d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Continue on-going campaign to replace lockmgr locks with sx locks where

Browse Source 
the specific semantics of ockmgr aren't required: update UFS1 extended
attributes to protect its data structures using an sx lock.

While here, update comments on lock granularity.

MFC after:	2 weeks
This commit is contained in:
rwatson 2008-03-04 12:50:11 +00:00
parent 96767d9090
commit 48aef5a3f0
2 changed files with 14 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
sys/ufs/ufs/extattr.h
View File

@ -113,6 +113,8 @@ struct extattr {
#ifdef _KERNEL
#include <sys/_sx.h>
#ifdef MALLOC_DECLARE
MALLOC_DECLARE(M_EXTATTR);
#endif
@ -127,10 +129,9 @@ struct ufs_extattr_list_entry {
struct vnode *uele_backing_vnode;
};
struct lock;
struct ucred;
struct ufs_extattr_per_mount {
struct lock uepm_lock;
struct sx uepm_lock;
struct ufs_extattr_list_head uepm_list;
struct ucred *uepm_ucred;
int uepm_flags;

21
sys/ufs/ufs/ufs_extattr.c
View File

@ -55,6 +55,7 @@ __FBSDID("$FreeBSD$");
#include <sys/lock.h>
#include <sys/dirent.h>
#include <sys/extattr.h>
#include <sys/sx.h>
#include <sys/sysctl.h>
#include <vm/uma.h>
@ -95,23 +96,23 @@ static int ufs_extattr_rm(struct vnode *vp, int attrnamespace,
/*
* Per-FS attribute lock protecting attribute operations.
* XXX Right now there is a lot of lock contention due to having a single
* lock per-FS; really, this should be far more fine-grained.
*
* XXXRW: Perhaps something more fine-grained would be appropriate, but at
* the end of the day we're going to contend on the vnode lock for the
* backing file anyway.
*/
static void
ufs_extattr_uepm_lock(struct ufsmount *ump, struct thread *td)
{
/* Ideally, LK_CANRECURSE would not be used, here. */
lockmgr(&ump->um_extattr.uepm_lock, LK_EXCLUSIVE | LK_RETRY |
LK_CANRECURSE, 0);
sx_xlock(&ump->um_extattr.uepm_lock);
}
static void
ufs_extattr_uepm_unlock(struct ufsmount *ump, struct thread *td)
{
lockmgr(&ump->um_extattr.uepm_lock, LK_RELEASE, 0);
sx_xunlock(&ump->um_extattr.uepm_lock);
}
/*-
@ -143,6 +144,8 @@ ufs_extattr_find_attr(struct ufsmount *ump, int attrnamespace,
{
struct ufs_extattr_list_entry *search_attribute;
sx_assert(&ump->um_extattr.uepm_lock, SA_XLOCKED);
for (search_attribute = LIST_FIRST(&ump->um_extattr.uepm_list);
search_attribute != NULL;
search_attribute = LIST_NEXT(search_attribute, uele_entries)) {
@ -165,10 +168,8 @@ ufs_extattr_uepm_init(struct ufs_extattr_per_mount *uepm)
{
uepm->uepm_flags = 0;
LIST_INIT(&uepm->uepm_list);
/* XXX is PVFS right, here? */
lockinit(&uepm->uepm_lock, PVFS, "extattr", 0, 0);
sx_init(&uepm->uepm_lock, "ufs_extattr_sx");
uepm->uepm_flags |= UFS_EXTATTR_UEPM_INITIALIZED;
}
@ -192,7 +193,7 @@ ufs_extattr_uepm_destroy(struct ufs_extattr_per_mount *uepm)
* during unmount, and with vfs_busy().
*/
uepm->uepm_flags &= ~UFS_EXTATTR_UEPM_INITIALIZED;
lockdestroy(&uepm->uepm_lock);
sx_destroy(&uepm->uepm_lock);
}
/*