int
zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
{
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
znode_t *xzp;
dmu_tx_t *tx;
int error;
zfs_fuid_info_t *fuidp = NULL;
*xvpp = NULL;
/*
* In FreeBSD, access checking for creating an EA is being done
* in zfs_setextattr(),
*/
#ifndef __FreeBSD__
if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr))
return (error);
#endif
tx = dmu_tx_create(zfsvfs->z_os);
dmu_tx_hold_bonus(tx, zp->z_id);
dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
if (IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))) {
if (zfsvfs->z_fuid_obj == 0) {
dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
FUID_SIZE_ESTIMATE(zfsvfs));
dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
} else {
dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
FUID_SIZE_ESTIMATE(zfsvfs));
}
}
error = dmu_tx_assign(tx, zfsvfs->z_assign);
if (error) {
if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT)
dmu_tx_wait(tx);
dmu_tx_abort(tx);
return (error);
}
zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, 0, NULL, &fuidp);
ASSERT(xzp->z_phys->zp_parent == zp->z_id);
dmu_buf_will_dirty(zp->z_dbuf, tx);
zp->z_phys->zp_xattr = xzp->z_id;
(void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp,
xzp, "", NULL, fuidp, vap);
if (fuidp)
zfs_fuid_info_free(fuidp);
dmu_tx_commit(tx);
*xvpp = ZTOV(xzp);
return (0);
}
void
zfs_fuid_txhold(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
{
if (zfsvfs->z_fuid_obj == 0) {
dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
FUID_SIZE_ESTIMATE(zfsvfs));
dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
} else {
dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
FUID_SIZE_ESTIMATE(zfsvfs));
}
}
/*
* Replay a TX_WRITE ZIL transaction that didn't get committed
* after a system failure
*/
static int
zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
{
objset_t *os = zv->zv_objset;
char *data = (char *)(lr + 1); /* data follows lr_write_t */
uint64_t off = lr->lr_offset;
uint64_t len = lr->lr_length;
dmu_tx_t *tx;
int error;
if (byteswap)
byteswap_uint64_array(lr, sizeof (*lr));
tx = dmu_tx_create(os);
dmu_tx_hold_write(tx, ZVOL_OBJ, off, len);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
} else {
dmu_write(os, ZVOL_OBJ, off, len, data, tx);
dmu_tx_commit(tx);
}
return (SET_ERROR(error));
}
static uint64_t
zpios_dmu_object_create(run_args_t *run_args, objset_t *os)
{
struct dmu_tx *tx;
uint64_t obj = 0ULL;
int rc;
tx = dmu_tx_create(os);
dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, OBJ_SIZE);
rc = dmu_tx_assign(tx, TXG_WAIT);
if (rc) {
zpios_print(run_args->file,
"dmu_tx_assign() failed: %d\n", rc);
dmu_tx_abort(tx);
return obj;
}
obj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0,
DMU_OT_NONE, 0, tx);
rc = dmu_object_set_blocksize(os, obj, 128ULL << 10, 0, tx);
if (rc) {
zpios_print(run_args->file,
"dmu_object_set_blocksize() failed: %d\n", rc);
dmu_tx_abort(tx);
return obj;
}
dmu_tx_commit(tx);
return obj;
}
int
zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
{
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
znode_t *xzp;
dmu_tx_t *tx;
int error;
zfs_fuid_info_t *fuidp = NULL;
*xvpp = NULL;
if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr))
return (error);
tx = dmu_tx_create(zfsvfs->z_os);
dmu_tx_hold_bonus(tx, zp->z_id);
dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
if (zfsvfs->z_fuid_obj == 0) {
dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
} else {
dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, SPA_MAXBLOCKSIZE);
}
error = dmu_tx_assign(tx, zfsvfs->z_assign);
if (error) {
if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT)
dmu_tx_wait(tx);
dmu_tx_abort(tx);
return (error);
}
zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, 0, NULL, &fuidp);
ASSERT(xzp->z_phys->zp_parent == zp->z_id);
dmu_buf_will_dirty(zp->z_dbuf, tx);
zp->z_phys->zp_xattr = xzp->z_id;
(void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp,
xzp, "", NULL, fuidp, vap);
if (fuidp)
zfs_fuid_info_free(fuidp);
dmu_tx_commit(tx);
*xvpp = ZTOV(xzp);
return (0);
}
static int
zvol_write(struct bio *bio)
{
zvol_state_t *zv = bio->bi_bdev->bd_disk->private_data;
uint64_t offset = BIO_BI_SECTOR(bio) << 9;
uint64_t size = BIO_BI_SIZE(bio);
int error = 0;
dmu_tx_t *tx;
rl_t *rl;
uio_t uio;
if (bio->bi_rw & VDEV_REQ_FLUSH)
zil_commit(zv->zv_zilog, ZVOL_OBJ);
/*
* Some requests are just for flush and nothing else.
