Archive-Team
/
zfs
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases 103 Wiki Activity

Share zfs_fsync, zfs_read, zfs_write, et al between Linux and FreeBSD 

The zfs_fsync, zfs_read, and zfs_write function are almost identical
between Linux and FreeBSD.  With a little refactoring they can be
moved to the common code which is what is done by this commit.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Matt Macy <mmacy@FreeBSD.org>
Closes #11078
This commit is contained in:
Matthew Macy 2020-10-21 14:08:06 -07:00 committed by GitHub
parent 666aa69f32
commit e53d678d4a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
29 changed files with 768 additions and 1268 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
include/os/freebsd/spl/sys/misc.h
View File

@ -53,4 +53,7 @@ struct opensolaris_utsname {
extern char hw_serial[11];
#define task_io_account_read(n)
#define task_io_account_write(n)
#endif /* _OPENSOLARIS_SYS_MISC_H_ */

3
include/os/freebsd/spl/sys/policy.h
View File

@ -34,6 +34,7 @@
#include <sys/vnode.h>
struct mount;
struct vattr;
struct znode;
int secpolicy_nfs(cred_t *cr);
int secpolicy_zfs(cred_t *crd);
@ -57,7 +58,7 @@ int secpolicy_vnode_setattr(cred_t *cr, vnode_t *vp, struct vattr *vap,
int unlocked_access(void *, int, cred_t *), void *node);
int secpolicy_vnode_create_gid(cred_t *cr);
int secpolicy_vnode_setids_setgids(vnode_t *vp, cred_t *cr, gid_t gid);
int secpolicy_vnode_setid_retain(vnode_t *vp, cred_t *cr,
int secpolicy_vnode_setid_retain(struct znode *zp, cred_t *cr,
boolean_t issuidroot);
void secpolicy_setid_clear(struct vattr *vap, vnode_t *vp, cred_t *cr);
int secpolicy_setid_setsticky_clear(vnode_t *vp, struct vattr *vap,

1
include/os/freebsd/spl/sys/uio.h
View File

@ -82,6 +82,7 @@ void uioskip(uio_t *uiop, size_t n);
#define uio_iovcnt(uio) (uio)->uio_iovcnt
#define uio_iovlen(uio, idx) (uio)->uio_iov[(idx)].iov_len
#define uio_iovbase(uio, idx) (uio)->uio_iov[(idx)].iov_base
#define uio_fault_disable(uio, set)
static inline void
uio_iov_at_index(uio_t *uio, uint_t idx, void **base, uint64_t *len)

2
include/os/freebsd/zfs/sys/Makefile.am
View File

@ -8,7 +8,7 @@ KERNEL_H = \
zfs_dir.h \
zfs_ioctl_compat.h \
zfs_vfsops_os.h \
zfs_vnops.h \
zfs_vnops_os.h \
zfs_znode_impl.h \
zpl.h

1
include/os/freebsd/zfs/sys/zfs_context_os.h
View File

@ -42,6 +42,7 @@
#include <linux/types.h>
#define cond_resched() kern_yield(PRI_USER)
#define uio_prefaultpages(size, uio) (0)
#define taskq_create_sysdc(a, b, d, e, p, dc, f) \
(taskq_create(a, b, maxclsyspri, d, e, f))

5
include/os/freebsd/zfs/sys/zfs_vnops.h → include/os/freebsd/zfs/sys/zfs_vnops_os.h
View File

@ -26,8 +26,9 @@
* $FreeBSD$
*/
#ifndef _SYS_ZFS_VNOPS_H_
#define _SYS_ZFS_VNOPS_H_
#ifndef _SYS_FS_ZFS_VNOPS_OS_H
#define _SYS_FS_ZFS_VNOPS_OS_H
int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
uint64_t size, struct vm_page **ppa, dmu_tx_t *tx);
int dmu_read_pages(objset_t *os, uint64_t object, vm_page_t *ma, int count,

6
include/os/freebsd/zfs/sys/zfs_znode_impl.h
View File

@ -39,6 +39,7 @@
#include <sys/zfs_acl.h>
#include <sys/zil.h>
#include <sys/zfs_project.h>
#include <vm/vm_object.h>
#ifdef __cplusplus
extern "C" {
@ -113,7 +114,10 @@ extern minor_t zfsdev_minor_alloc(void);
#define Z_ISBLK(type) ((type) == VBLK)
#define Z_ISCHR(type) ((type) == VCHR)
#define Z_ISLNK(type) ((type) == VLNK)
#define Z_ISDIR(type) ((type) == VDIR)
#define zn_has_cached_data(zp) vn_has_cached_data(ZTOV(zp))
#define zn_rlimit_fsize(zp, uio, td) vn_rlimit_fsize(ZTOV(zp), (uio), (td))
/* Called on entry to each ZFS vnode and vfs operation */
#define ZFS_ENTER(zfsvfs) \
@ -175,7 +179,7 @@ extern int zfsfstype;
extern int zfs_znode_parent_and_name(struct znode *zp, struct znode **dzpp,
char *buf);
extern void zfs_inode_update(struct znode *);
#ifdef __cplusplus
}
#endif

1
include/os/linux/kernel/linux/mod_compat.h
View File

@ -74,6 +74,7 @@ enum scope_prefix_types {
zfs_vdev_cache,
zfs_vdev_file,
zfs_vdev_mirror,
zfs_vnops,
zfs_zevent,
zfs_zio,
zfs_zil

2
include/os/linux/spl/sys/uio.h
View File

@ -59,7 +59,6 @@ typedef struct uio {
boolean_t uio_fault_disable;
uint16_t uio_fmode;
uint16_t uio_extflg;
offset_t uio_limit;
ssize_t uio_resid;
size_t uio_skip;
} uio_t;
@ -113,6 +112,7 @@ typedef struct xuio {
#define uio_iovcnt(uio) (uio)->uio_iovcnt
#define uio_iovlen(uio, idx) (uio)->uio_iov[(idx)].iov_len
#define uio_iovbase(uio, idx) (uio)->uio_iov[(idx)].iov_base
#define uio_fault_disable(uio, set) (uio)->uio_fault_disable = set
static inline void
uio_iov_at_index(uio_t *uio, uint_t idx, void **base, uint64_t *len)

2
include/os/linux/zfs/sys/Makefile.am
View File

@ -21,7 +21,7 @@ KERNEL_H = \
zfs_ctldir.h \
zfs_dir.h \
zfs_vfsops_os.h \
zfs_vnops.h \
zfs_vnops_os.h \
zfs_znode_impl.h \
zpl.h

4
include/os/linux/zfs/sys/policy.h
View File

@ -35,6 +35,8 @@
#include <sys/xvattr.h>
#include <sys/zpl.h>
struct znode;
int secpolicy_nfs(const cred_t *);
int secpolicy_sys_config(const cred_t *, boolean_t);
int secpolicy_vnode_access2(const cred_t *, struct inode *,
@ -44,7 +46,7 @@ int secpolicy_vnode_chown(const cred_t *, uid_t);
int secpolicy_vnode_create_gid(const cred_t *);
int secpolicy_vnode_remove(const cred_t *);
int secpolicy_vnode_setdac(const cred_t *, uid_t);
int secpolicy_vnode_setid_retain(const cred_t *, boolean_t);
int secpolicy_vnode_setid_retain(struct znode *, const cred_t *, boolean_t);
int secpolicy_vnode_setids_setgids(const cred_t *, gid_t);
int secpolicy_zinject(const cred_t *);
int secpolicy_zfs(const cred_t *);

11
include/os/linux/zfs/sys/zfs_vnops.h → include/os/linux/zfs/sys/zfs_vnops_os.h
View File

@ -22,8 +22,8 @@
* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
*/
#ifndef _SYS_FS_ZFS_VNOPS_H
#define _SYS_FS_ZFS_VNOPS_H
#ifndef _SYS_FS_ZFS_VNOPS_OS_H
#define _SYS_FS_ZFS_VNOPS_OS_H
#include <sys/vnode.h>
#include <sys/xvattr.h>
@ -41,8 +41,6 @@ extern "C" {
extern int zfs_open(struct inode *ip, int mode, int flag, cred_t *cr);
extern int zfs_close(struct inode *ip, int flag, cred_t *cr);
extern int zfs_holey(struct inode *ip, int cmd, loff_t *off);
extern int zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr);
extern int zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr);
extern int zfs_write_simple(znode_t *zp, const void *data, size_t len,
loff_t pos, size_t *resid);
extern int zfs_access(struct inode *ip, int mode, int flag, cred_t *cr);
@ -58,7 +56,6 @@ extern int zfs_mkdir(znode_t *dzp, char *dirname, vattr_t *vap,
extern int zfs_rmdir(znode_t *dzp, char *name, znode_t *cwd,
cred_t *cr, int flags);
extern int zfs_readdir(struct inode *ip, zpl_dir_context_t *ctx, cred_t *cr);
extern int zfs_fsync(znode_t *zp, int syncflag, cred_t *cr);
extern int zfs_getattr_fast(struct inode *ip, struct kstat *sp);
extern int zfs_setattr(znode_t *zp, vattr_t *vap, int flag, cred_t *cr);
extern int zfs_rename(znode_t *sdzp, char *snm, znode_t *tdzp,
@ -72,10 +69,6 @@ extern void zfs_inactive(struct inode *ip);
extern int zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag,
offset_t offset, cred_t *cr);
extern int zfs_fid(struct inode *ip, fid_t *fidp);
extern int zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag,
cred_t *cr);
extern int zfs_setsecattr(znode_t *zp, vsecattr_t *vsecp, int flag,
cred_t *cr);
extern int zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages);
extern int zfs_putpage(struct inode *ip, struct page *pp,
struct writeback_control *wbc);

