dmu.c revision 3baa08fc5b6bea08a475b0cfe3ad161d74c5864b
4015N/A/*
4015N/A * CDDL HEADER START
4015N/A *
4015N/A * The contents of this file are subject to the terms of the
4015N/A * Common Development and Distribution License (the "License").
4015N/A * You may not use this file except in compliance with the License.
4015N/A *
4015N/A * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
4015N/A * or http://www.opensolaris.org/os/licensing.
4015N/A * See the License for the specific language governing permissions
4015N/A * and limitations under the License.
4015N/A *
4015N/A * When distributing Covered Code, include this CDDL HEADER in each
4015N/A * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
4015N/A * If applicable, add the following below this CDDL HEADER, with the
4015N/A * fields enclosed by brackets "[]" replaced with your own identifying
4015N/A * information: Portions Copyright [yyyy] [name of copyright owner]
4015N/A *
4015N/A * CDDL HEADER END
4015N/A */
4015N/A/*
4015N/A * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
4015N/A * Use is subject to license terms.
4015N/A */
4015N/A
4015N/A#pragma ident "%Z%%M% %I% %E% SMI"
4015N/A
4015N/A#include <sys/dmu.h>
4015N/A#include <sys/dmu_impl.h>
4015N/A#include <sys/dmu_tx.h>
4015N/A#include <sys/dbuf.h>
4015N/A#include <sys/dnode.h>
4015N/A#include <sys/zfs_context.h>
4015N/A#include <sys/dmu_objset.h>
4015N/A#include <sys/dmu_traverse.h>
4015N/A#include <sys/dsl_dataset.h>
4015N/A#include <sys/dsl_dir.h>
4015N/A#include <sys/dsl_pool.h>
4015N/A#include <sys/dsl_synctask.h>
4015N/A#include <sys/dsl_prop.h>
4015N/A#include <sys/dmu_zfetch.h>
4015N/A#include <sys/zfs_ioctl.h>
4015N/A#include <sys/zap.h>
4015N/A#include <sys/zio_checksum.h>
4015N/A#ifdef _KERNEL
4015N/A#include <sys/vmsystm.h>
4015N/A#endif
4015N/A
4015N/Aconst dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
4015N/A { byteswap_uint8_array, TRUE, "unallocated" },
4015N/A { zap_byteswap, TRUE, "object directory" },
4015N/A { byteswap_uint64_array, TRUE, "object array" },
4015N/A { byteswap_uint8_array, TRUE, "packed nvlist" },
4015N/A { byteswap_uint64_array, TRUE, "packed nvlist size" },
4015N/A { byteswap_uint64_array, TRUE, "bplist" },
4015N/A { byteswap_uint64_array, TRUE, "bplist header" },
4015N/A { byteswap_uint64_array, TRUE, "SPA space map header" },
4015N/A { byteswap_uint64_array, TRUE, "SPA space map" },
4015N/A { byteswap_uint64_array, TRUE, "ZIL intent log" },
4015N/A { dnode_buf_byteswap, TRUE, "DMU dnode" },
4015N/A { dmu_objset_byteswap, TRUE, "DMU objset" },
4015N/A { byteswap_uint64_array, TRUE, "DSL directory" },
4015N/A { zap_byteswap, TRUE, "DSL directory child map"},
4015N/A { zap_byteswap, TRUE, "DSL dataset snap map" },
4015N/A { zap_byteswap, TRUE, "DSL props" },
4015N/A { byteswap_uint64_array, TRUE, "DSL dataset" },
4015N/A { zfs_znode_byteswap, TRUE, "ZFS znode" },
4015N/A { zfs_oldacl_byteswap, TRUE, "ZFS V0 ACL" },
4015N/A { byteswap_uint8_array, FALSE, "ZFS plain file" },
4015N/A { zap_byteswap, TRUE, "ZFS directory" },
4015N/A { zap_byteswap, TRUE, "ZFS master node" },
4015N/A { zap_byteswap, TRUE, "ZFS delete queue" },
4015N/A { byteswap_uint8_array, FALSE, "zvol object" },
4015N/A { zap_byteswap, TRUE, "zvol prop" },
4015N/A { byteswap_uint8_array, FALSE, "other uint8[]" },