*/
if (size == 0)
goto out;
uio.uio_bvec = &bio->bi_io_vec[BIO_BI_IDX(bio)];
uio.uio_skip = BIO_BI_SKIP(bio);
uio.uio_resid = size;
uio.uio_iovcnt = bio->bi_vcnt - BIO_BI_IDX(bio);
uio.uio_loffset = offset;
uio.uio_limit = MAXOFFSET_T;
uio.uio_segflg = UIO_BVEC;
rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_WRITER);
tx = dmu_tx_create(zv->zv_objset);
dmu_tx_hold_write(tx, ZVOL_OBJ, offset, size);
/* This will only fail for ENOSPC */
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
zfs_range_unlock(rl);
goto out;
}
error = dmu_write_uio(zv->zv_objset, ZVOL_OBJ, &uio, size, tx);
if (error == 0)
zvol_log_write(zv, tx, offset, size,
!!(bio->bi_rw & VDEV_REQ_FUA));
dmu_tx_commit(tx);
zfs_range_unlock(rl);
if ((bio->bi_rw & VDEV_REQ_FUA) ||
zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
zil_commit(zv->zv_zilog, ZVOL_OBJ);
out:
return (error);
}
/*
* Common write path running under the zvol taskq context. This function
* is responsible for copying the request structure data in to the DMU and
* signaling the request queue with the result of the copy.
*/
static void
zvol_write(void *arg)
{
struct request *req = (struct request *)arg;
struct request_queue *q = req->q;
zvol_state_t *zv = q->queuedata;
fstrans_cookie_t cookie = spl_fstrans_mark();
uint64_t offset = blk_rq_pos(req) << 9;
uint64_t size = blk_rq_bytes(req);
int error = 0;
dmu_tx_t *tx;
rl_t *rl;
if (req->cmd_flags & VDEV_REQ_FLUSH)
zil_commit(zv->zv_zilog, ZVOL_OBJ);
/*
* Some requests are just for flush and nothing else.
*/
if (size == 0) {
error = 0;
goto out;
}
rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_WRITER);
tx = dmu_tx_create(zv->zv_objset);
dmu_tx_hold_write(tx, ZVOL_OBJ, offset, size);
/* This will only fail for ENOSPC */
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
zfs_range_unlock(rl);
goto out;
}
error = dmu_write_req(zv->zv_objset, ZVOL_OBJ, req, tx);
if (error == 0)
zvol_log_write(zv, tx, offset, size,
req->cmd_flags & VDEV_REQ_FUA);
dmu_tx_commit(tx);
zfs_range_unlock(rl);
if ((req->cmd_flags & VDEV_REQ_FUA) ||
zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
zil_commit(zv->zv_zilog, ZVOL_OBJ);
out:
blk_end_request(req, -error, size);
spl_fstrans_unmark(cookie);
}
static ssize_t osd_declare_write(const struct lu_env *env, struct dt_object *dt,
const struct lu_buf *buf, loff_t pos,
struct thandle *th)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
struct osd_thandle *oh;
uint64_t oid;
ENTRY;
oh = container_of0(th, struct osd_thandle, ot_super);
/* in some cases declare can race with creation (e.g. llog)
* and we need to wait till object is initialized. notice
* LOHA_EXISTs is supposed to be the last step in the
* initialization */
/* declare possible size change. notice we can't check
* current size here as another thread can change it */
if (dt_object_exists(dt)) {
LASSERT(obj->oo_db);
oid = obj->oo_db->db_object;
dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
} else {
oid = DMU_NEW_OBJECT;
dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
}
/* XXX: we still miss for append declaration support in ZFS
* -1 means append which is used by llog mostly, llog
* can grow upto LLOG_MIN_CHUNK_SIZE*8 records */
if (pos == -1)
pos = max_t(loff_t, 256 * 8 * LLOG_MIN_CHUNK_SIZE,
obj->oo_attr.la_size + (2 << 20));
dmu_tx_hold_write(oh->ot_tx, oid, pos, buf->lb_len);
/* dt_declare_write() is usually called for system objects, such
* as llog or last_rcvd files. We needn't enforce quota on those
* objects, so always set the lqi_space as 0. */
RETURN(osd_declare_quota(env, osd, obj->oo_attr.la_uid,
obj->oo_attr.la_gid, 0, oh, true, NULL,
false));
}
static int
zvol_write(zvol_state_t *zv, uio_t *uio, boolean_t sync)
{
uint64_t volsize = zv->zv_volsize;