6
include/os/linux/zfs/sys/zfs_znode_impl.h
View File

@ -68,6 +68,10 @@ extern "C" {
#define Z_ISCHR(type) S_ISCHR(type)
#define Z_ISLNK(type) S_ISLNK(type)
#define Z_ISDEV(type) (S_ISCHR(type) || S_ISBLK(type) || S_ISFIFO(type))
#define Z_ISDIR(type) S_ISDIR(type)
#define zn_has_cached_data(zp) ((zp)->z_is_mapped)
#define zn_rlimit_fsize(zp, uio, td) (0)
#define zhold(zp) igrab(ZTOI((zp)))
#define zrele(zp) iput(ZTOI((zp)))
@ -143,6 +147,8 @@ do { \
} while (0)
#endif /* HAVE_INODE_TIMESPEC64_TIMES */
#define ZFS_ACCESSTIME_STAMP(zfsvfs, zp)
struct znode;
extern int zfs_sync(struct super_block *, int, cred_t *);

1
include/sys/Makefile.am
View File

@ -117,6 +117,7 @@ COMMON_H = \
zfs_stat.h \
zfs_sysfs.h \
zfs_vfsops.h \
zfs_vnops.h \
zfs_znode.h \
zil.h \
zil_impl.h \

39
include/sys/zfs_vnops.h Normal file
View File

@ -0,0 +1,39 @@
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
*/
#ifndef _SYS_FS_ZFS_VNOPS_H
#define _SYS_FS_ZFS_VNOPS_H
#include <sys/zfs_vnops_os.h>
extern int zfs_fsync(znode_t *, int, cred_t *);
extern int zfs_read(znode_t *, uio_t *, int, cred_t *);
extern int zfs_write(znode_t *, uio_t *, int, cred_t *);
extern int zfs_getsecattr(znode_t *zp, vsecattr_t *vsecp, int flag, cred_t *cr);
extern int zfs_setsecattr(znode_t *zp, vsecattr_t *vsecp, int flag, cred_t *cr);
extern int mappedread(znode_t *, int, uio_t *);
extern int mappedread_sf(znode_t *, int, uio_t *);
extern void update_pages(znode_t *, int64_t, int, objset_t *, uint64_t);
#endif

1
lib/libspl/include/sys/uio.h
View File

@ -73,7 +73,6 @@ typedef struct uio {
uio_seg_t uio_segflg; /* address space (kernel or user) */
uint16_t uio_fmode; /* file mode flags */
uint16_t uio_extflg; /* extended flags */
offset_t uio_limit; /* u-limit (maximum byte offset) */
ssize_t uio_resid; /* residual count */
} uio_t;

27
module/Makefile.bsd
View File

@ -101,9 +101,10 @@ SRCS+= nvpair.c \
#os/freebsd/spl
SRCS+= acl_common.c \
btree.c \
callb.c \
list.c \
sha256c.c \
sha512c.c \
spl_acl.c \
spl_cmn_err.c \
spl_dtrace.c \
@ -111,6 +112,7 @@ SRCS+= acl_common.c \
spl_kstat.c \
spl_misc.c \
spl_policy.c \
spl_procfs_list.c \
spl_string.c \
spl_sunddi.c \
spl_sysevent.c \
@ -118,11 +120,8 @@ SRCS+= acl_common.c \
spl_uio.c \
spl_vfs.c \
spl_vm.c \
spl_zone.c \
sha256c.c \
sha512c.c \
spl_procfs_list.c \
spl_zlib.c
spl_zlib.c \
spl_zone.c
.if ${MACHINE_ARCH} == "i386" || ${MACHINE_ARCH} == "powerpc" || \
@ -132,6 +131,7 @@ SRCS+= spl_atomic.c
#os/freebsd/zfs
SRCS+= abd_os.c \
arc_os.c \
crypto_os.c \
dmu_os.c \
hkdf.c \
@ -139,17 +139,16 @@ SRCS+= abd_os.c \
spa_os.c \
sysctl_os.c \
vdev_file.c \
vdev_label_os.c \
vdev_geom.c \
vdev_label_os.c \
zfs_acl.c \
zfs_ctldir.c \
zfs_debug.c \
zfs_dir.c \
zfs_ioctl_compat.c \
zfs_ioctl_os.c \
zfs_log.c \
zfs_replay.c \
zfs_vfsops.c \
zfs_vnops.c \
zfs_vnops_os.c \
zfs_znode.c \
zio_crypt.c \
zvol_os.c
@ -177,10 +176,10 @@ SRCS+= zfeature_common.c \
SRCS+= abd.c \
aggsum.c \
arc.c \
arc_os.c \
blkptr.c \
bplist.c \
bpobj.c \
btree.c \
cityhash.c \
dbuf.c \
dbuf_stats.c \
@ -275,16 +274,18 @@ SRCS+= abd.c \
zcp_synctask.c \
zfeature.c \
zfs_byteswap.c \
zfs_debug.c \
zfs_file_os.c \
zfs_fm.c \
zfs_fuid.c \
zfs_ioctl.c \
zfs_log.c \
zfs_onexit.c \
zfs_quota.c \
zfs_ratelimit.c \
zfs_replay.c \
zfs_rlock.c \
zfs_sa.c \
zfs_vnops.c \
zil.c \
zio.c \
zio_checksum.c \
@ -322,7 +323,7 @@ CFLAGS.spl_vm.c= -Wno-cast-qual
CFLAGS.spl_zlib.c= -Wno-cast-qual
CFLAGS.abd.c= -Wno-cast-qual
CFLAGS.zfs_log.c= -Wno-cast-qual
CFLAGS.zfs_vnops.c= -Wno-pointer-arith
CFLAGS.zfs_vnops_os.c= -Wno-pointer-arith
CFLAGS.u8_textprep.c= -Wno-cast-qual
CFLAGS.zfs_fletcher.c= -Wno-cast-qual -Wno-pointer-arith
CFLAGS.zfs_fletcher_intel.c= -Wno-cast-qual -Wno-pointer-arith

5
module/os/freebsd/spl/spl_policy.c
View File

@ -37,6 +37,7 @@ __FBSDID("$FreeBSD$");
#include <sys/jail.h>
#include <sys/policy.h>
#include <sys/zfs_vfsops.h>
#include <sys/zfs_znode.h>
int
@ -312,11 +313,11 @@ secpolicy_vnode_setids_setgids(vnode_t *vp, cred_t *cr, gid_t gid)
}
int
secpolicy_vnode_setid_retain(vnode_t *vp, cred_t *cr,
secpolicy_vnode_setid_retain(znode_t *zp, cred_t *cr,
boolean_t issuidroot __unused)
{
if (secpolicy_fs_owner(vp->v_mount, cr) == 0)
if (secpolicy_fs_owner(ZTOV(zp)->v_mount, cr) == 0)
return (0);
return (spl_priv_check_cred(cr, PRIV_VFS_RETAINSUGID));
}

1
module/os/freebsd/zfs/sysctl_os.c
View File

@ -114,6 +114,7 @@ SYSCTL_NODE(_vfs_zfs, OID_AUTO, spa, CTLFLAG_RW, 0, "ZFS space allocation");
SYSCTL_NODE(_vfs_zfs, OID_AUTO, trim, CTLFLAG_RW, 0, "ZFS TRIM");
SYSCTL_NODE(_vfs_zfs, OID_AUTO, txg, CTLFLAG_RW, 0, "ZFS transaction group");
SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
SYSCTL_NODE(_vfs_zfs, OID_AUTO, vnops, CTLFLAG_RW, 0, "ZFS VNOPS");
SYSCTL_NODE(_vfs_zfs, OID_AUTO, zevent, CTLFLAG_RW, 0, "ZFS event");
SYSCTL_NODE(_vfs_zfs, OID_AUTO, zil, CTLFLAG_RW, 0, "ZFS ZIL");
SYSCTL_NODE(_vfs_zfs, OID_AUTO, zio, CTLFLAG_RW, 0, "ZFS ZIO");

605
module/os/freebsd/zfs/zfs_vnops.c → module/os/freebsd/zfs/zfs_vnops_os.c
View File