4015N/A { byteswap_uint64_array, FALSE, "other uint64[]" },
4015N/A { zap_byteswap, TRUE, "other ZAP" },
4015N/A { zap_byteswap, TRUE, "persistent error log" },
4015N/A { byteswap_uint8_array, TRUE, "SPA history" },
4015N/A { byteswap_uint64_array, TRUE, "SPA history offsets" },
4015N/A { zap_byteswap, TRUE, "Pool properties" },
4015N/A { zap_byteswap, TRUE, "DSL permissions" },
4030N/A { zfs_acl_byteswap, TRUE, "ZFS ACL" },
4030N/A { byteswap_uint8_array, TRUE, "ZFS SYSACL" },
4015N/A { byteswap_uint8_array, TRUE, "FUID table" },
4030N/A { byteswap_uint64_array, TRUE, "FUID table size" },
4030N/A { zap_byteswap, TRUE, "DSL dataset next clones"},
4015N/A { zap_byteswap, TRUE, "scrub work queue" },
4015N/A};
4015N/A
4015N/Aint
4015N/Admu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
4015N/A void *tag, dmu_buf_t **dbp)
4015N/A{
4015N/A dnode_t *dn;
4015N/A uint64_t blkid;
4015N/A dmu_buf_impl_t *db;
4015N/A int err;
4015N/A
4015N/A err = dnode_hold(os->os, object, FTAG, &dn);
4015N/A if (err)
4015N/A return (err);
4015N/A blkid = dbuf_whichblock(dn, offset);
4015N/A rw_enter(&dn->dn_struct_rwlock, RW_READER);
4015N/A db = dbuf_hold(dn, blkid, tag);
4015N/A rw_exit(&dn->dn_struct_rwlock);
4015N/A if (db == NULL) {
4015N/A err = EIO;
4015N/A } else {
4015N/A err = dbuf_read(db, NULL, DB_RF_CANFAIL);
4015N/A if (err) {
4015N/A dbuf_rele(db, tag);
4015N/A db = NULL;
4015N/A }
4015N/A }
4015N/A
4015N/A dnode_rele(dn, FTAG);
4015N/A *dbp = &db->db;
4015N/A return (err);
4015N/A}
4015N/A
4015N/Aint
4015N/Admu_bonus_max(void)
4015N/A{
4015N/A return (DN_MAX_BONUSLEN);
4015N/A}
4015N/A
4015N/Aint
4015N/Admu_set_bonus(dmu_buf_t *db, int newsize, dmu_tx_t *tx)
4015N/A{
4015N/A dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode;
4015N/A
4015N/A if (dn->dn_bonus != (dmu_buf_impl_t *)db)
4015N/A return (EINVAL);
4015N/A if (newsize < 0 || newsize > db->db_size)
4015N/A return (EINVAL);
4015N/A dnode_setbonuslen(dn, newsize, tx);
4015N/A return (0);
4015N/A}
4015N/A
4015N/A/*
4015N/A * returns ENOENT, EIO, or 0.
4015N/A */
4015N/Aint
4015N/Admu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
4015N/A{
4015N/A dnode_t *dn;
4015N/A dmu_buf_impl_t *db;
4015N/A int error;
4015N/A
4015N/A error = dnode_hold(os->os, object, FTAG, &dn);
4015N/A if (error)
4015N/A return (error);
4015N/A
4015N/A rw_enter(&dn->dn_struct_rwlock, RW_READER);
4015N/A if (dn->dn_bonus == NULL) {
4015N/A rw_exit(&dn->dn_struct_rwlock);
4015N/A rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
4015N/A if (dn->dn_bonus == NULL)
4015N/A dbuf_create_bonus(dn);
4015N/A }
4015N/A db = dn->dn_bonus;
4015N/A rw_exit(&dn->dn_struct_rwlock);
4015N/A
4015N/A /* as long as the bonus buf is held, the dnode will be held */
4015N/A if (refcount_add(&db->db_holds, tag) == 1)
4015N/A VERIFY(dnode_add_ref(dn, db));
4015N/A
4015N/A dnode_rele(dn, FTAG);
4015N/A
4015N/A VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED));
4015N/A
4015N/A *dbp = &db->db;
4015N/A return (0);
4015N/A}
4015N/A
4015N/A/*
4015N/A * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
4015N/A * to take a held dnode rather than <os, object> -- the lookup is wasteful,
4015N/A * and can induce severe lock contention when writing to several files
4015N/A * whose dnodes are in the same block.
