fdat,
return EINVAL;
}
- nd_flag = FOLLOW | LOCKLEAF | TRYEMULROOT;
+ nd_flag = FOLLOW | LOCKLEAF | LOCKSHARED | TRYEMULROOT;
if (flags & AT_SYMLINK_NOFOLLOW)
nd_flag &= ~FOLLOW;
@@ -3221,7 +3221,7 @@ do_sys_readlinkat(struct lwp *l, int fda
if (error) {
return error;
}
- NDINIT(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | TRYEMULROOT, pb);
+ NDINIT(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED | TRYEMULROOT, pb);
if ((error = fd_nameiat(l, fdat, &nd)) != 0) {
pathbuf_destroy(pb);
return error;
@@ -4687,7 +4687,7 @@ dorevoke(struct vnode *vp, kauth_cred_t
struct vattr vattr;
int error, fs_decision;
- vn_lock(vp, LK_SHARED | LK_RETRY);
+ vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
error = VOP_GETATTR(vp, &vattr, cred);
VOP_UNLOCK(vp);
if (error != 0)
Index: src/sys/kern/vfs_vnode.c
diff -u src/sys/kern/vfs_vnode.c:1.116 src/sys/kern/vfs_vnode.c:1.117
--- src/sys/kern/vfs_vnode.c:1.116 Sun Mar 22 18:45:28 2020
+++ src/sys/kern/vfs_vnode.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: vfs_vnode.c,v 1.116 2020/03/22 18:45:28 ad Exp $ */
+/* $NetBSD: vfs_vnode.c,v 1.117 2020/04/04 20:49:30 ad Exp $ */
/*-
* Copyright (c) 1997-2011, 2019, 2020 The NetBSD Foundation, Inc.
@@ -155,7 +155,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: vfs_vnode.c,v 1.116 2020/03/22 18:45:28 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: vfs_vnode.c,v 1.117 2020/04/04 20:49:30 ad Exp $");
#ifdef _KERNEL_OPT
#include "opt_pax.h"
@@ -828,9 +828,6 @@ vrelel(vnode_t *vp, int flags, int lktyp
if (VSTATE_GET(vp) == VS_RECLAIMED) {
VOP_UNLOCK(vp);
} else {
- VSTATE_CHANGE(vp, VS_LOADED, VS_BLOCKED);
- mutex_exit(vp->v_interlock);
-
/*
* The vnode must not gain another reference while being
* deactivated. If VOP_INACTIVE() indicates that
@@ -839,19 +836,19 @@ vrelel(vnode_t *vp, int flags, int lktyp
*
* Note that VOP_INACTIVE() will not drop the vnode lock.
*/
+ mutex_exit(vp->v_interlock);
recycle = false;
VOP_INACTIVE(vp, &recycle);
- if (!recycle)
- VOP_UNLOCK(vp);
rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
mutex_enter(vp->v_interlock);
- VSTATE_CHANGE(vp, VS_BLOCKED, VS_LOADED);
+ if (vtryrele(vp)) {
+ VOP_UNLOCK(vp);
+ mutex_exit(vp->v_interlock);
+ rw_exit(vp->v_uobj.vmobjlock);
+ return;
+ }
if (!recycle) {
- if (vtryrele(vp)) {
- mutex_exit(vp->v_interlock);
- rw_exit(vp->v_uobj.vmobjlock);
- return;
- }
+ VOP_UNLOCK(vp);
}
/* Take care of space accounting. */
Index: src/sys/miscfs/genfs/layer_vnops.c
diff -u src/sys/miscfs/genfs/layer_vnops.c:1.68 src/sys/miscfs/genfs/layer_vnops.c:1.69
--- src/sys/miscfs/genfs/layer_vnops.c:1.68 Sun Feb 23 15:46:41 2020
+++ src/sys/miscfs/genfs/layer_vnops.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: layer_vnops.c,v 1.68 2020/02/23 15:46:41 ad Exp $ */
+/* $NetBSD: layer_vnops.c,v 1.69 2020/04/04 20:49:30 ad Exp $ */
/*
* Copyright (c) 1999 National Aeronautics & Space Administration
@@ -170,7 +170,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: layer_vnops.c,v 1.68 2020/02/23 15:46:41 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: layer_vnops.c,v 1.69 2020/04/04 20:49:30 ad Exp $");
#include <sys/param.h>
#include <sys/systm.h>
@@ -384,6 +384,7 @@ layer_lookup(void *v)