@ -525,16 +525,15 @@ page_unhold(vm_page_t pp)
* On Write: If we find a memory mapped page, we write to *both*
* the page and the dmu buffer.
*/
static void
update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
int segflg, dmu_tx_t *tx)
void
update_pages(znode_t *zp, int64_t start, int len, objset_t *os, uint64_t oid)
{
vm_object_t obj;
struct sf_buf *sf;
vnode_t *vp = ZTOV(zp);
caddr_t va;
int off;
ASSERT(segflg != UIO_NOCOPY);
ASSERT(vp->v_mount != NULL);
obj = vp->v_object;
ASSERT(obj != NULL);
@ -579,10 +578,10 @@ update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
* map them into contiguous KVA region and populate them
* in one single dmu_read() call.
*/
static int
mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
int
mappedread_sf(znode_t *zp, int nbytes, uio_t *uio)
{
znode_t *zp = VTOZ(vp);
vnode_t *vp = ZTOV(zp);
objset_t *os = zp->z_zfsvfs->z_os;
struct sf_buf *sf;
vm_object_t obj;
@ -664,10 +663,10 @@ mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
* NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
* the file is memory mapped.
*/
static int
mappedread(vnode_t *vp, int nbytes, uio_t *uio)
int
mappedread(znode_t *zp, int nbytes, uio_t *uio)
{
znode_t *zp = VTOZ(vp);
vnode_t *vp = ZTOV(zp);
vm_object_t obj;
int64_t start;
int len = nbytes;
@ -710,523 +709,6 @@ mappedread(vnode_t *vp, int nbytes, uio_t *uio)
return (error);
}
offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
/*
* Read bytes from specified file into supplied buffer.
*
* IN: vp - vnode of file to be read from.
* uio - structure supplying read location, range info,
* and return buffer.
* ioflag - SYNC flags; used to provide FRSYNC semantics.
* cr - credentials of caller.
* ct - caller context
*
* OUT: uio - updated offset and range, buffer filled.
*
* RETURN: 0 on success, error code on failure.
*
* Side Effects:
* vp - atime updated if byte count > 0
*/
/* ARGSUSED */
static int
zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr)
{
znode_t *zp = VTOZ(vp);
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
ssize_t n, nbytes, start_resid;
int error = 0;
int64_t nread;
zfs_locked_range_t *lr;
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
/* We don't copy out anything useful for directories. */
if (vp->v_type == VDIR) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EISDIR));
}
if (zp->z_pflags & ZFS_AV_QUARANTINED) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EACCES));
}
/*
* Validate file offset
*/
if (uio->uio_loffset < (offset_t)0) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EINVAL));
}
/*
* Fasttrack empty reads
*/
if (uio->uio_resid == 0) {
ZFS_EXIT(zfsvfs);
return (0);
}
/*
* If we're in FRSYNC mode, sync out this znode before reading it.
*/
if (zfsvfs->z_log &&
(ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
zil_commit(zfsvfs->z_log, zp->z_id);
/*
* Lock the range against changes.
*/
lr = zfs_rangelock_enter(&zp->z_rangelock, uio->uio_loffset,
uio->uio_resid, RL_READER);
/*
* If we are reading past end-of-file we can skip
* to the end; but we might still need to set atime.
*/
if (uio->uio_loffset >= zp->z_size) {
error = 0;
goto out;
}
ASSERT(uio->uio_loffset < zp->z_size);
n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
start_resid = n;
while (n > 0) {
nbytes = MIN(n, zfs_read_chunk_size -
P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
if (uio->uio_segflg == UIO_NOCOPY)
error = mappedread_sf(vp, nbytes, uio);
else if (vn_has_cached_data(vp)) {
error = mappedread(vp, nbytes, uio);
} else {
error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
uio, nbytes);
}
if (error) {
/* convert checksum errors into IO errors */
if (error == ECKSUM)
error = SET_ERROR(EIO);
break;
}
n -= nbytes;
}
nread = start_resid - n;
dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, nread);
out:
zfs_rangelock_exit(lr);
ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
ZFS_EXIT(zfsvfs);
return (error);
}
/*
* Write the bytes to a file.
*
* IN: vp - vnode of file to be written to.
* uio - structure supplying write location, range info,
* and data buffer.
* ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
* set if in append mode.
* cr - credentials of caller.
* ct - caller context (NFS/CIFS fem monitor only)
*
* OUT: uio - updated offset and range.
*
* RETURN: 0 on success, error code on failure.
*
* Timestamps:
* vp - ctime|mtime updated if byte count > 0
*/
/* ARGSUSED */
static int
zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr)
{
znode_t *zp = VTOZ(vp);
rlim64_t limit = MAXOFFSET_T;
ssize_t start_resid = uio->uio_resid;
ssize_t tx_bytes;
uint64_t end_size;
dmu_buf_impl_t *db;
dmu_tx_t *tx;
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
zilog_t *zilog;
offset_t woff;
ssize_t n, nbytes;
zfs_locked_range_t *lr;
int max_blksz = zfsvfs->z_max_blksz;
int error = 0;
arc_buf_t *abuf;
iovec_t *aiov = NULL;
xuio_t *xuio = NULL;
int i_iov = 0;
int iovcnt __unused = uio->uio_iovcnt;
iovec_t *iovp = uio->uio_iov;
int write_eof;
int count = 0;
sa_bulk_attr_t bulk[4];
uint64_t mtime[2], ctime[2];
uint64_t uid, gid, projid;
int64_t nwritten;
/*
* Fasttrack empty write
*/
n = start_resid;
if (n == 0)
return (0);
if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
limit = MAXOFFSET_T;
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
&zp->z_size, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
&zp->z_pflags, 8);
/*
* Callers might not be able to detect properly that we are read-only,
* so check it explicitly here.
*/
if (zfs_is_readonly(zfsvfs)) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EROFS));
}
/*
* If immutable or not appending then return EPERM.
* Intentionally allow ZFS_READONLY through here.
* See zfs_zaccess_common()
*/
if ((zp->z_pflags & ZFS_IMMUTABLE) ||
((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
(uio->uio_loffset < zp->z_size))) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EPERM));
}
zilog = zfsvfs->z_log;
/*
* Validate file offset
*/
woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
if (woff < 0) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EINVAL));
}
/*
* If in append mode, set the io offset pointer to eof.
*/
if (ioflag & FAPPEND) {
/*
* Obtain an appending range lock to guarantee file append
* semantics. We reset the write offset once we have the lock.
*/
lr = zfs_rangelock_enter(&zp->z_rangelock, 0, n, RL_APPEND);
woff = lr->lr_offset;
if (lr->lr_length == UINT64_MAX) {
/*
* We overlocked the file because this write will cause
* the file block size to increase.
* Note that zp_size cannot change with this lock held.
*/
woff = zp->z_size;
}
uio->uio_loffset = woff;
} else {
/*
* Note that if the file block size will change as a result of
* this write, then this range lock will lock the entire file
* so that we can re-write the block safely.
*/
lr = zfs_rangelock_enter(&zp->z_rangelock, woff, n, RL_WRITER);
}
if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
zfs_rangelock_exit(lr);
ZFS_EXIT(zfsvfs);
return (EFBIG);
}
if (woff >= limit) {
zfs_rangelock_exit(lr);
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EFBIG));
}
if ((woff + n) > limit || woff > (limit - n))
n = limit - woff;
/* Will this write extend the file length? */
write_eof = (woff + n > zp->z_size);
end_size = MAX(zp->z_size, woff + n);
uid = zp->z_uid;
gid = zp->z_gid;
projid = zp->z_projid;
/*
* Write the file in reasonable size chunks. Each chunk is written
* in a separate transaction; this keeps the intent log records small
* and allows us to do more fine-grained space accounting.
*/
while (n > 0) {
woff = uio->uio_loffset;
if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, uid) ||
zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, gid) ||
(projid != ZFS_DEFAULT_PROJID &&
zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
projid))) {
error = SET_ERROR(EDQUOT);
break;
}
abuf = NULL;
if (xuio) {
ASSERT(i_iov < iovcnt);
aiov = &iovp[i_iov];
abuf = dmu_xuio_arcbuf(xuio, i_iov);
dmu_xuio_clear(xuio, i_iov);
DTRACE_PROBE3(zfs_cp_write, int, i_iov,
iovec_t *, aiov, arc_buf_t *, abuf);
ASSERT((aiov->iov_base == abuf->b_data) ||
((char *)aiov->iov_base - (char *)abuf->b_data +
aiov->iov_len == arc_buf_size(abuf)));
i_iov++;
} else if (n >= max_blksz &&
woff >= zp->z_size &&
P2PHASE(woff, max_blksz) == 0 &&
zp->z_blksz == max_blksz) {
/*
* This write covers a full block. "Borrow" a buffer
* from the dmu so that we can fill it before we enter
* a transaction. This avoids the possibility of
* holding up the transaction if the data copy hangs
* up on a pagefault (e.g., from an NFS server mapping).
*/
size_t cbytes;
abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
max_blksz);
ASSERT(abuf != NULL);
ASSERT(arc_buf_size(abuf) == max_blksz);
if ((error = uiocopy(abuf->b_data, max_blksz,
UIO_WRITE, uio, &cbytes))) {
dmu_return_arcbuf(abuf);
break;
}
ASSERT(cbytes == max_blksz);
}
/*
* Start a transaction.
*/
tx = dmu_tx_create(zfsvfs->z_os);
dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
db = (dmu_buf_impl_t *)sa_get_db(zp->z_sa_hdl);
DB_DNODE_ENTER(db);
dmu_tx_hold_write_by_dnode(tx, DB_DNODE(db), woff,
MIN(n, max_blksz));
DB_DNODE_EXIT(db);
zfs_sa_upgrade_txholds(tx, zp);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
if (abuf != NULL)
dmu_return_arcbuf(abuf);
break;
}
/*
* If zfs_range_lock() over-locked we grow the blocksize
* and then reduce the lock range. This will only happen
* on the first iteration since zfs_range_reduce() will
* shrink down r_len to the appropriate size.
*/
if (lr->lr_length == UINT64_MAX) {
uint64_t new_blksz;
if (zp->z_blksz > max_blksz) {
/*
* File's blocksize is already larger than the
* "recordsize" property. Only let it grow to
* the next power of 2.
*/
ASSERT(!ISP2(zp->z_blksz));
new_blksz = MIN(end_size,
1 << highbit64(zp->z_blksz));
} else {
new_blksz = MIN(end_size, max_blksz);
}
zfs_grow_blocksize(zp, new_blksz, tx);
zfs_rangelock_reduce(lr, woff, n);
}
/*
* XXX - should we really limit each write to z_max_blksz?
* Perhaps we should use SPA_MAXBLOCKSIZE chunks?
*/
nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
if (woff + nbytes > zp->z_size)
vnode_pager_setsize(vp, woff + nbytes);
if (abuf == NULL) {
tx_bytes = uio->uio_resid;
error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
uio, nbytes, tx);
tx_bytes -= uio->uio_resid;
} else {
tx_bytes = nbytes;
ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
/*
* If this is not a full block write, but we are
* extending the file past EOF and this data starts
* block-aligned, use assign_arcbuf(). Otherwise,
* write via dmu_write().
*/
if (tx_bytes < max_blksz && (!write_eof ||
aiov->iov_base != abuf->b_data)) {
ASSERT(xuio);
dmu_write(zfsvfs->z_os, zp->z_id, woff,
aiov->iov_len, aiov->iov_base, tx);
dmu_return_arcbuf(abuf);
xuio_stat_wbuf_copied();
} else {
ASSERT(xuio || tx_bytes == max_blksz);
dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), woff,
abuf, tx);
}
ASSERT(tx_bytes <= uio->uio_resid);
uioskip(uio, tx_bytes);
}
if (tx_bytes && vn_has_cached_data(vp)) {
update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
zp->z_id, uio->uio_segflg, tx);
}
/*
* If we made no progress, we're done. If we made even
* partial progress, update the znode and ZIL accordingly.
*/
if (tx_bytes == 0) {
(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
(void *)&zp->z_size, sizeof (uint64_t), tx);
dmu_tx_commit(tx);
ASSERT(error != 0);
break;
}
/*
* Clear Set-UID/Set-GID bits on successful write if not
* privileged and at least one of the execute bits is set.
*
* It would be nice to to this after all writes have
* been done, but that would still expose the ISUID/ISGID
* to another app after the partial write is committed.
*
* Note: we don't call zfs_fuid_map_id() here because
* user 0 is not an ephemeral uid.
*/
mutex_enter(&zp->z_acl_lock);
if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
(S_IXUSR >> 6))) != 0 &&
(zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
secpolicy_vnode_setid_retain(vp, cr,
(zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
uint64_t newmode;
zp->z_mode &= ~(S_ISUID | S_ISGID);
newmode = zp->z_mode;
(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
(void *)&newmode, sizeof (uint64_t), tx);
}
mutex_exit(&zp->z_acl_lock);
zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
/*
* Update the file size (zp_size) if it has changed;
* account for possible concurrent updates.
*/
while ((end_size = zp->z_size) < uio->uio_loffset) {
(void) atomic_cas_64(&zp->z_size, end_size,
uio->uio_loffset);
ASSERT(error == 0 || error == EFAULT);
}
/*
* If we are replaying and eof is non zero then force
* the file size to the specified eof. Note, there's no
* concurrency during replay.
*/
if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
zp->z_size = zfsvfs->z_replay_eof;
if (error == 0)
error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
else
(void) sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes,
ioflag, NULL, NULL);
dmu_tx_commit(tx);
if (error != 0)
break;
ASSERT(tx_bytes == nbytes);
n -= nbytes;
}
zfs_rangelock_exit(lr);
/*
* If we're in replay mode, or we made no progress, return error.
* Otherwise, it's at least a partial write, so it's successful.
*/
if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
ZFS_EXIT(zfsvfs);
return (error);
}
/*
* EFAULT means that at least one page of the source buffer was not
* available. VFS will re-try remaining I/O upon this error.
*/
if (error == EFAULT) {
ZFS_EXIT(zfsvfs);
return (error);
}
if (ioflag & (FSYNC | FDSYNC) ||
zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
zil_commit(zilog, zp->z_id);
nwritten = start_resid - uio->uio_resid;
dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, nwritten);
ZFS_EXIT(zfsvfs);
return (0);
}
int
zfs_write_simple(znode_t *zp, const void *data, size_t len,
loff_t pos, size_t *presid)
@ -2704,27 +2186,6 @@ update:
return (error);
}
ulong_t zfs_fsync_sync_cnt = 4;
static int
zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
{
znode_t *zp = VTOZ(vp);
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
zil_commit(zfsvfs->z_log, zp->z_id);
ZFS_EXIT(zfsvfs);
}
tsd_set(zfs_fsyncer_key, NULL);
return (0);
}
/*
* Get the requested file attributes and place them in the provided
* vattr structure.
@ -4789,45 +4250,6 @@ zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
}
}
/*ARGSUSED*/
static int
zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
caller_context_t *ct)
{
znode_t *zp = VTOZ(vp);
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
int error;
boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
error = zfs_getacl(zp, vsecp, skipaclchk, cr);
ZFS_EXIT(zfsvfs);
return (error);
}
/*ARGSUSED*/
int
zfs_setsecattr(znode_t *zp, vsecattr_t *vsecp, int flag, cred_t *cr)
{
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
int error;
boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
zilog_t *zilog = zfsvfs->z_log;
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
error = zfs_setacl(zp, vsecp, skipaclchk, cr);
if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
zil_commit(zilog, 0);
ZFS_EXIT(zfsvfs);
return (error);
}
static int
zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind,
int *rahead)
@ -5221,7 +4643,7 @@ static int
zfs_freebsd_read(struct vop_read_args *ap)
{
return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
return (zfs_read(VTOZ(ap->a_vp), ap->a_uio, ioflags(ap->a_ioflag),
ap->a_cred));
}
@ -5238,7 +4660,7 @@ static int
zfs_freebsd_write(struct vop_write_args *ap)
{
return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
return (zfs_write(VTOZ(ap->a_vp), ap->a_uio, ioflags(ap->a_ioflag),
ap->a_cred));
}
@ -5508,7 +4930,7 @@ zfs_freebsd_fsync(struct vop_fsync_args *ap)
{
vop_stdfsync(ap);
return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
return (zfs_fsync(VTOZ(ap->a_vp), 0, ap->a_td->td_ucred));
}
#ifndef _SYS_SYSPROTO_H_
@ -6374,7 +5796,8 @@ zfs_freebsd_getacl(struct vop_getacl_args *ap)
return (EINVAL);
vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
if ((error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)))
if ((error = zfs_getsecattr(VTOZ(ap->a_vp),
&vsecattr, 0, ap->a_cred)))
return (error);
error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp,