4015N/A */
4015N/Astatic int
4015N/Admu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset,
4015N/A uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
4015N/A{
4015N/A dmu_buf_t **dbp;
4015N/A uint64_t blkid, nblks, i;
4015N/A uint32_t flags;
4015N/A int err;
4015N/A zio_t *zio;
4015N/A
4015N/A ASSERT(length <= DMU_MAX_ACCESS);
4015N/A
4015N/A flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT;
4015N/A if (length > zfetch_array_rd_sz)
4015N/A flags |= DB_RF_NOPREFETCH;
4015N/A
4015N/A rw_enter(&dn->dn_struct_rwlock, RW_READER);
4030N/A if (dn->dn_datablkshift) {
4015N/A int blkshift = dn->dn_datablkshift;
4015N/A nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) -
4015N/A P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift;
4015N/A } else {
4015N/A if (offset + length > dn->dn_datablksz) {
4015N/A zfs_panic_recover("zfs: accessing past end of object "
4015N/A "%llx/%llx (size=%u access=%llu+%llu)",
4015N/A (longlong_t)dn->dn_objset->
4015N/A os_dsl_dataset->ds_object,
4015N/A (longlong_t)dn->dn_object, dn->dn_datablksz,
4015N/A (longlong_t)offset, (longlong_t)length);
4015N/A return (EIO);
4015N/A }
4015N/A nblks = 1;
4015N/A }
4015N/A dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
4015N/A
4015N/A zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, TRUE);
4015N/A blkid = dbuf_whichblock(dn, offset);
4015N/A for (i = 0; i < nblks; i++) {
4015N/A dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag);
4015N/A if (db == NULL) {
4015N/A rw_exit(&dn->dn_struct_rwlock);
4015N/A dmu_buf_rele_array(dbp, nblks, tag);
4015N/A zio_nowait(zio);
4015N/A return (EIO);
4015N/A }
4015N/A /* initiate async i/o */
4015N/A if (read) {
4015N/A rw_exit(&dn->dn_struct_rwlock);
4015N/A (void) dbuf_read(db, zio, flags);
4015N/A rw_enter(&dn->dn_struct_rwlock, RW_READER);
4015N/A }
4015N/A dbp[i] = &db->db;
4015N/A }
4015N/A rw_exit(&dn->dn_struct_rwlock);
4015N/A
4015N/A /* wait for async i/o */
4015N/A err = zio_wait(zio);
4015N/A if (err) {
4015N/A dmu_buf_rele_array(dbp, nblks, tag);
4015N/A return (err);
4015N/A }
4015N/A
4015N/A /* wait for other io to complete */
4015N/A if (read) {
4015N/A for (i = 0; i < nblks; i++) {
4015N/A dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
4015N/A mutex_enter(&db->db_mtx);
4015N/A while (db->db_state == DB_READ ||
4015N/A db->db_state == DB_FILL)
4015N/A cv_wait(&db->db_changed, &db->db_mtx);
4015N/A if (db->db_state == DB_UNCACHED)
4015N/A err = EIO;
4015N/A mutex_exit(&db->db_mtx);
4015N/A if (err) {
4015N/A dmu_buf_rele_array(dbp, nblks, tag);
4015N/A return (err);
4015N/A }
4015N/A }
4015N/A }
4015N/A
4015N/A *numbufsp = nblks;
4015N/A *dbpp = dbp;
4015N/A return (0);
4015N/A}
4015N/A
4015N/Astatic int
4015N/Admu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
4015N/A uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
4015N/A{
4015N/A dnode_t *dn;
4015N/A int err;
4015N/A
4015N/A err = dnode_hold(os->os, object, FTAG, &dn);
4023N/A if (err)
4015N/A return (err);
4015N/A
4015N/A err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
4015N/A numbufsp, dbpp);
4015N/A
4015N/A dnode_rele(dn, FTAG);
4015N/A
4015N/A return (err);
4015N/A}
4015N/A
4015N/Aint
4015N/Admu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
4015N/A uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
4015N/A{
4015N/A dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode;
4015N/A int err;