vrele(lvp);
} else if (lvp != NULL) {
/* Note: dvp and ldvp are both locked. */
+ KASSERT(error != ENOLCK);
error = layer_node_create(dvp->v_mount, lvp, ap->a_vpp);
if (error) {
vrele(lvp);
Index: src/sys/miscfs/nullfs/null_vfsops.c
diff -u src/sys/miscfs/nullfs/null_vfsops.c:1.97 src/sys/miscfs/nullfs/null_vfsops.c:1.98
--- src/sys/miscfs/nullfs/null_vfsops.c:1.97 Mon Mar 16 21:20:11 2020
+++ src/sys/miscfs/nullfs/null_vfsops.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: null_vfsops.c,v 1.97 2020/03/16 21:20:11 pgoyette Exp $ */
+/* $NetBSD: null_vfsops.c,v 1.98 2020/04/04 20:49:30 ad Exp $ */
/*
* Copyright (c) 1999 National Aeronautics & Space Administration
@@ -76,7 +76,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: null_vfsops.c,v 1.97 2020/03/16 21:20:11 pgoyette Exp $");
+__KERNEL_RCSID(0, "$NetBSD: null_vfsops.c,v 1.98 2020/04/04 20:49:30 ad Exp $");
#include <sys/param.h>
#include <sys/systm.h>
@@ -139,6 +139,7 @@ nullfs_mount(struct mount *mp, const cha
nmp = kmem_zalloc(sizeof(struct null_mount), KM_SLEEP);
mp->mnt_data = nmp;
mp->mnt_iflag |= IMNT_MPSAFE;
+ mp->mnt_iflag |= lowerrootvp->v_mount->mnt_iflag & IMNT_SHRLOOKUP;
/*
* Make sure that the mount point is sufficiently initialized
Index: src/sys/miscfs/procfs/procfs_vfsops.c
diff -u src/sys/miscfs/procfs/procfs_vfsops.c:1.103 src/sys/miscfs/procfs/procfs_vfsops.c:1.104
--- src/sys/miscfs/procfs/procfs_vfsops.c:1.103 Mon Mar 16 21:20:11 2020
+++ src/sys/miscfs/procfs/procfs_vfsops.c Sat Apr 4 20:49:30 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: procfs_vfsops.c,v 1.103 2020/03/16 21:20:11 pgoyette Exp $ */
+/* $NetBSD: procfs_vfsops.c,v 1.104 2020/04/04 20:49:30 ad Exp $ */
/*
* Copyright (c) 1993
@@ -76,7 +76,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: procfs_vfsops.c,v 1.103 2020/03/16 21:20:11 pgoyette Exp $");
+__KERNEL_RCSID(0, "$NetBSD: procfs_vfsops.c,v 1.104 2020/04/04 20:49:30 ad Exp $");
#if defined(_KERNEL_OPT)
#include "opt_compat_netbsd.h"
@@ -171,7 +171,7 @@ procfs_mount(
else
pmnt->pmnt_flags = 0;
- mp->mnt_iflag |= IMNT_MPSAFE;
+ mp->mnt_iflag |= IMNT_MPSAFE | IMNT_SHRLOOKUP;
return error;
}
Index: src/sys/sys/fstypes.h
diff -u src/sys/sys/fstypes.h:1.37 src/sys/sys/fstypes.h:1.38
--- src/sys/sys/fstypes.h:1.37 Wed Feb 20 10:07:27 2019
+++ src/sys/sys/fstypes.h Sat Apr 4 20:49:31 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: fstypes.h,v 1.37 2019/02/20 10:07:27 hannken Exp $ */
+/* $NetBSD: fstypes.h,v 1.38 2020/04/04 20:49:31 ad Exp $ */
/*
* Copyright (c) 1989, 1991, 1993
@@ -220,7 +220,9 @@ typedef struct fhandle fhandle_t;
#define IMNT_UNMOUNT 0x00000002 /* unmount in progress */
#define IMNT_WANTRDWR 0x00000004 /* upgrade to read/write requested */
#define IMNT_WANTRDONLY 0x00000008 /* upgrade to readonly requested */
+#define IMNT_NCLOOKUP 0x00000020 /* can do lookop direct in namecache */
#define IMNT_DTYPE 0x00000040 /* returns d_type fields */
+#define IMNT_SHRLOOKUP 0x00000080 /* can do LK_SHARED lookups */
#define IMNT_MPSAFE 0x00000100 /* file system code MP safe */
#define IMNT_CAN_RWTORO 0x00000200 /* can downgrade fs to from rw to r/o */
#define IMNT_ONWORKLIST 0x00000400 /* on syncer worklist */