rl_t *rl;
int error = 0;
ASSERT(zv && zv->zv_open_count > 0);
rl = zfs_range_lock(&zv->zv_range_lock, uio->uio_loffset,
uio->uio_resid, RL_WRITER);
while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
uint64_t off = uio->uio_loffset;
dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
if (bytes > volsize - off) /* don't write past the end */
bytes = volsize - off;
dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
/* This will only fail for ENOSPC */
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
break;
}
error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
if (error == 0)
zvol_log_write(zv, tx, off, bytes, sync);
dmu_tx_commit(tx);
if (error)
break;
}
zfs_range_unlock(rl);
if (sync)
zil_commit(zv->zv_zilog, ZVOL_OBJ);
return (error);
}
static uint64_t
zpios_dmu_object_create(run_args_t *run_args, objset_t *os)
{
struct dmu_tx *tx;
uint64_t obj = 0ULL;
uint64_t blksize = run_args->block_size;
int rc;
if (blksize < SPA_MINBLOCKSIZE ||
blksize > spa_maxblocksize(dmu_objset_spa(os)) ||
!ISP2(blksize)) {
zpios_print(run_args->file,
"invalid block size for pool: %d\n", (int)blksize);
return (obj);
}
tx = dmu_tx_create(os);
dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, OBJ_SIZE);
rc = dmu_tx_assign(tx, TXG_WAIT);
if (rc) {
zpios_print(run_args->file,
"dmu_tx_assign() failed: %d\n", rc);
dmu_tx_abort(tx);
return (obj);
}
obj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, DMU_OT_NONE, 0, tx);
rc = dmu_object_set_blocksize(os, obj, blksize, 0, tx);
if (rc) {
zpios_print(run_args->file,
"dmu_object_set_blocksize to %d failed: %d\n",
(int)blksize, rc);
dmu_tx_abort(tx);
return (obj);
}
dmu_tx_commit(tx);
return (obj);
}
static ssize_t osd_declare_write(const struct lu_env *env, struct dt_object *dt,
const loff_t size, loff_t pos,
struct thandle *th)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
struct osd_thandle *oh;
uint64_t oid;
ENTRY;
oh = container_of0(th, struct osd_thandle, ot_super);
/* in some cases declare can race with creation (e.g. llog)
* and we need to wait till object is initialized. notice
* LOHA_EXISTs is supposed to be the last step in the
* initialization */
/* declare possible size change. notice we can't check
* current size here as another thread can change it */
if (dt_object_exists(dt)) {
LASSERT(obj->oo_db);
oid = obj->oo_db->db_object;
dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
} else {
oid = DMU_NEW_OBJECT;
dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
}
dmu_tx_hold_write(oh->ot_tx, oid, pos, size);
/* dt_declare_write() is usually called for system objects, such
* as llog or last_rcvd files. We needn't enforce quota on those
* objects, so always set the lqi_space as 0. */
RETURN(osd_declare_quota(env, osd, obj->oo_attr.la_uid,
obj->oo_attr.la_gid, 0, oh, true, NULL,
false));
}
static int
zpios_dmu_write(run_args_t *run_args, objset_t *os, uint64_t object,
uint64_t offset, uint64_t size, const void *buf)
{
struct dmu_tx *tx;
int rc, how = TXG_WAIT;
// int flags = 0;
if (run_args->flags & DMU_WRITE_NOWAIT)
how = TXG_NOWAIT;
while (1) {
tx = dmu_tx_create(os);
dmu_tx_hold_write(tx, object, offset, size);
rc = dmu_tx_assign(tx, how);
if (rc) {
if (rc == ERESTART && how == TXG_NOWAIT) {
dmu_tx_wait(tx);
dmu_tx_abort(tx);
continue;
}
zpios_print(run_args->file,
"Error in dmu_tx_assign(), %d", rc);
dmu_tx_abort(tx);
return (rc);
}
break;
}
// if (run_args->flags & DMU_WRITE_ZC)
// flags |= DMU_WRITE_ZEROCOPY;
dmu_write(os, object, offset, size, buf, tx);
dmu_tx_commit(tx);
return (0);
}
/*
* Common write path running under the zvol taskq context. This function
* is responsible for copying the request structure data in to the DMU and
* signaling the request queue with the result of the copy.