4015N/A
4015N/A err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
4015N/A numbufsp, dbpp);
4015N/A
4015N/A return (err);
4015N/A}
4015N/A
4015N/Avoid
4015N/Admu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
4015N/A{
4015N/A int i;
4015N/A dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
4015N/A
4015N/A if (numbufs == 0)
4015N/A return;
4015N/A
4015N/A for (i = 0; i < numbufs; i++) {
4015N/A if (dbp[i])
4015N/A dbuf_rele(dbp[i], tag);
4015N/A }
4015N/A
4015N/A kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
4015N/A}
4015N/A
4015N/Avoid
4015N/Admu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
4015N/A{
4015N/A dnode_t *dn;
4015N/A uint64_t blkid;
4015N/A int nblks, i, err;
4015N/A
4015N/A if (zfs_prefetch_disable)
4015N/A return;
4015N/A
4015N/A if (len == 0) { /* they're interested in the bonus buffer */
4015N/A dn = os->os->os_meta_dnode;
4015N/A
4015N/A if (object == 0 || object >= DN_MAX_OBJECT)
4015N/A return;
4015N/A
4015N/A rw_enter(&dn->dn_struct_rwlock, RW_READER);
4015N/A blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t));
4015N/A dbuf_prefetch(dn, blkid);
4015N/A rw_exit(&dn->dn_struct_rwlock);
4015N/A return;
4015N/A }
4015N/A
4015N/A /*
4015N/A * XXX - Note, if the dnode for the requested object is not
4015N/A * already cached, we will do a *synchronous* read in the
4015N/A * dnode_hold() call. The same is true for any indirects.
4015N/A */
4015N/A err = dnode_hold(os->os, object, FTAG, &dn);
4015N/A if (err != 0)
4015N/A return;
4015N/A
4015N/A rw_enter(&dn->dn_struct_rwlock, RW_READER);
4015N/A if (dn->dn_datablkshift) {
4015N/A int blkshift = dn->dn_datablkshift;
4015N/A nblks = (P2ROUNDUP(offset+len, 1<<blkshift) -
4015N/A P2ALIGN(offset, 1<<blkshift)) >> blkshift;
4015N/A } else {
4015N/A nblks = (offset < dn->dn_datablksz);
4015N/A }
4015N/A
4015N/A if (nblks != 0) {
4015N/A blkid = dbuf_whichblock(dn, offset);
4015N/A for (i = 0; i < nblks; i++)
4015N/A dbuf_prefetch(dn, blkid+i);
4015N/A }
4015N/A
4015N/A rw_exit(&dn->dn_struct_rwlock);
4015N/A
dnode_rele(dn, FTAG);
}
static int
get_next_chunk(dnode_t *dn, uint64_t *offset, uint64_t limit)
{
uint64_t len = limit - *offset;
uint64_t chunk_len = dn->dn_datablksz * DMU_MAX_DELETEBLKCNT;
uint64_t dn_used;
int err;
ASSERT(limit <= *offset);
dn_used = dn->dn_phys->dn_used <<
(dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES ? 0 : DEV_BSHIFT);
if (len <= chunk_len || dn_used <= chunk_len) {
*offset = limit;
return (0);
}
while (*offset > limit) {
uint64_t initial_offset = *offset;
uint64_t delta;
/* skip over allocated data */
err = dnode_next_offset(dn,
DNODE_FIND_HOLE|DNODE_FIND_BACKWARDS, offset, 1, 1, 0);
if (err == ESRCH)
*offset = limit;
else if (err)
return (err);
ASSERT3U(*offset, <=, initial_offset);
delta = initial_offset - *offset;
if (delta >= chunk_len) {
*offset += delta - chunk_len;
return (0);
}
chunk_len -= delta;
/* skip over unallocated data */
err = dnode_next_offset(dn,
DNODE_FIND_BACKWARDS, offset, 1, 1, 0);
if (err == ESRCH)
*offset = limit;
else if (err)
return (err);
if (*offset < limit)
*offset = limit;
ASSERT3U(*offset, <, initial_offset);
}
return (0);
}
static int
dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
uint64_t length, boolean_t free_dnode)
{
dmu_tx_t *tx;
uint64_t object_size, start, end, len;
boolean_t trunc = (length == DMU_OBJECT_END);
int align, err;
align = 1 << dn->dn_datablkshift;