@@ -271,7 +273,9 @@ typedef struct fhandle fhandle_t;
"\13IMNT_ONWORKLIST" \
"\12IMNT_CAN_RWTORO" \
"\11IMNT_MPSAFE" \
+ "\10IMNT_SHRLOOKUP" \
"\07IMNT_DTYPE" \
+ "\06IMNT_NCLOOKUP" \
"\04IMNT_WANTRDONLY" \
"\03IMNT_WANTRDWR" \
"\02IMNT_UNMOUNT" \
Index: src/sys/sys/namei.src
diff -u src/sys/sys/namei.src:1.52 src/sys/sys/namei.src:1.53
--- src/sys/sys/namei.src:1.52 Mon Mar 23 23:28:11 2020
+++ src/sys/sys/namei.src Sat Apr 4 20:49:31 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: namei.src,v 1.52 2020/03/23 23:28:11 ad Exp $ */
+/* $NetBSD: namei.src,v 1.53 2020/04/04 20:49:31 ad Exp $ */
/*
* Copyright (c) 1985, 1989, 1991, 1993
@@ -152,13 +152,14 @@ NAMEIFL NOFOLLOW 0x00000000 /* do not fo
(pseudo) */
NAMEIFL EMULROOTSET 0x00000080 /* emulation root already
in ni_erootdir */
+NAMEIFL LOCKSHARED 0x00000100 /* want shared locks if possible */
NAMEIFL NOCHROOT 0x01000000 /* no chroot on abs path lookups */
-NAMEIFL MODMASK 0x010000fc /* mask of operational modifiers */
+NAMEIFL MODMASK 0x010001fc /* mask of operational modifiers */
/*
* Namei parameter descriptors.
*/
-NAMEIFL NOCROSSMOUNT 0x0000100 /* do not cross mount points */
-NAMEIFL RDONLY 0x0000200 /* lookup with read-only semantics */
+NAMEIFL NOCROSSMOUNT 0x0000800 /* do not cross mount points */
+NAMEIFL RDONLY 0x0001000 /* lookup with read-only semantics */
NAMEIFL ISDOTDOT 0x0002000 /* current component name is .. */
NAMEIFL MAKEENTRY 0x0004000 /* entry is to be added to name cache */
NAMEIFL ISLASTCN 0x0008000 /* this is last component of pathname */
@@ -166,7 +167,7 @@ NAMEIFL ISWHITEOUT 0x0020000 /* found wh
NAMEIFL DOWHITEOUT 0x0040000 /* do whiteouts */
NAMEIFL REQUIREDIR 0x0080000 /* must be a directory */
NAMEIFL CREATEDIR 0x0200000 /* trailing slashes are ok */
-NAMEIFL PARAMASK 0x02ee300 /* mask of parameter descriptors */
+NAMEIFL PARAMASK 0x02ef800 /* mask of parameter descriptors */
/*
* Initialization of a nameidata structure.
@@ -207,12 +208,11 @@ NAMEIFL PARAMASK 0x02ee300 /* mask of pa
*
* Field markings and their corresponding locks:
*
- * - stable throught the lifetime of the namecache entry
+ * - stable throughout the lifetime of the namecache entry
* d protected by nc_dvp->vi_nc_lock
* v protected by nc_vp->vi_nc_listlock
* l protected by cache_lru_lock
*/
-struct nchnode;
struct namecache {
struct rb_node nc_tree; /* d red-black tree, must be first */
uint64_t nc_key; /* - hashed key value */
@@ -220,7 +220,7 @@ struct namecache {
TAILQ_ENTRY(namecache) nc_lru; /* l pseudo-lru chain */
struct vnode *nc_dvp; /* - vnode of parent of name */
struct vnode *nc_vp; /* - vnode the name refers to */
- int nc_lrulist; /* l which LRU list its on */
+ int nc_lrulist; /* l which LRU list it's on */
u_short nc_nlen; /* - length of the name */
char nc_whiteout; /* - true if a whiteout */
char nc_name[41]; /* - segment name */
@@ -313,18 +313,17 @@ void namecache_print(struct vnode *, voi
* Stats on usefulness of namei caches. A couple of structures are
* used for counting, with members having the same names but different
* types. Containerize member names with the preprocessor to avoid
- * cut-'n'-paste. A (U) in the comment documents values that are
- * incremented unlocked; we may treat these specially.