14
module/os/freebsd/zfs/zfs_znode.c
View File

@ -2034,6 +2034,20 @@ zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
return (error);
}
void
zfs_inode_update(znode_t *zp)
{
vm_object_t object;
if ((object = ZTOV(zp)->v_object) == NULL ||
zp->z_size == object->un_pager.vnp.vnp_size)
return;
vnode_pager_setsize(ZTOV(zp), zp->z_size);
}
#ifdef _KERNEL
int
zfs_znode_parent_and_name(znode_t *zp, znode_t **dzpp, char *buf)

2
module/os/linux/zfs/Makefile.in
View File

@ -24,7 +24,7 @@ $(MODULE)-objs += ../os/linux/zfs/zfs_file_os.o
$(MODULE)-objs += ../os/linux/zfs/zfs_ioctl_os.o
$(MODULE)-objs += ../os/linux/zfs/zfs_sysfs.o
$(MODULE)-objs += ../os/linux/zfs/zfs_vfsops.o
$(MODULE)-objs += ../os/linux/zfs/zfs_vnops.o
$(MODULE)-objs += ../os/linux/zfs/zfs_vnops_os.o
$(MODULE)-objs += ../os/linux/zfs/zfs_znode.o
$(MODULE)-objs += ../os/linux/zfs/zio_crypt.o
$(MODULE)-objs += ../os/linux/zfs/zpl_ctldir.o

5
module/os/linux/zfs/policy.c
View File

@ -204,7 +204,8 @@ secpolicy_vnode_setdac(const cred_t *cr, uid_t owner)
* Enforced in the Linux VFS.
*/
int
secpolicy_vnode_setid_retain(const cred_t *cr, boolean_t issuidroot)
secpolicy_vnode_setid_retain(struct znode *zp __maybe_unused, const cred_t *cr,
boolean_t issuidroot)
{
return (priv_policy_user(cr, CAP_FSETID, EPERM));
}
@ -271,7 +272,7 @@ void
secpolicy_setid_clear(vattr_t *vap, cred_t *cr)
{
if ((vap->va_mode & (S_ISUID | S_ISGID)) != 0 &&
secpolicy_vnode_setid_retain(cr,
secpolicy_vnode_setid_retain(NULL, cr,
(vap->va_mode & S_ISUID) != 0 &&
(vap->va_mask & AT_UID) != 0 && vap->va_uid == 0) != 0) {
vap->va_mask |= AT_MODE;

643
module/os/linux/zfs/zfs_vnops.c → module/os/linux/zfs/zfs_vnops_os.c
View File

