xfs: allocate xattr buffer on demand

When doing file lookups and checking for permissions, we end up in
xfs_get_acl() to see if there are any ACLs on the inode. This
requires and xattr lookup, and to do that we have to supply a buffer
large enough to hold an maximum sized xattr.

On workloads were we are accessing a wide range of cache cold files
under memory pressure (e.g. NFS fileservers) we end up spending a
lot of time allocating the buffer. The buffer is 64k in length, so
is a contiguous multi-page allocation, and if that then fails we
fall back to vmalloc(). Hence the allocation here is /expensive/
when we are looking up hundreds of thousands of files a second.

Initial numbers from a bpf trace show average time in xfs_get_acl()
is ~32us, with ~19us of that in the memory allocation. Note these
are average times, so there are going to be affected by the worst
case allocations more than the common fast case...

To avoid this, we could just do a "null"  lookup to see if the ACL
xattr exists and then only do the allocation if it exists. This,
however, optimises the path for the "no ACL present" case at the
expense of the "acl present" case. i.e. we can halve the time in
xfs_get_acl() for the no acl case (i.e down to ~10-15us), but that
then increases the ACL case by 30% (i.e. up to 40-45us).

To solve this and speed up both cases, drive the xattr buffer
allocation into the attribute code once we know what the actual
xattr length is. For the no-xattr case, we avoid the allocation
completely, speeding up that case. For the common ACL case, we'll
end up with a fast heap allocation (because it'll be smaller than a
page), and only for the rarer "we have a remote xattr" will we have
a multi-page allocation occur. Hence the common ACL case will be
much faster, too.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
This commit is contained in:
Dave Chinner 2019-08-29 09:04:10 -07:00 committed by Darrick J. Wong
parent 9df243a1a9
commit ddbca70cc4
7 changed files with 55 additions and 19 deletions

View File

@ -118,12 +118,28 @@ xfs_attr_get_ilocked(
return xfs_attr_node_get(args); return xfs_attr_node_get(args);
} }
/* Retrieve an extended attribute by name, and its value. */ /*
* Retrieve an extended attribute by name, and its value if requested.
*
* If ATTR_KERNOVAL is set in @flags, then the caller does not want the value,
* just an indication whether the attribute exists and the size of the value if
* it exists. The size is returned in @valuelenp,
*
* If the attribute is found, but exceeds the size limit set by the caller in
* @valuelenp, return -ERANGE with the size of the attribute that was found in
* @valuelenp.
*
* If ATTR_ALLOC is set in @flags, allocate the buffer for the value after
* existence of the attribute has been determined. On success, return that
* buffer to the caller and leave them to free it. On failure, free any
* allocated buffer and ensure the buffer pointer returned to the caller is
* null.
*/
int int
xfs_attr_get( xfs_attr_get(
struct xfs_inode *ip, struct xfs_inode *ip,
const unsigned char *name, const unsigned char *name,
unsigned char *value, unsigned char **value,
int *valuelenp, int *valuelenp,
int flags) int flags)
{ {
@ -131,6 +147,8 @@ xfs_attr_get(
uint lock_mode; uint lock_mode;
int error; int error;
ASSERT((flags & (ATTR_ALLOC | ATTR_KERNOVAL)) || *value);
XFS_STATS_INC(ip->i_mount, xs_attr_get); XFS_STATS_INC(ip->i_mount, xs_attr_get);
if (XFS_FORCED_SHUTDOWN(ip->i_mount)) if (XFS_FORCED_SHUTDOWN(ip->i_mount))
@ -140,17 +158,29 @@ xfs_attr_get(
if (error) if (error)
return error; return error;
args.value = value;
args.valuelen = *valuelenp;
/* Entirely possible to look up a name which doesn't exist */ /* Entirely possible to look up a name which doesn't exist */
args.op_flags = XFS_DA_OP_OKNOENT; args.op_flags = XFS_DA_OP_OKNOENT;
if (flags & ATTR_ALLOC)
args.op_flags |= XFS_DA_OP_ALLOCVAL;
else
args.value = *value;
args.valuelen = *valuelenp;
lock_mode = xfs_ilock_attr_map_shared(ip); lock_mode = xfs_ilock_attr_map_shared(ip);
error = xfs_attr_get_ilocked(ip, &args); error = xfs_attr_get_ilocked(ip, &args);
xfs_iunlock(ip, lock_mode); xfs_iunlock(ip, lock_mode);
*valuelenp = args.valuelen; *valuelenp = args.valuelen;
return error;
/* on error, we have to clean up allocated value buffers */
if (error) {
if (flags & ATTR_ALLOC) {
kmem_free(args.value);
*value = NULL;
}
return error;
}
*value = args.value;
return 0;
} }
/* /*