+ * cut-'n'-paste.
*/
#define _NAMEI_CACHE_STATS(type) { \
- type ncs_goodhits; /* hits that we can really use (U) */ \
+ type ncs_goodhits; /* hits that we can really use */ \
type ncs_neghits; /* negative hits that we can use */ \
type ncs_badhits; /* hits we must drop */ \
- type ncs_falsehits; /* hits with id mismatch (U) */ \
+ type ncs_falsehits; /* hits with id mismatch */ \
type ncs_miss; /* misses */ \
type ncs_long; /* long names that ignore cache */ \
- type ncs_pass2; /* names found with passes == 2 (U) */ \
- type ncs_2passes; /* number of times we attempt it (U) */ \
+ type ncs_pass2; /* names found with passes == 2 */ \
+ type ncs_2passes; /* number of times we attempt it */ \
type ncs_revhits; /* reverse-cache hits */ \
type ncs_revmiss; /* reverse-cache misses */ \
type ncs_denied; /* access denied */ \
Index: src/sys/ufs/chfs/chfs_vnops.c
diff -u src/sys/ufs/chfs/chfs_vnops.c:1.36 src/sys/ufs/chfs/chfs_vnops.c:1.37
--- src/sys/ufs/chfs/chfs_vnops.c:1.36 Sun Feb 23 15:46:42 2020
+++ src/sys/ufs/chfs/chfs_vnops.c Sat Apr 4 20:49:31 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: chfs_vnops.c,v 1.36 2020/02/23 15:46:42 ad Exp $ */
+/* $NetBSD: chfs_vnops.c,v 1.37 2020/04/04 20:49:31 ad Exp $ */
/*-
* Copyright (c) 2010 Department of Software Engineering,
@@ -90,6 +90,10 @@ chfs_lookup(void *v)
return (*vpp == NULLVP ? ENOENT : 0);
}
+ /* May need to restart the lookup with an exclusive lock. */
+ if (VOP_ISLOCKED(dvp) != LK_EXCLUSIVE)
+ return ENOLCK;
+
ip = VTOI(dvp);
ump = VFSTOUFS(dvp->v_mount);
chmp = ump->um_chfs;
Index: src/sys/ufs/ext2fs/ext2fs_lookup.c
diff -u src/sys/ufs/ext2fs/ext2fs_lookup.c:1.89 src/sys/ufs/ext2fs/ext2fs_lookup.c:1.90
--- src/sys/ufs/ext2fs/ext2fs_lookup.c:1.89 Sat Mar 14 18:08:40 2020
+++ src/sys/ufs/ext2fs/ext2fs_lookup.c Sat Apr 4 20:49:31 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: ext2fs_lookup.c,v 1.89 2020/03/14 18:08:40 ad Exp $ */
+/* $NetBSD: ext2fs_lookup.c,v 1.90 2020/04/04 20:49:31 ad Exp $ */
/*
* Modified for NetBSD 1.2E
@@ -48,7 +48,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ext2fs_lookup.c,v 1.89 2020/03/14 18:08:40 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ext2fs_lookup.c,v 1.90 2020/04/04 20:49:31 ad Exp $");
#include <sys/param.h>
#include <sys/systm.h>
@@ -313,14 +313,6 @@ ext2fs_lookup(void *v)
*vpp = NULL;
/*
- * Produce the auxiliary lookup results into i_crap. Increment
- * its serial number so elsewhere we can tell if we're using
- * stale results. This should not be done this way. XXX.
- */
- results = &dp->i_crap;
- dp->i_crapcounter++;
-
- /*
* Check accessiblity of directory.