@ -320,10 +320,11 @@ zfs_holey(struct inode *ip, int cmd, loff_t *off)
* On Write: If we find a memory mapped page, we write to *both*
* the page and the dmu buffer.
*/
static void
update_pages(struct inode *ip, int64_t start, int len,
void
update_pages(znode_t *zp, int64_t start, int len,
objset_t *os, uint64_t oid)
{
struct inode *ip = ZTOI(zp);
struct address_space *mp = ip->i_mapping;
struct page *pp;
uint64_t nbytes;
@ -369,12 +370,12 @@ update_pages(struct inode *ip, int64_t start, int len,
* NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
* the file is memory mapped.
*/
static int
mappedread(struct inode *ip, int nbytes, uio_t *uio)
int
mappedread(znode_t *zp, int nbytes, uio_t *uio)
{
struct inode *ip = ZTOI(zp);
struct address_space *mp = ip->i_mapping;
struct page *pp;
znode_t *zp = ITOZ(ip);
int64_t start, off;
uint64_t bytes;
int len = nbytes;
@ -414,574 +415,8 @@ mappedread(struct inode *ip, int nbytes, uio_t *uio)
}
#endif /* _KERNEL */
unsigned long zfs_read_chunk_size = 1024 * 1024; /* Tunable */
unsigned long zfs_delete_blocks = DMU_MAX_DELETEBLKCNT;
/*
* Read bytes from specified file into supplied buffer.
*
* IN: ip - inode of file to be read from.
* uio - structure supplying read location, range info,
* and return buffer.
* ioflag - O_SYNC flags; used to provide FRSYNC semantics.
* O_DIRECT flag; used to bypass page cache.
* cr - credentials of caller.
*
* OUT: uio - updated offset and range, buffer filled.
*
* RETURN: 0 on success, error code on failure.
*
* Side Effects:
* inode - atime updated if byte count > 0
*/
/* ARGSUSED */
int
zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
{
int error = 0;
boolean_t frsync = B_FALSE;
znode_t *zp = ITOZ(ip);
zfsvfs_t *zfsvfs = ITOZSB(ip);
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
if (zp->z_pflags & ZFS_AV_QUARANTINED) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EACCES));
}
/*
* Validate file offset
*/
if (uio->uio_loffset < (offset_t)0) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EINVAL));
}
/*
* Fasttrack empty reads
*/
if (uio->uio_resid == 0) {
ZFS_EXIT(zfsvfs);
return (0);
}
#ifdef FRSYNC
/*
* If we're in FRSYNC mode, sync out this znode before reading it.
* Only do this for non-snapshots.
*
* Some platforms do not support FRSYNC and instead map it
* to O_SYNC, which results in unnecessary calls to zil_commit. We
* only honor FRSYNC requests on platforms which support it.
*/
frsync = !!(ioflag & FRSYNC);
#endif
if (zfsvfs->z_log &&
(frsync || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
zil_commit(zfsvfs->z_log, zp->z_id);
/*
* Lock the range against changes.
*/
zfs_locked_range_t *lr = zfs_rangelock_enter(&zp->z_rangelock,
uio->uio_loffset, uio->uio_resid, RL_READER);
/*
* If we are reading past end-of-file we can skip
* to the end; but we might still need to set atime.
*/
if (uio->uio_loffset >= zp->z_size) {
error = 0;
goto out;
}
ASSERT(uio->uio_loffset < zp->z_size);
ssize_t n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
ssize_t start_resid = n;
#ifdef HAVE_UIO_ZEROCOPY
xuio_t *xuio = NULL;
if ((uio->uio_extflg == UIO_XUIO) &&
(((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
int nblk;
int blksz = zp->z_blksz;
uint64_t offset = uio->uio_loffset;
xuio = (xuio_t *)uio;
if ((ISP2(blksz))) {
nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
blksz)) / blksz;
} else {
ASSERT(offset + n <= blksz);
nblk = 1;
}
(void) dmu_xuio_init(xuio, nblk);
if (vn_has_cached_data(ip)) {
/*
* For simplicity, we always allocate a full buffer
* even if we only expect to read a portion of a block.
*/
while (--nblk >= 0) {
(void) dmu_xuio_add(xuio,
dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
blksz), 0, blksz);
}
}
}
#endif /* HAVE_UIO_ZEROCOPY */
while (n > 0) {
ssize_t nbytes = MIN(n, zfs_read_chunk_size -
P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
if (zp->z_is_mapped && !(ioflag & O_DIRECT)) {
error = mappedread(ip, nbytes, uio);
} else {
error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
uio, nbytes);
}
if (error) {
/* convert checksum errors into IO errors */
if (error == ECKSUM)
error = SET_ERROR(EIO);
break;
}
n -= nbytes;
}
int64_t nread = start_resid - n;
dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, nread);
task_io_account_read(nread);
out:
zfs_rangelock_exit(lr);
ZFS_EXIT(zfsvfs);
return (error);
}
/*
* Write the bytes to a file.
*
* IN: ip - inode of file to be written to.
* uio - structure supplying write location, range info,
* and data buffer.
* ioflag - O_APPEND flag set if in append mode.
* O_DIRECT flag; used to bypass page cache.
* cr - credentials of caller.
*
* OUT: uio - updated offset and range.
*
* RETURN: 0 if success
* error code if failure
*
* Timestamps:
* ip - ctime|mtime updated if byte count > 0
*/
/* ARGSUSED */
int
zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
{
int error = 0;
ssize_t start_resid = uio->uio_resid;
/*
* Fasttrack empty write
*/
ssize_t n = start_resid;
if (n == 0)
return (0);
rlim64_t limit = uio->uio_limit;
if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
limit = MAXOFFSET_T;
znode_t *zp = ITOZ(ip);
zfsvfs_t *zfsvfs = ZTOZSB(zp);
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
sa_bulk_attr_t bulk[4];
int count = 0;
uint64_t mtime[2], ctime[2];
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
&zp->z_size, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
&zp->z_pflags, 8);
/*
* Callers might not be able to detect properly that we are read-only,
* so check it explicitly here.
*/
if (zfs_is_readonly(zfsvfs)) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EROFS));
}
/*
* If immutable or not appending then return EPERM
*/
if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & O_APPEND) &&
(uio->uio_loffset < zp->z_size))) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EPERM));
}
/*
* Validate file offset
*/
offset_t woff = ioflag & O_APPEND ? zp->z_size : uio->uio_loffset;
if (woff < 0) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EINVAL));
}
int max_blksz = zfsvfs->z_max_blksz;
xuio_t *xuio = NULL;
/*
* Pre-fault the pages to ensure slow (eg NFS) pages
* don't hold up txg.
* Skip this if uio contains loaned arc_buf.
*/
#ifdef HAVE_UIO_ZEROCOPY
if ((uio->uio_extflg == UIO_XUIO) &&
(((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
xuio = (xuio_t *)uio;
else
#endif
if (uio_prefaultpages(MIN(n, max_blksz), uio)) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EFAULT));
}
/*
* If in append mode, set the io offset pointer to eof.
*/
zfs_locked_range_t *lr;
if (ioflag & O_APPEND) {
/*
* Obtain an appending range lock to guarantee file append
* semantics. We reset the write offset once we have the lock.
*/
lr = zfs_rangelock_enter(&zp->z_rangelock, 0, n, RL_APPEND);
woff = lr->lr_offset;
if (lr->lr_length == UINT64_MAX) {
/*
* We overlocked the file because this write will cause
* the file block size to increase.
* Note that zp_size cannot change with this lock held.
*/
woff = zp->z_size;
}
uio->uio_loffset = woff;
} else {
/*
* Note that if the file block size will change as a result of
* this write, then this range lock will lock the entire file
* so that we can re-write the block safely.
*/
lr = zfs_rangelock_enter(&zp->z_rangelock, woff, n, RL_WRITER);
}
if (woff >= limit) {
zfs_rangelock_exit(lr);
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EFBIG));
}
if ((woff + n) > limit || woff > (limit - n))
n = limit - woff;
/* Will this write extend the file length? */
int write_eof = (woff + n > zp->z_size);
uint64_t end_size = MAX(zp->z_size, woff + n);
zilog_t *zilog = zfsvfs->z_log;
#ifdef HAVE_UIO_ZEROCOPY
int i_iov = 0;
const iovec_t *iovp = uio->uio_iov;
int iovcnt __maybe_unused = uio->uio_iovcnt;
#endif
/*
* Write the file in reasonable size chunks. Each chunk is written
* in a separate transaction; this keeps the intent log records small
* and allows us to do more fine-grained space accounting.
*/
while (n > 0) {
woff = uio->uio_loffset;
if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT,
KUID_TO_SUID(ip->i_uid)) ||
zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT,
KGID_TO_SGID(ip->i_gid)) ||
(zp->z_projid != ZFS_DEFAULT_PROJID &&
zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
zp->z_projid))) {
error = SET_ERROR(EDQUOT);
break;
}
arc_buf_t *abuf = NULL;
const iovec_t *aiov = NULL;
if (xuio) {
#ifdef HAVE_UIO_ZEROCOPY
ASSERT(i_iov < iovcnt);
ASSERT3U(uio->uio_segflg, !=, UIO_BVEC);
aiov = &iovp[i_iov];
abuf = dmu_xuio_arcbuf(xuio, i_iov);
dmu_xuio_clear(xuio, i_iov);
ASSERT((aiov->iov_base == abuf->b_data) ||
((char *)aiov->iov_base - (char *)abuf->b_data +
aiov->iov_len == arc_buf_size(abuf)));
i_iov++;
#endif