View File

@ -37,6 +37,7 @@ struct xfs_attr_list_context;
#define ATTR_KERNOVAL 0x2000 /* [kernel] get attr size only, not value */ #define ATTR_KERNOVAL 0x2000 /* [kernel] get attr size only, not value */
#define ATTR_INCOMPLETE 0x4000 /* [kernel] return INCOMPLETE attr keys */ #define ATTR_INCOMPLETE 0x4000 /* [kernel] return INCOMPLETE attr keys */
#define ATTR_ALLOC 0x8000 /* allocate xattr buffer on demand */
#define XFS_ATTR_FLAGS \ #define XFS_ATTR_FLAGS \
{ ATTR_DONTFOLLOW, "DONTFOLLOW" }, \ { ATTR_DONTFOLLOW, "DONTFOLLOW" }, \
@ -47,7 +48,8 @@ struct xfs_attr_list_context;
{ ATTR_REPLACE, "REPLACE" }, \ { ATTR_REPLACE, "REPLACE" }, \
{ ATTR_KERNOTIME, "KERNOTIME" }, \ { ATTR_KERNOTIME, "KERNOTIME" }, \
{ ATTR_KERNOVAL, "KERNOVAL" }, \ { ATTR_KERNOVAL, "KERNOVAL" }, \
{ ATTR_INCOMPLETE, "INCOMPLETE" } { ATTR_INCOMPLETE, "INCOMPLETE" }, \
{ ATTR_ALLOC, "ALLOC" }
/* /*
* The maximum size (into the kernel or returned from the kernel) of an * The maximum size (into the kernel or returned from the kernel) of an
@ -143,7 +145,7 @@ int xfs_attr_list_int(struct xfs_attr_list_context *);
int xfs_inode_hasattr(struct xfs_inode *ip); int xfs_inode_hasattr(struct xfs_inode *ip);
int xfs_attr_get_ilocked(struct xfs_inode *ip, struct xfs_da_args *args); int xfs_attr_get_ilocked(struct xfs_inode *ip, struct xfs_da_args *args);
int xfs_attr_get(struct xfs_inode *ip, const unsigned char *name, int xfs_attr_get(struct xfs_inode *ip, const unsigned char *name,
unsigned char *value, int *valuelenp, int flags); unsigned char **value, int *valuelenp, int flags);
int xfs_attr_set(struct xfs_inode *dp, const unsigned char *name, int xfs_attr_set(struct xfs_inode *dp, const unsigned char *name,
unsigned char *value, int valuelen, int flags); unsigned char *value, int valuelen, int flags);
int xfs_attr_set_args(struct xfs_da_args *args); int xfs_attr_set_args(struct xfs_da_args *args);

View File

@ -414,6 +414,12 @@ xfs_attr_copy_value(
args->valuelen = valuelen; args->valuelen = valuelen;
return -ERANGE; return -ERANGE;
} }
if (args->op_flags & XFS_DA_OP_ALLOCVAL) {
args->value = kmem_alloc_large(valuelen, 0);
if (!args->value)
return -ENOMEM;
}
args->valuelen = valuelen; args->valuelen = valuelen;
/* remote block xattr requires IO for copy-in */ /* remote block xattr requires IO for copy-in */