*/
static void
zvol_write(void *arg)
{
struct request *req = (struct request *)arg;
struct request_queue *q = req->q;
zvol_state_t *zv = q->queuedata;
uint64_t offset = blk_rq_pos(req) << 9;
uint64_t size = blk_rq_bytes(req);
int error = 0;
dmu_tx_t *tx;
rl_t *rl;
rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_WRITER);
tx = dmu_tx_create(zv->zv_objset);
dmu_tx_hold_write(tx, ZVOL_OBJ, offset, size);
/* This will only fail for ENOSPC */
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
zfs_range_unlock(rl);
blk_end_request(req, -error, size);
return;
}
error = dmu_write_req(zv->zv_objset, ZVOL_OBJ, req, tx);
if (error == 0)
zvol_log_write(zv, tx, offset, size, rq_is_sync(req));
dmu_tx_commit(tx);
zfs_range_unlock(rl);
if (rq_is_sync(req))
zil_commit(zv->zv_zilog, ZVOL_OBJ);
blk_end_request(req, -error, size);
}
void __osd_xattr_declare_set(const struct lu_env *env, struct osd_object *obj,
int vallen, const char *name,
struct osd_thandle *oh)
{
struct osd_device *osd = osd_obj2dev(obj);
dmu_buf_t *db = obj->oo_db;
dmu_tx_t *tx = oh->ot_tx;
uint64_t xa_data_obj;
int rc = 0;
int here;
if (unlikely(obj->oo_destroyed))
return;
here = dt_object_exists(&obj->oo_dt);
/* object may be not yet created */
if (here) {
LASSERT(db);
LASSERT(obj->oo_sa_hdl);
/* we might just update SA_ZPL_DXATTR */
dmu_tx_hold_sa(tx, obj->oo_sa_hdl, 1);
if (obj->oo_xattr == ZFS_NO_OBJECT)
rc = -ENOENT;
}
if (!here || rc == -ENOENT) {
/* we'll be updating SA_ZPL_XATTR */
if (here) {
LASSERT(obj->oo_sa_hdl);
dmu_tx_hold_sa(tx, obj->oo_sa_hdl, 1);
}
/* xattr zap + entry */
dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, (char *) name);
/* xattr value obj */
dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, vallen);
return;
}
rc = -zap_lookup(osd->od_os, obj->oo_xattr, name, sizeof(uint64_t), 1,
&xa_data_obj);
if (rc == 0) {
/*
* Entry already exists.
* We'll truncate the existing object.
*/
dmu_tx_hold_bonus(tx, xa_data_obj);
dmu_tx_hold_free(tx, xa_data_obj, vallen, DMU_OBJECT_END);
dmu_tx_hold_write(tx, xa_data_obj, 0, vallen);
return;
} else if (rc == -ENOENT) {
/*
* Entry doesn't exist, we need to create a new one and a new
* object to store the value.
*/
dmu_tx_hold_bonus(tx, obj->oo_xattr);
dmu_tx_hold_zap(tx, obj->oo_xattr, TRUE, (char *) name);
dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, vallen);
return;
}
/* An error happened */
tx->tx_err = -rc;
}
/*
* Common write path running under the zvol taskq context. This function
* is responsible for copying the request structure data in to the DMU and
* signaling the request queue with the result of the copy.
*/
static void
zvol_write(void *arg)
{
struct request *req = (struct request *)arg;
struct request_queue *q = req->q;
zvol_state_t *zv = q->queuedata;
uint64_t offset = blk_rq_pos(req) << 9;
uint64_t size = blk_rq_bytes(req);
int error = 0;
dmu_tx_t *tx;
rl_t *rl;
/*
* Annotate this call path with a flag that indicates that it is
* unsafe to use KM_SLEEP during memory allocations due to the
* potential for a deadlock. KM_PUSHPAGE should be used instead.