*/
if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
@@ -342,6 +334,18 @@ ext2fs_lookup(void *v)
return *vpp == NULLVP ? ENOENT : 0;
}
+ /* May need to restart the lookup with an exclusive lock. */
+ if (VOP_ISLOCKED(vdp) != LK_EXCLUSIVE)
+ return ENOLCK;
+
+ /*
+ * Produce the auxiliary lookup results into i_crap. Increment
+ * its serial number so elsewhere we can tell if we're using
+ * stale results. This should not be done this way. XXX.
+ */
+ results = &dp->i_crap;
+ dp->i_crapcounter++;
+
/*
* Suppress search for slots unless creating
* file and at end of pathname, in which case
Index: src/sys/ufs/ext2fs/ext2fs_vfsops.c
diff -u src/sys/ufs/ext2fs/ext2fs_vfsops.c:1.217 src/sys/ufs/ext2fs/ext2fs_vfsops.c:1.218
--- src/sys/ufs/ext2fs/ext2fs_vfsops.c:1.217 Mon Mar 16 21:20:12 2020
+++ src/sys/ufs/ext2fs/ext2fs_vfsops.c Sat Apr 4 20:49:31 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: ext2fs_vfsops.c,v 1.217 2020/03/16 21:20:12 pgoyette Exp $ */
+/* $NetBSD: ext2fs_vfsops.c,v 1.218 2020/04/04 20:49:31 ad Exp $ */
/*
* Copyright (c) 1989, 1991, 1993, 1994
@@ -60,7 +60,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ext2fs_vfsops.c,v 1.217 2020/03/16 21:20:12 pgoyette Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ext2fs_vfsops.c,v 1.218 2020/04/04 20:49:31 ad Exp $");
#if defined(_KERNEL_OPT)
#include "opt_compat_netbsd.h"
@@ -738,7 +738,7 @@ ext2fs_mountfs(struct vnode *devvp, stru
mp->mnt_flag |= MNT_LOCAL;
mp->mnt_dev_bshift = DEV_BSHIFT; /* XXX */
mp->mnt_fs_bshift = m_fs->e2fs_bshift;
- mp->mnt_iflag |= IMNT_DTYPE;
+ mp->mnt_iflag |= IMNT_DTYPE | IMNT_SHRLOOKUP;
ump->um_flags = 0;
ump->um_mountp = mp;
ump->um_dev = dev;
Index: src/sys/ufs/ffs/ffs_vfsops.c
diff -u src/sys/ufs/ffs/ffs_vfsops.c:1.366 src/sys/ufs/ffs/ffs_vfsops.c:1.367
--- src/sys/ufs/ffs/ffs_vfsops.c:1.366 Mon Mar 16 21:20:12 2020
+++ src/sys/ufs/ffs/ffs_vfsops.c Sat Apr 4 20:49:31 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: ffs_vfsops.c,v 1.366 2020/03/16 21:20:12 pgoyette Exp $ */
+/* $NetBSD: ffs_vfsops.c,v 1.367 2020/04/04 20:49:31 ad Exp $ */
/*-
* Copyright (c) 2008, 2009 The NetBSD Foundation, Inc.