View File

@ -81,13 +81,15 @@ typedef struct xfs_da_args {
#define XFS_DA_OP_ADDNAME 0x0004 /* this is an add operation */ #define XFS_DA_OP_ADDNAME 0x0004 /* this is an add operation */
#define XFS_DA_OP_OKNOENT 0x0008 /* lookup/add op, ENOENT ok, else die */ #define XFS_DA_OP_OKNOENT 0x0008 /* lookup/add op, ENOENT ok, else die */
#define XFS_DA_OP_CILOOKUP 0x0010 /* lookup to return CI name if found */ #define XFS_DA_OP_CILOOKUP 0x0010 /* lookup to return CI name if found */
#define XFS_DA_OP_ALLOCVAL 0x0020 /* lookup to alloc buffer if found */
#define XFS_DA_OP_FLAGS \ #define XFS_DA_OP_FLAGS \
{ XFS_DA_OP_JUSTCHECK, "JUSTCHECK" }, \ { XFS_DA_OP_JUSTCHECK, "JUSTCHECK" }, \
{ XFS_DA_OP_RENAME, "RENAME" }, \ { XFS_DA_OP_RENAME, "RENAME" }, \
{ XFS_DA_OP_ADDNAME, "ADDNAME" }, \ { XFS_DA_OP_ADDNAME, "ADDNAME" }, \
{ XFS_DA_OP_OKNOENT, "OKNOENT" }, \ { XFS_DA_OP_OKNOENT, "OKNOENT" }, \
{ XFS_DA_OP_CILOOKUP, "CILOOKUP" } { XFS_DA_OP_CILOOKUP, "CILOOKUP" }, \
{ XFS_DA_OP_ALLOCVAL, "ALLOCVAL" }
/* /*
* Storage for holding state during Btree searches and split/join ops. * Storage for holding state during Btree searches and split/join ops.

View File

@ -112,7 +112,7 @@ xfs_get_acl(struct inode *inode, int type)
{ {
struct xfs_inode *ip = XFS_I(inode); struct xfs_inode *ip = XFS_I(inode);
struct posix_acl *acl = NULL; struct posix_acl *acl = NULL;
struct xfs_acl *xfs_acl; struct xfs_acl *xfs_acl = NULL;
unsigned char *ea_name; unsigned char *ea_name;
int error; int error;
int len; int len;
@ -135,12 +135,8 @@ xfs_get_acl(struct inode *inode, int type)
* go out to the disk. * go out to the disk.
*/ */
len = XFS_ACL_MAX_SIZE(ip->i_mount); len = XFS_ACL_MAX_SIZE(ip->i_mount);
xfs_acl = kmem_zalloc_large(len, 0); error = xfs_attr_get(ip, ea_name, (unsigned char **)&xfs_acl, &len,
if (!xfs_acl) ATTR_ALLOC | ATTR_ROOT);
return ERR_PTR(-ENOMEM);
error = xfs_attr_get(ip, ea_name, (unsigned char *)xfs_acl,
&len, ATTR_ROOT);
if (error) { if (error) {
/* /*
* If the attribute doesn't exist make sure we have a negative * If the attribute doesn't exist make sure we have a negative
@ -151,8 +147,8 @@ xfs_get_acl(struct inode *inode, int type)
} else { } else {
acl = xfs_acl_from_disk(xfs_acl, len, acl = xfs_acl_from_disk(xfs_acl, len,
XFS_ACL_MAX_ENTRIES(ip->i_mount)); XFS_ACL_MAX_ENTRIES(ip->i_mount));
kmem_free(xfs_acl);
} }
kmem_free(xfs_acl);
return acl; return acl;
} }

View File

@ -438,7 +438,7 @@ xfs_attrmulti_attr_get(
if (!kbuf) if (!kbuf)
return -ENOMEM; return -ENOMEM;
error = xfs_attr_get(XFS_I(inode), name, kbuf, (int *)len, flags); error = xfs_attr_get(XFS_I(inode), name, &kbuf, (int *)len, flags);
if (error) if (error)
goto out_kfree; goto out_kfree;

View File

@ -30,7 +30,7 @@ xfs_xattr_get(const struct xattr_handler *handler, struct dentry *unused,
value = NULL; value = NULL;
} }
error = xfs_attr_get(ip, (unsigned char *)name, value, &asize, xflags); error = xfs_attr_get(ip, name, (unsigned char **)&value, &asize, xflags);
if (error) if (error)
return error; return error;
return asize; return asize;