*/
ASSERT(!(current->flags & PF_NOFS));
current->flags |= PF_NOFS;
if (req->cmd_flags & VDEV_REQ_FLUSH)
zil_commit(zv->zv_zilog, ZVOL_OBJ);
/*
* Some requests are just for flush and nothing else.
*/
if (size == 0) {
blk_end_request(req, 0, size);
goto out;
}
rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_WRITER);
tx = dmu_tx_create(zv->zv_objset);
dmu_tx_hold_write(tx, ZVOL_OBJ, offset, size);
/* This will only fail for ENOSPC */
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
zfs_range_unlock(rl);
blk_end_request(req, -error, size);
goto out;
}
error = dmu_write_req(zv->zv_objset, ZVOL_OBJ, req, tx);
if (error == 0)
zvol_log_write(zv, tx, offset, size,
req->cmd_flags & VDEV_REQ_FUA);
dmu_tx_commit(tx);
zfs_range_unlock(rl);
if ((req->cmd_flags & VDEV_REQ_FUA) ||
zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
zil_commit(zv->zv_zilog, ZVOL_OBJ);
blk_end_request(req, -error, size);
out:
current->flags &= ~PF_NOFS;
}
/*
* Lookup/Create an extended attribute entry.
*
* Input arguments:
* dzp - znode for hidden attribute directory
* name - name of attribute
* flag - ZNEW: if the entry already exists, fail with EEXIST.
* ZEXISTS: if the entry does not exist, fail with ENOENT.
*
* Output arguments:
* vpp - pointer to the vnode for the entry (NULL if there isn't one)
*
* Return value: 0 on success or errno value on failure.
*/
int
zfs_obtain_xattr(znode_t *dzp, const char *name, mode_t mode, cred_t *cr,
vnode_t **vpp, int flag)
{
znode_t *xzp = NULL;
zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
zilog_t *zilog;
zfs_dirlock_t *dl;
dmu_tx_t *tx;
struct vnode_attr vattr;
int error;
struct componentname cn;
zfs_acl_ids_t acl_ids;
/* zfs_dirent_lock() expects a component name */
bzero(&cn, sizeof (cn));
cn.cn_nameiop = LOOKUP;
cn.cn_flags = ISLASTCN;
cn.cn_nameptr = (char *)name;
cn.cn_namelen = strlen(name);
ZFS_ENTER(zfsvfs);
ZFS_VERIFY_ZP(dzp);
zilog = zfsvfs->z_log;
VATTR_INIT(&vattr);
VATTR_SET(&vattr, va_type, VREG);
VATTR_SET(&vattr, va_mode, mode & ~S_IFMT);
if ((error = zfs_acl_ids_create(dzp, 0,
&vattr, cr, NULL, &acl_ids)) != 0) {
ZFS_EXIT(zfsvfs);
return (error);
}
top:
/* Lock the attribute entry name. */
if ( (error = zfs_dirent_lock(&dl, dzp, (char *)name, &xzp, flag,
NULL, &cn)) ) {
goto out;
}
/* If the name already exists, we're done. */
if (xzp != NULL) {
zfs_dirent_unlock(dl);
goto out;
}
tx = dmu_tx_create(zfsvfs->z_os);
dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
//dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
//dmu_tx_hold_bonus(tx, dzp->z_id);
dmu_tx_hold_zap(tx, dzp->z_id, TRUE, (char *)name);
#if 1 // FIXME
if (dzp->z_pflags & ZFS_INHERIT_ACE) {
dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
}
#endif
zfs_sa_upgrade_txholds(tx, dzp);
error = dmu_tx_assign(tx, TXG_NOWAIT);
if (error) {
zfs_dirent_unlock(dl);
if (error == ERESTART) {
dmu_tx_wait(tx);
dmu_tx_abort(tx);
goto top;
}
dmu_tx_abort(tx);
goto out;
}
zfs_mknode(dzp, &vattr, tx, cr, 0, &xzp, &acl_ids);
/*
ASSERT(xzp->z_id == zoid);
*/
(void) zfs_link_create(dl, xzp, tx, ZNEW);
zfs_log_create(zilog, tx, TX_CREATE, dzp, xzp, (char *)name,
NULL /* vsecp */, 0 /*acl_ids.z_fuidp*/, &vattr);
zfs_acl_ids_free(&acl_ids);
dmu_tx_commit(tx);
zfs_znode_wait_vnode(xzp);
zfs_dirent_unlock(dl);
out:
if (error == EEXIST)
error = ENOATTR;
if (xzp)
*vpp = ZTOV(xzp);
ZFS_EXIT(zfsvfs);
return (error);
}
/*
* Increase the file length
*
* IN: zp - znode of file to free data in.