} else if (n >= max_blksz && woff >= zp->z_size &&
P2PHASE(woff, max_blksz) == 0 &&
zp->z_blksz == max_blksz) {
/*
* This write covers a full block. "Borrow" a buffer
* from the dmu so that we can fill it before we enter
* a transaction. This avoids the possibility of
* holding up the transaction if the data copy hangs
* up on a pagefault (e.g., from an NFS server mapping).
*/
size_t cbytes;
abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
max_blksz);
ASSERT(abuf != NULL);
ASSERT(arc_buf_size(abuf) == max_blksz);
if ((error = uiocopy(abuf->b_data, max_blksz,
UIO_WRITE, uio, &cbytes))) {
dmu_return_arcbuf(abuf);
break;
}
ASSERT(cbytes == max_blksz);
}
/*
* Start a transaction.
*/
dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
dmu_buf_impl_t *db = (dmu_buf_impl_t *)sa_get_db(zp->z_sa_hdl);
DB_DNODE_ENTER(db);
dmu_tx_hold_write_by_dnode(tx, DB_DNODE(db), woff,
MIN(n, max_blksz));
DB_DNODE_EXIT(db);
zfs_sa_upgrade_txholds(tx, zp);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
if (abuf != NULL)
dmu_return_arcbuf(abuf);
break;
}
/*
* If rangelock_enter() over-locked we grow the blocksize
* and then reduce the lock range. This will only happen
* on the first iteration since rangelock_reduce() will
* shrink down lr_length to the appropriate size.
*/
if (lr->lr_length == UINT64_MAX) {
uint64_t new_blksz;
if (zp->z_blksz > max_blksz) {
/*
* File's blocksize is already larger than the
* "recordsize" property. Only let it grow to
* the next power of 2.
*/
ASSERT(!ISP2(zp->z_blksz));
new_blksz = MIN(end_size,
1 << highbit64(zp->z_blksz));
} else {
new_blksz = MIN(end_size, max_blksz);
}
zfs_grow_blocksize(zp, new_blksz, tx);
zfs_rangelock_reduce(lr, woff, n);
}
/*
* XXX - should we really limit each write to z_max_blksz?
* Perhaps we should use SPA_MAXBLOCKSIZE chunks?
*/
ssize_t nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
ssize_t tx_bytes;
if (abuf == NULL) {
tx_bytes = uio->uio_resid;
uio->uio_fault_disable = B_TRUE;
error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
uio, nbytes, tx);
uio->uio_fault_disable = B_FALSE;
if (error == EFAULT) {
dmu_tx_commit(tx);
/*
* Account for partial writes before
* continuing the loop.
* Update needs to occur before the next
* uio_prefaultpages, or prefaultpages may
* error, and we may break the loop early.
*/
if (tx_bytes != uio->uio_resid)
n -= tx_bytes - uio->uio_resid;
if (uio_prefaultpages(MIN(n, max_blksz), uio)) {
break;
}
continue;
} else if (error != 0) {
dmu_tx_commit(tx);
break;
}
tx_bytes -= uio->uio_resid;
} else {
tx_bytes = nbytes;
ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
/*
* If this is not a full block write, but we are
* extending the file past EOF and this data starts
* block-aligned, use assign_arcbuf(). Otherwise,
* write via dmu_write().
*/
if (tx_bytes < max_blksz && (!write_eof ||
aiov->iov_base != abuf->b_data)) {
ASSERT(xuio);
dmu_write(zfsvfs->z_os, zp->z_id, woff,
/* cppcheck-suppress nullPointer */
aiov->iov_len, aiov->iov_base, tx);
dmu_return_arcbuf(abuf);
xuio_stat_wbuf_copied();
} else {
ASSERT(xuio || tx_bytes == max_blksz);
error = dmu_assign_arcbuf_by_dbuf(
sa_get_db(zp->z_sa_hdl), woff, abuf, tx);
if (error != 0) {
dmu_return_arcbuf(abuf);
dmu_tx_commit(tx);
break;
}
}
ASSERT(tx_bytes <= uio->uio_resid);
uioskip(uio, tx_bytes);
}
if (tx_bytes && zp->z_is_mapped && !(ioflag & O_DIRECT)) {
update_pages(ip, woff,
tx_bytes, zfsvfs->z_os, zp->z_id);
}
/*
* If we made no progress, we're done. If we made even
* partial progress, update the znode and ZIL accordingly.
*/
if (tx_bytes == 0) {
(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
(void *)&zp->z_size, sizeof (uint64_t), tx);
dmu_tx_commit(tx);
ASSERT(error != 0);
break;
}
/*
* Clear Set-UID/Set-GID bits on successful write if not
* privileged and at least one of the execute bits is set.
*
* It would be nice to do this after all writes have
* been done, but that would still expose the ISUID/ISGID
* to another app after the partial write is committed.
*
* Note: we don't call zfs_fuid_map_id() here because
* user 0 is not an ephemeral uid.
*/
mutex_enter(&zp->z_acl_lock);
uint32_t uid = KUID_TO_SUID(ip->i_uid);
if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
(S_IXUSR >> 6))) != 0 &&
(zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
secpolicy_vnode_setid_retain(cr,
((zp->z_mode & S_ISUID) != 0 && uid == 0)) != 0) {
uint64_t newmode;
zp->z_mode &= ~(S_ISUID | S_ISGID);
ip->i_mode = newmode = zp->z_mode;
(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
(void *)&newmode, sizeof (uint64_t), tx);
}
mutex_exit(&zp->z_acl_lock);
zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
/*
* Update the file size (zp_size) if it has changed;
* account for possible concurrent updates.
*/
while ((end_size = zp->z_size) < uio->uio_loffset) {
(void) atomic_cas_64(&zp->z_size, end_size,
uio->uio_loffset);
ASSERT(error == 0);
}
/*
* If we are replaying and eof is non zero then force
* the file size to the specified eof. Note, there's no
* concurrency during replay.
*/
if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
zp->z_size = zfsvfs->z_replay_eof;
error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag,
NULL, NULL);
dmu_tx_commit(tx);
if (error != 0)
break;
ASSERT(tx_bytes == nbytes);
n -= nbytes;
if (!xuio && n > 0) {
if (uio_prefaultpages(MIN(n, max_blksz), uio)) {
error = EFAULT;
break;
}
}
}
zfs_inode_update(zp);
zfs_rangelock_exit(lr);
/*
* If we're in replay mode, or we made no progress, return error.
* Otherwise, it's at least a partial write, so it's successful.
*/
if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
ZFS_EXIT(zfsvfs);
return (error);
}
if (ioflag & (O_SYNC | O_DSYNC) ||
zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
zil_commit(zilog, zp->z_id);
int64_t nwritten = start_resid - uio->uio_resid;
dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, nwritten);
task_io_account_write(nwritten);
ZFS_EXIT(zfsvfs);
return (0);
}
/*
* Write the bytes to a file.
*
@ -2440,26 +1875,6 @@ out:
return (error);
}
ulong_t zfs_fsync_sync_cnt = 4;
int
zfs_fsync(znode_t *zp, int syncflag, cred_t *cr)
{
zfsvfs_t *zfsvfs = ZTOZSB(zp);
(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
zil_commit(zfsvfs->z_log, zp->z_id);
ZFS_EXIT(zfsvfs);
}
tsd_set(zfs_fsyncer_key, NULL);
return (0);
}
/*
* Get the basic file attributes and place them in the provided kstat
* structure. The inode is assumed to be the authoritative source
@ -4796,44 +4211,6 @@ zfs_fid(struct inode *ip, fid_t *fidp)
return (0);
}
/*ARGSUSED*/
int
zfs_getsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, cred_t *cr)
{
znode_t *zp = ITOZ(ip);
zfsvfs_t *zfsvfs = ITOZSB(ip);
int error;
boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
error = zfs_getacl(zp, vsecp, skipaclchk, cr);
ZFS_EXIT(zfsvfs);
return (error);
}
/*ARGSUSED*/
int
zfs_setsecattr(znode_t *zp, vsecattr_t *vsecp, int flag, cred_t *cr)
{
zfsvfs_t *zfsvfs = ZTOZSB(zp);
int error;
boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
zilog_t *zilog = zfsvfs->z_log;
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
error = zfs_setacl(zp, vsecp, skipaclchk, cr);
if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
zil_commit(zilog, 0);
ZFS_EXIT(zfsvfs);
return (error);
}
#ifdef HAVE_UIO_ZEROCOPY
/*
* The smallest read we may consider to loan out an arcbuf.
@ -4846,6 +4223,7 @@ int zcr_blksz_min = (1 << 10); /* 1K */
*/
int zcr_blksz_max = (1 << 17); /* 128K */
/*ARGSUSED*/
static int
zfs_reqzcbuf(struct inode *ip, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr)
@ -4994,8 +4372,6 @@ zfs_retzcbuf(struct inode *ip, xuio_t *xuio, cred_t *cr)
#if defined(_KERNEL)
EXPORT_SYMBOL(zfs_open);
EXPORT_SYMBOL(zfs_close);
EXPORT_SYMBOL(zfs_read);
EXPORT_SYMBOL(zfs_write);
EXPORT_SYMBOL(zfs_access);
EXPORT_SYMBOL(zfs_lookup);
EXPORT_SYMBOL(zfs_create);
@ -5004,7 +4380,6 @@ EXPORT_SYMBOL(zfs_remove);
EXPORT_SYMBOL(zfs_mkdir);
EXPORT_SYMBOL(zfs_rmdir);
EXPORT_SYMBOL(zfs_readdir);
EXPORT_SYMBOL(zfs_fsync);
EXPORT_SYMBOL(zfs_getattr_fast);
EXPORT_SYMBOL(zfs_setattr);
EXPORT_SYMBOL(zfs_rename);
@ -5014,8 +4389,6 @@ EXPORT_SYMBOL(zfs_link);
EXPORT_SYMBOL(zfs_inactive);
EXPORT_SYMBOL(zfs_space);
EXPORT_SYMBOL(zfs_fid);
EXPORT_SYMBOL(zfs_getsecattr);
EXPORT_SYMBOL(zfs_setsecattr);
EXPORT_SYMBOL(zfs_getpage);
EXPORT_SYMBOL(zfs_putpage);
EXPORT_SYMBOL(zfs_dirty_inode);
@ -5024,8 +4397,6 @@ EXPORT_SYMBOL(zfs_map);
/* BEGIN CSTYLED */
module_param(zfs_delete_blocks, ulong, 0644);
MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async");
module_param(zfs_read_chunk_size, ulong, 0644);
MODULE_PARM_DESC(zfs_read_chunk_size, "Bytes to read per chunk");
/* END CSTYLED */
#endif