@@ -61,7 +61,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ffs_vfsops.c,v 1.366 2020/03/16 21:20:12 pgoyette Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ffs_vfsops.c,v 1.367 2020/04/04 20:49:31 ad Exp $");
#if defined(_KERNEL_OPT)
#include "opt_ffs.h"
@@ -1454,7 +1454,8 @@ ffs_mountfs(struct vnode *devvp, struct
mp->mnt_fs_bshift = fs->fs_bshift;
mp->mnt_dev_bshift = DEV_BSHIFT; /* XXX */
mp->mnt_flag |= MNT_LOCAL;
- mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO;
+ mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO | IMNT_SHRLOOKUP |
+ IMNT_NCLOOKUP;
#ifdef FFS_EI
if (needswap)
ump->um_flags |= UFS_NEEDSWAP;
@@ -2083,6 +2084,7 @@ ffs_loadvnode(struct mount *mp, struct v
ip->i_gid = ip->i_ffs1_ogid; /* XXX */
} /* XXX */
uvm_vnp_setsize(vp, ip->i_size);
+ cache_enter_id(vp, ip->i_mode, ip->i_uid, ip->i_gid);
*new_key = &ip->i_number;
return 0;
}
@@ -2204,6 +2206,7 @@ ffs_newvnode(struct mount *mp, struct vn
}
uvm_vnp_setsize(vp, ip->i_size);
+ cache_enter_id(vp, ip->i_mode, ip->i_uid, ip->i_gid);
*new_key = &ip->i_number;
return 0;
}
Index: src/sys/ufs/lfs/lfs_vfsops.c
diff -u src/sys/ufs/lfs/lfs_vfsops.c:1.377 src/sys/ufs/lfs/lfs_vfsops.c:1.378
--- src/sys/ufs/lfs/lfs_vfsops.c:1.377 Mon Mar 16 21:20:13 2020
+++ src/sys/ufs/lfs/lfs_vfsops.c Sat Apr 4 20:49:31 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: lfs_vfsops.c,v 1.377 2020/03/16 21:20:13 pgoyette Exp $ */
+/* $NetBSD: lfs_vfsops.c,v 1.378 2020/04/04 20:49:31 ad Exp $ */
/*-
* Copyright (c) 1999, 2000, 2001, 2002, 2003, 2007, 2007
@@ -61,7 +61,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.377 2020/03/16 21:20:13 pgoyette Exp $");
+__KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.378 2020/04/04 20:49:31 ad Exp $");
#if defined(_KERNEL_OPT)
#include "opt_lfs.h"
@@ -1125,6 +1125,7 @@ lfs_mountfs(struct vnode *devvp, struct
mp->mnt_stat.f_namemax = LFS_MAXNAMLEN;
mp->mnt_stat.f_iosize = lfs_sb_getbsize(fs);
mp->mnt_flag |= MNT_LOCAL;
+ mp->mnt_iflag |= IMNT_SHRLOOKUP;
mp->mnt_fs_bshift = lfs_sb_getbshift(fs);
mp->mnt_iflag |= IMNT_CAN_RWTORO;
if (fs->um_maxsymlinklen > 0)
Index: src/sys/ufs/lfs/ulfs_lookup.c
diff -u src/sys/ufs/lfs/ulfs_lookup.c:1.42 src/sys/ufs/lfs/ulfs_lookup.c:1.43
--- src/sys/ufs/lfs/ulfs_lookup.c:1.42 Sat Mar 14 18:08:40 2020
+++ src/sys/ufs/lfs/ulfs_lookup.c Sat Apr 4 20:49:31 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: ulfs_lookup.c,v 1.42 2020/03/14 18:08:40 ad Exp $ */
+/* $NetBSD: ulfs_lookup.c,v 1.43 2020/04/04 20:49:31 ad Exp $ */
/* from NetBSD: ufs_lookup.c,v 1.135 2015/07/11 11:04:48 mlelstv */
/*
@@ -38,7 +38,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ulfs_lookup.c,v 1.42 2020/03/14 18:08:40 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ulfs_lookup.c,v 1.43 2020/04/04 20:49:31 ad Exp $");
#ifdef _KERNEL_OPT
#include "opt_lfs.h"
@@ -162,14 +162,6 @@ ulfs_lookup(void *v)
endsearch = 0; /* silence compiler warning */
/*
- * Produce the auxiliary lookup results into i_crap. Increment
- * its serial number so elsewhere we can tell if we're using
- * stale results. This should not be done this way. XXX.
- */
- results = &dp->i_crap;
- dp->i_crapcounter++;
-
- /*
* Check accessiblity of directory.
*/
if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
@@ -193,6 +185,19 @@ ulfs_lookup(void *v)
}
return *vpp == NULLVP ? ENOENT : 0;
}
+
+ /* May need to restart the lookup with an exclusive lock. */
+ if (VOP_ISLOCKED(vdp) != LK_EXCLUSIVE)
+ return ENOLCK;
+
+ /*
+ * Produce the auxiliary lookup results into i_crap. Increment
+ * its serial number so elsewhere we can tell if we're using
+ * stale results. This should not be done this way. XXX.