* end - new end-of-file
*
* RETURN: 0 on success, error code on failure
*/
static int
zfs_extend(znode_t *zp, uint64_t end)
{
zfs_sb_t *zsb = ZTOZSB(zp);
dmu_tx_t *tx;
rl_t *rl;
uint64_t newblksz;
int error;
/*
* We will change zp_size, lock the whole file.
*/
rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
/*
* Nothing to do if file already at desired length.
*/
if (end <= zp->z_size) {
zfs_range_unlock(rl);
return (0);
}
tx = dmu_tx_create(zsb->z_os);
dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
zfs_sa_upgrade_txholds(tx, zp);
if (end > zp->z_blksz &&
(!ISP2(zp->z_blksz) || zp->z_blksz < zsb->z_max_blksz)) {
/*
* We are growing the file past the current block size.
*/
if (zp->z_blksz > ZTOZSB(zp)->z_max_blksz) {
ASSERT(!ISP2(zp->z_blksz));
newblksz = MIN(end, SPA_MAXBLOCKSIZE);
} else {
newblksz = MIN(end, ZTOZSB(zp)->z_max_blksz);
}
dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
} else {
newblksz = 0;
}
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
zfs_range_unlock(rl);
return (error);
}
if (newblksz)
zfs_grow_blocksize(zp, newblksz, tx);
zp->z_size = end;
VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(ZTOZSB(zp)),
&zp->z_size, sizeof (zp->z_size), tx));
zfs_range_unlock(rl);
dmu_tx_commit(tx);
return (0);
}
int
zvol_strategy(buf_t *bp)
{
zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(bp->b_edev));
uint64_t off, volsize;
size_t size, resid;
char *addr;
objset_t *os;
int error = 0;
int sync;
int reading;
int txg_sync_needed = B_FALSE;
if (zv == NULL) {
bioerror(bp, ENXIO);
biodone(bp);
return (0);
}
if (getminor(bp->b_edev) == 0) {
bioerror(bp, EINVAL);
biodone(bp);
return (0);
}
if (zv->zv_readonly && !(bp->b_flags & B_READ)) {
bioerror(bp, EROFS);
biodone(bp);
return (0);
}
off = ldbtob(bp->b_blkno);
volsize = zv->zv_volsize;
os = zv->zv_objset;
ASSERT(os != NULL);
sync = !(bp->b_flags & B_ASYNC) && !(zil_disable);
bp_mapin(bp);
addr = bp->b_un.b_addr;
resid = bp->b_bcount;
/*
* There must be no buffer changes when doing a dmu_sync() because
* we can't change the data whilst calculating the checksum.
* A better approach than a per zvol rwlock would be to lock ranges.
*/
reading = bp->b_flags & B_READ;
if (reading || resid <= zvol_immediate_write_sz)
rw_enter(&zv->zv_dslock, RW_READER);
else
rw_enter(&zv->zv_dslock, RW_WRITER);
while (resid != 0 && off < volsize) {
size = MIN(resid, 1UL << 20); /* cap at 1MB per tx */
if (size > volsize - off) /* don't write past the end */
size = volsize - off;
if (reading) {
error = dmu_read(os, ZVOL_OBJ, off, size, addr);
} else {
dmu_tx_t *tx = dmu_tx_create(os);
dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
} else {
dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
if (sync) {
/* use the ZIL to commit this write */
if (zvol_log_write(zv, tx, off, size,
addr) != 0) {
txg_sync_needed = B_TRUE;
}
}
dmu_tx_commit(tx);
}
}
if (error)
break;
off += size;
addr += size;
resid -= size;
}
rw_exit(&zv->zv_dslock);
if ((bp->b_resid = resid) == bp->b_bcount)
bioerror(bp, off > volsize ? EINVAL : error);
biodone(bp);
if (sync) {
if (txg_sync_needed)
txg_wait_synced(dmu_objset_pool(os), 0);
else
zil_commit(zv->zv_zilog, UINT64_MAX, 0);
}
return (0);
}