1
module/os/linux/zfs/zfs_znode.c
View File

@ -505,6 +505,7 @@ zfs_inode_update(znode_t *zp)
dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &blksize, &i_blocks);
spin_lock(&ip->i_lock);
ip->i_mode = zp->z_mode;
ip->i_blocks = i_blocks;
i_size_write(ip, zp->z_size);
spin_unlock(&ip->i_lock);

6
module/os/linux/zfs/zpl_file.c
View File

@ -226,12 +226,11 @@ zpl_read_common_iovec(struct inode *ip, const struct iovec *iovp, size_t count,
uio.uio_iovcnt = nr_segs;
uio.uio_loffset = *ppos;
uio.uio_segflg = segment;
uio.uio_limit = MAXOFFSET_T;
uio.uio_resid = count;
uio.uio_skip = skip;
cookie = spl_fstrans_mark();
error = -zfs_read(ip, &uio, flags, cr);
error = -zfs_read(ITOZ(ip), &uio, flags, cr);
spl_fstrans_unmark(cookie);
if (error < 0)
return (error);
@ -339,12 +338,11 @@ zpl_write_common_iovec(struct inode *ip, const struct iovec *iovp, size_t count,
uio.uio_iovcnt = nr_segs;
uio.uio_loffset = *ppos;
uio.uio_segflg = segment;
uio.uio_limit = MAXOFFSET_T;
uio.uio_resid = count;
uio.uio_skip = skip;
cookie = spl_fstrans_mark();
error = -zfs_write(ip, &uio, flags, cr);
error = -zfs_write(ITOZ(ip), &uio, flags, cr);
spl_fstrans_unmark(cookie);
if (error < 0)
return (error);

1
module/os/linux/zfs/zvol_os.c
View File

@ -91,7 +91,6 @@ uio_from_bio(uio_t *uio, struct bio *bio)
uio->uio_iovcnt = bio->bi_vcnt - BIO_BI_IDX(bio);
uio->uio_loffset = BIO_BI_SECTOR(bio) << 9;
uio->uio_segflg = UIO_BVEC;
uio->uio_limit = MAXOFFSET_T;
uio->uio_resid = BIO_BI_SIZE(bio);
uio->uio_skip = BIO_BI_SKIP(bio);
}

1
module/zfs/Makefile.in
View File

@ -120,6 +120,7 @@ $(MODULE)-objs += zfs_ratelimit.o
$(MODULE)-objs += zfs_replay.o
$(MODULE)-objs += zfs_rlock.o
$(MODULE)-objs += zfs_sa.o
$(MODULE)-objs += zfs_vnops.o
$(MODULE)-objs += zil.o
$(MODULE)-objs += zio.o
$(MODULE)-objs += zio_checksum.o