+ */
+ results = &dp->i_crap;
+ dp->i_crapcounter++;
+
if (iswhiteout) {
/*
* The namecache set iswhiteout without finding a
Index: src/sys/ufs/ufs/ufs_lookup.c
diff -u src/sys/ufs/ufs/ufs_lookup.c:1.151 src/sys/ufs/ufs/ufs_lookup.c:1.152
--- src/sys/ufs/ufs/ufs_lookup.c:1.151 Sat Mar 14 18:08:40 2020
+++ src/sys/ufs/ufs/ufs_lookup.c Sat Apr 4 20:49:31 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: ufs_lookup.c,v 1.151 2020/03/14 18:08:40 ad Exp $ */
+/* $NetBSD: ufs_lookup.c,v 1.152 2020/04/04 20:49:31 ad Exp $ */
/*
* Copyright (c) 1989, 1993
@@ -37,7 +37,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ufs_lookup.c,v 1.151 2020/03/14 18:08:40 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ufs_lookup.c,v 1.152 2020/04/04 20:49:31 ad Exp $");
#ifdef _KERNEL_OPT
#include "opt_ffs.h"
@@ -330,14 +330,6 @@ ufs_lookup(void *v)
endsearch = 0; /* silence compiler warning */
/*
- * Produce the auxiliary lookup results into i_crap. Increment
- * its serial number so elsewhere we can tell if we're using
- * stale results. This should not be done this way. XXX.
- */
- results = &dp->i_crap;
- dp->i_crapcounter++;
-
- /*
* Check accessiblity of directory.
*/
if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
@@ -361,6 +353,20 @@ ufs_lookup(void *v)
}
return *vpp == NULLVP ? ENOENT : 0;
}
+
+ /* May need to restart the lookup with an exclusive lock. */
+ if (VOP_ISLOCKED(vdp) != LK_EXCLUSIVE) {
+ return ENOLCK;
+ }
+
+ /*
+ * Produce the auxiliary lookup results into i_crap. Increment
+ * its serial number so elsewhere we can tell if we're using
+ * stale results. This should not be done this way. XXX.
+ */
+ results = &dp->i_crap;
+ dp->i_crapcounter++;
+
if (iswhiteout) {
/*
* The namecache set iswhiteout without finding a
Index: src/sys/ufs/ufs/ufs_vnops.c
diff -u src/sys/ufs/ufs/ufs_vnops.c:1.249 src/sys/ufs/ufs/ufs_vnops.c:1.250
--- src/sys/ufs/ufs/ufs_vnops.c:1.249 Wed Feb 26 18:00:12 2020
+++ src/sys/ufs/ufs/ufs_vnops.c Sat Apr 4 20:49:31 2020
@@ -1,7 +1,7 @@
-/* $NetBSD: ufs_vnops.c,v 1.249 2020/02/26 18:00:12 maxv Exp $ */
+/* $NetBSD: ufs_vnops.c,v 1.250 2020/04/04 20:49:31 ad Exp $ */
/*-
- * Copyright (c) 2008 The NetBSD Foundation, Inc.
+ * Copyright (c) 2008, 2020 The NetBSD Foundation, Inc.
* All rights reserved.
*
* This code is derived from software contributed to The NetBSD Foundation
@@ -66,7 +66,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ufs_vnops.c,v 1.249 2020/02/26 18:00:12 maxv Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ufs_vnops.c,v 1.250 2020/04/04 20:49:31 ad Exp $");
#if defined(_KERNEL_OPT)
#include "opt_ffs.h"
@@ -621,6 +621,7 @@ ufs_setattr(void *v)
}
VN_KNOTE(vp, NOTE_ATTRIB);
out:
+ cache_enter_id(vp, ip->i_mode, ip->i_uid, ip->i_gid);
return (error);
}
@@ -648,6 +649,7 @@ ufs_chmod(struct vnode *vp, int mode, ka
ip->i_flag |= IN_CHANGE;
DIP_ASSIGN(ip, mode, ip->i_mode);
UFS_WAPBL_UPDATE(vp, NULL, NULL, 0);
+ cache_enter_id(vp, ip->i_mode, ip->i_uid, ip->i_gid);
return (0);
}
@@ -708,6 +710,7 @@ ufs_chown(struct vnode *vp, uid_t uid, g
#endif /* QUOTA || QUOTA2 */
ip->i_flag |= IN_CHANGE;
UFS_WAPBL_UPDATE(vp, NULL, NULL, 0);
+ cache_enter_id(vp, ip->i_mode, ip->i_uid, ip->i_gid);
return (0);
}