637
module/zfs/zfs_vnops.c Normal file
View File

@ -0,0 +1,637 @@
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012, 2018 by Delphix. All rights reserved.
* Copyright (c) 2015 by Chunwei Chen. All rights reserved.
* Copyright 2017 Nexenta Systems, Inc.
*/
/* Portions Copyright 2007 Jeremy Teo */
/* Portions Copyright 2010 Robert Milkowski */
#include <sys/types.h>
#include <sys/param.h>
#include <sys/time.h>
#include <sys/sysmacros.h>
#include <sys/vfs.h>
#include <sys/uio.h>
#include <sys/file.h>
#include <sys/stat.h>
#include <sys/kmem.h>
#include <sys/cmn_err.h>
#include <sys/errno.h>
#include <sys/zfs_dir.h>
#include <sys/zfs_acl.h>
#include <sys/zfs_ioctl.h>
#include <sys/fs/zfs.h>
#include <sys/dmu.h>
#include <sys/dmu_objset.h>
#include <sys/spa.h>
#include <sys/txg.h>
#include <sys/dbuf.h>
#include <sys/policy.h>
#include <sys/zfs_vnops.h>
#include <sys/zfs_quota.h>
static ulong_t zfs_fsync_sync_cnt = 4;
int
zfs_fsync(znode_t *zp, int syncflag, cred_t *cr)
{
zfsvfs_t *zfsvfs = ZTOZSB(zp);
(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
zil_commit(zfsvfs->z_log, zp->z_id);
ZFS_EXIT(zfsvfs);
}
tsd_set(zfs_fsyncer_key, NULL);
return (0);
}
static unsigned long zfs_vnops_read_chunk_size = 1024 * 1024; /* Tunable */
/*
* Read bytes from specified file into supplied buffer.
*
* IN: zp - inode of file to be read from.
* uio - structure supplying read location, range info,
* and return buffer.
* ioflag - O_SYNC flags; used to provide FRSYNC semantics.
* O_DIRECT flag; used to bypass page cache.
* cr - credentials of caller.
*
* OUT: uio - updated offset and range, buffer filled.
*
* RETURN: 0 on success, error code on failure.
*
* Side Effects:
* inode - atime updated if byte count > 0
*/
/* ARGSUSED */
int
zfs_read(struct znode *zp, uio_t *uio, int ioflag, cred_t *cr)
{
int error = 0;
boolean_t frsync = B_FALSE;
zfsvfs_t *zfsvfs = ZTOZSB(zp);
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
if (zp->z_pflags & ZFS_AV_QUARANTINED) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EACCES));
}
/* We don't copy out anything useful for directories. */
if (Z_ISDIR(ZTOTYPE(zp))) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EISDIR));
}
/*
* Validate file offset
*/
if (uio->uio_loffset < (offset_t)0) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EINVAL));
}
/*
* Fasttrack empty reads
*/
if (uio->uio_resid == 0) {
ZFS_EXIT(zfsvfs);
return (0);
}
#ifdef FRSYNC
/*
* If we're in FRSYNC mode, sync out this znode before reading it.
* Only do this for non-snapshots.
*
* Some platforms do not support FRSYNC and instead map it
* to O_SYNC, which results in unnecessary calls to zil_commit. We
* only honor FRSYNC requests on platforms which support it.
*/
frsync = !!(ioflag & FRSYNC);
#endif
if (zfsvfs->z_log &&
(frsync || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
zil_commit(zfsvfs->z_log, zp->z_id);
/*
* Lock the range against changes.
*/
zfs_locked_range_t *lr = zfs_rangelock_enter(&zp->z_rangelock,
uio->uio_loffset, uio->uio_resid, RL_READER);
/*
* If we are reading past end-of-file we can skip
* to the end; but we might still need to set atime.
*/
if (uio->uio_loffset >= zp->z_size) {
error = 0;
goto out;
}
ASSERT(uio->uio_loffset < zp->z_size);
ssize_t n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
ssize_t start_resid = n;
while (n > 0) {
ssize_t nbytes = MIN(n, zfs_vnops_read_chunk_size -
P2PHASE(uio->uio_loffset, zfs_vnops_read_chunk_size));
#ifdef UIO_NOCOPY
if (uio->uio_segflg == UIO_NOCOPY)
error = mappedread_sf(zp, nbytes, uio);
else
#endif
if (zn_has_cached_data(zp) && !(ioflag & O_DIRECT)) {
error = mappedread(zp, nbytes, uio);
} else {
error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
uio, nbytes);
}
if (error) {
/* convert checksum errors into IO errors */
if (error == ECKSUM)
error = SET_ERROR(EIO);
break;
}
n -= nbytes;
}
int64_t nread = start_resid - n;
dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, nread);
task_io_account_read(nread);
out:
zfs_rangelock_exit(lr);
ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
ZFS_EXIT(zfsvfs);
return (error);
}
/*
* Write the bytes to a file.
*
* IN: zp - znode of file to be written to.
* uio - structure supplying write location, range info,
* and data buffer.
* ioflag - O_APPEND flag set if in append mode.
* O_DIRECT flag; used to bypass page cache.
* cr - credentials of caller.
*
* OUT: uio - updated offset and range.
*
* RETURN: 0 if success
* error code if failure
*
* Timestamps:
* ip - ctime|mtime updated if byte count > 0
*/
/* ARGSUSED */
int
zfs_write(znode_t *zp, uio_t *uio, int ioflag, cred_t *cr)
{
int error = 0;
ssize_t start_resid = uio->uio_resid;
/*
* Fasttrack empty write
*/
ssize_t n = start_resid;
if (n == 0)
return (0);
rlim64_t limit = MAXOFFSET_T;
zfsvfs_t *zfsvfs = ZTOZSB(zp);
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
sa_bulk_attr_t bulk[4];
int count = 0;
uint64_t mtime[2], ctime[2];
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
&zp->z_size, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
&zp->z_pflags, 8);
/*
* Callers might not be able to detect properly that we are read-only,
* so check it explicitly here.
*/
if (zfs_is_readonly(zfsvfs)) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EROFS));
}
/*
* If immutable or not appending then return EPERM
*/
if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & O_APPEND) &&
(uio->uio_loffset < zp->z_size))) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EPERM));
}
/*
* Validate file offset
*/
offset_t woff = ioflag & O_APPEND ? zp->z_size : uio->uio_loffset;
if (woff < 0) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EINVAL));
}
int max_blksz = zfsvfs->z_max_blksz;
/*
* Pre-fault the pages to ensure slow (eg NFS) pages
* don't hold up txg.
* Skip this if uio contains loaned arc_buf.
*/
if (uio_prefaultpages(MIN(n, max_blksz), uio)) {
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EFAULT));
}
/*
* If in append mode, set the io offset pointer to eof.
*/
zfs_locked_range_t *lr;
if (ioflag & O_APPEND) {
/*
* Obtain an appending range lock to guarantee file append
* semantics. We reset the write offset once we have the lock.
*/
lr = zfs_rangelock_enter(&zp->z_rangelock, 0, n, RL_APPEND);
woff = lr->lr_offset;
if (lr->lr_length == UINT64_MAX) {
/*
* We overlocked the file because this write will cause
* the file block size to increase.
* Note that zp_size cannot change with this lock held.
*/
woff = zp->z_size;
}
uio->uio_loffset = woff;
} else {
/*
* Note that if the file block size will change as a result of
* this write, then this range lock will lock the entire file
* so that we can re-write the block safely.
*/
lr = zfs_rangelock_enter(&zp->z_rangelock, woff, n, RL_WRITER);
}
if (zn_rlimit_fsize(zp, uio, uio->uio_td)) {
zfs_rangelock_exit(lr);
ZFS_EXIT(zfsvfs);
return (EFBIG);
}
if (woff >= limit) {
zfs_rangelock_exit(lr);
ZFS_EXIT(zfsvfs);
return (SET_ERROR(EFBIG));
}
if ((woff + n) > limit || woff > (limit - n))
n = limit - woff;
uint64_t end_size = MAX(zp->z_size, woff + n);
zilog_t *zilog = zfsvfs->z_log;
/*
* Write the file in reasonable size chunks. Each chunk is written
* in a separate transaction; this keeps the intent log records small
* and allows us to do more fine-grained space accounting.
*/
while (n > 0) {
woff = uio->uio_loffset;
if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT,
KUID_TO_SUID(ZTOUID(zp))) ||
zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT,
KGID_TO_SGID(ZTOGID(zp))) ||
(zp->z_projid != ZFS_DEFAULT_PROJID &&
zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
zp->z_projid))) {
error = SET_ERROR(EDQUOT);
break;
}
arc_buf_t *abuf = NULL;
if (n >= max_blksz && woff >= zp->z_size &&
P2PHASE(woff, max_blksz) == 0 &&
zp->z_blksz == max_blksz) {
/*
* This write covers a full block. "Borrow" a buffer
* from the dmu so that we can fill it before we enter
* a transaction. This avoids the possibility of
* holding up the transaction if the data copy hangs
* up on a pagefault (e.g., from an NFS server mapping).
*/
size_t cbytes;
abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
max_blksz);
ASSERT(abuf != NULL);
ASSERT(arc_buf_size(abuf) == max_blksz);
if ((error = uiocopy(abuf->b_data, max_blksz,
UIO_WRITE, uio, &cbytes))) {
dmu_return_arcbuf(abuf);
break;
}
ASSERT(cbytes == max_blksz);
}
/*
* Start a transaction.
*/
dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
dmu_buf_impl_t *db = (dmu_buf_impl_t *)sa_get_db(zp->z_sa_hdl);
DB_DNODE_ENTER(db);
dmu_tx_hold_write_by_dnode(tx, DB_DNODE(db), woff,
MIN(n, max_blksz));
DB_DNODE_EXIT(db);
zfs_sa_upgrade_txholds(tx, zp);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
if (abuf != NULL)
dmu_return_arcbuf(abuf);
break;
}
/*
* If rangelock_enter() over-locked we grow the blocksize
* and then reduce the lock range. This will only happen
* on the first iteration since rangelock_reduce() will
* shrink down lr_length to the appropriate size.
*/
if (lr->lr_length == UINT64_MAX) {
uint64_t new_blksz;
if (zp->z_blksz > max_blksz) {
/*
* File's blocksize is already larger than the
* "recordsize" property. Only let it grow to
* the next power of 2.
*/
ASSERT(!ISP2(zp->z_blksz));
new_blksz = MIN(end_size,
1 << highbit64(zp->z_blksz));
} else {
new_blksz = MIN(end_size, max_blksz);
}
zfs_grow_blocksize(zp, new_blksz, tx);
zfs_rangelock_reduce(lr, woff, n);
}
/*
* XXX - should we really limit each write to z_max_blksz?
* Perhaps we should use SPA_MAXBLOCKSIZE chunks?
*/
ssize_t nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
ssize_t tx_bytes;
if (abuf == NULL) {
tx_bytes = uio->uio_resid;
uio_fault_disable(uio, B_TRUE);
error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
uio, nbytes, tx);
uio_fault_disable(uio, B_FALSE);
#ifdef __linux__
if (error == EFAULT) {
dmu_tx_commit(tx);
/*
* Account for partial writes before
* continuing the loop.
* Update needs to occur before the next
* uio_prefaultpages, or prefaultpages may
* error, and we may break the loop early.
*/
if (tx_bytes != uio->uio_resid)
n -= tx_bytes - uio->uio_resid;
if (uio_prefaultpages(MIN(n, max_blksz), uio)) {
break;
}
continue;
}
#endif
if (error != 0) {
dmu_tx_commit(tx);
break;
}
tx_bytes -= uio->uio_resid;
} else {
/*
* Is this block ever reached?
*/
tx_bytes = nbytes;
/*
* If this is not a full block write, but we are
* extending the file past EOF and this data starts
* block-aligned, use assign_arcbuf(). Otherwise,
* write via dmu_write().
*/
if (tx_bytes == max_blksz) {
error = dmu_assign_arcbuf_by_dbuf(
sa_get_db(zp->z_sa_hdl), woff, abuf, tx);
if (error != 0) {
dmu_return_arcbuf(abuf);
dmu_tx_commit(tx);
break;
}
}
ASSERT(tx_bytes <= uio->uio_resid);
uioskip(uio, tx_bytes);
}
if (tx_bytes && zn_has_cached_data(zp) &&
!(ioflag & O_DIRECT)) {
update_pages(zp, woff,
tx_bytes, zfsvfs->z_os, zp->z_id);
}
/*
* If we made no progress, we're done. If we made even
* partial progress, update the znode and ZIL accordingly.
*/
if (tx_bytes == 0) {
(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
(void *)&zp->z_size, sizeof (uint64_t), tx);
dmu_tx_commit(tx);
ASSERT(error != 0);
break;
}
/*
* Clear Set-UID/Set-GID bits on successful write if not
* privileged and at least one of the execute bits is set.
*
* It would be nice to do this after all writes have
* been done, but that would still expose the ISUID/ISGID
* to another app after the partial write is committed.
*
* Note: we don't call zfs_fuid_map_id() here because
* user 0 is not an ephemeral uid.
*/
mutex_enter(&zp->z_acl_lock);
uint32_t uid = KUID_TO_SUID(ZTOUID(zp));
if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
(S_IXUSR >> 6))) != 0 &&
(zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
secpolicy_vnode_setid_retain(zp, cr,
((zp->z_mode & S_ISUID) != 0 && uid == 0)) != 0) {
uint64_t newmode;
zp->z_mode &= ~(S_ISUID | S_ISGID);
(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
(void *)&newmode, sizeof (uint64_t), tx);
}
mutex_exit(&zp->z_acl_lock);
zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
/*
* Update the file size (zp_size) if it has changed;
* account for possible concurrent updates.
*/
while ((end_size = zp->z_size) < uio->uio_loffset) {
(void) atomic_cas_64(&zp->z_size, end_size,
uio->uio_loffset);
ASSERT(error == 0);
}
/*
* If we are replaying and eof is non zero then force
* the file size to the specified eof. Note, there's no
* concurrency during replay.
*/
if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
zp->z_size = zfsvfs->z_replay_eof;
error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag,
NULL, NULL);
dmu_tx_commit(tx);
if (error != 0)
break;
ASSERT(tx_bytes == nbytes);
n -= nbytes;
if (n > 0) {
if (uio_prefaultpages(MIN(n, max_blksz), uio)) {
error = EFAULT;
break;
}
}
}
zfs_inode_update(zp);
zfs_rangelock_exit(lr);
/*
* If we're in replay mode, or we made no progress, return error.
* Otherwise, it's at least a partial write, so it's successful.
*/
if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
ZFS_EXIT(zfsvfs);
return (error);
}
if (ioflag & (O_SYNC | O_DSYNC) ||
zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
zil_commit(zilog, zp->z_id);
int64_t nwritten = start_resid - uio->uio_resid;
dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, nwritten);
task_io_account_write(nwritten);
ZFS_EXIT(zfsvfs);
return (0);
}
/*ARGSUSED*/
int
zfs_getsecattr(znode_t *zp, vsecattr_t *vsecp, int flag, cred_t *cr)
{
zfsvfs_t *zfsvfs = ZTOZSB(zp);
int error;
boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
error = zfs_getacl(zp, vsecp, skipaclchk, cr);
ZFS_EXIT(zfsvfs);
return (error);
}
/*ARGSUSED*/
int
zfs_setsecattr(znode_t *zp, vsecattr_t *vsecp, int flag, cred_t *cr)
{
zfsvfs_t *zfsvfs = ZTOZSB(zp);
int error;
boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
zilog_t *zilog = zfsvfs->z_log;
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(zp);
error = zfs_setacl(zp, vsecp, skipaclchk, cr);
if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
zil_commit(zilog, 0);
ZFS_EXIT(zfsvfs);
return (error);
}
EXPORT_SYMBOL(zfs_fsync);
EXPORT_SYMBOL(zfs_read);
EXPORT_SYMBOL(zfs_write);
EXPORT_SYMBOL(zfs_getsecattr);
EXPORT_SYMBOL(zfs_setsecattr);
ZFS_MODULE_PARAM(zfs_vnops, zfs_vnops_, read_chunk_size, ULONG, ZMOD_RW,
"Bytes to read per chunk");