ASSERT(align > 0);
object_size = align == 1 ? dn->dn_datablksz :
(dn->dn_maxblkid + 1) << dn->dn_datablkshift;
if (trunc || (end = offset + length) > object_size)
end = object_size;
if (end <= offset)
return (0);
length = end - offset;
while (length) {
start = end;
err = get_next_chunk(dn, &start, offset);
if (err)
return (err);
len = trunc ? DMU_OBJECT_END : end - start;
tx = dmu_tx_create(os);
dmu_tx_hold_free(tx, dn->dn_object, start, len);
err = dmu_tx_assign(tx, TXG_WAIT);
if (err) {
dmu_tx_abort(tx);
return (err);
}
dnode_free_range(dn, start, trunc ? -1 : len, tx);
if (start == 0 && trunc && free_dnode)
dnode_free(dn, tx);
length -= end - start;
dmu_tx_commit(tx);
end = start;
trunc = FALSE;
}
return (0);
}
int
dmu_free_long_range(objset_t *os, uint64_t object,
uint64_t offset, uint64_t length)
{
dnode_t *dn;
int err;
err = dnode_hold(os->os, object, FTAG, &dn);
if (err != 0)
return (err);
err = dmu_free_long_range_impl(os, dn, offset, length, FALSE);
dnode_rele(dn, FTAG);
return (err);
}
int
dmu_free_object(objset_t *os, uint64_t object)
{
dnode_t *dn;
dmu_tx_t *tx;
int err;
err = dnode_hold_impl(os->os, object, DNODE_MUST_BE_ALLOCATED,
FTAG, &dn);
if (err != 0)
return (err);
if (dn->dn_nlevels == 1) {
tx = dmu_tx_create(os);
dmu_tx_hold_bonus(tx, object);
dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END);
err = dmu_tx_assign(tx, TXG_WAIT);
if (err == 0) {
dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
dnode_free(dn, tx);
dmu_tx_commit(tx);
} else {
dmu_tx_abort(tx);
}
} else {
err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE);
}
dnode_rele(dn, FTAG);
return (err);
}
int
dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
uint64_t size, dmu_tx_t *tx)
{
dnode_t *dn;
int err = dnode_hold(os->os, object, FTAG, &dn);
if (err)
return (err);
ASSERT(offset < UINT64_MAX);
ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
dnode_free_range(dn, offset, size, tx);
dnode_rele(dn, FTAG);
return (0);
}
int
dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
void *buf)
{
dnode_t *dn;
dmu_buf_t **dbp;
int numbufs, i, err;
err = dnode_hold(os->os, object, FTAG, &dn);
if (err)
return (err);
/*
* Deal with odd block sizes, where there can't be data past the first
* block. If we ever do the tail block optimization, we will need to
* handle that here as well.
*/
if (dn->dn_datablkshift == 0) {
int newsz = offset > dn->dn_datablksz ? 0 :
MIN(size, dn->dn_datablksz - offset);
bzero((char *)buf + newsz, size - newsz);
size = newsz;
}
while (size > 0) {
uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
/*
* NB: we could do this block-at-a-time, but it's nice
* to be reading in parallel.
*/
err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
TRUE, FTAG, &numbufs, &dbp);
if (err)
break;
for (i = 0; i < numbufs; i++) {
int tocpy;
int bufoff;
dmu_buf_t *db = dbp[i];
ASSERT(size > 0);
bufoff = offset - db->db_offset;
tocpy = (int)MIN(db->db_size - bufoff, size);
bcopy((char *)db->db_data + bufoff, buf, tocpy);
offset += tocpy;
size -= tocpy;
buf = (char *)buf + tocpy;
}
dmu_buf_rele_array(dbp, numbufs, FTAG);
}
dnode_rele(dn, FTAG);
return (err);
}
void
dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
const void *buf, dmu_tx_t *tx)
{
dmu_buf_t **dbp;
int numbufs, i;
if (size == 0)
return;
VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
FALSE, FTAG, &numbufs, &dbp));
for (i = 0; i < numbufs; i++) {
int tocpy;
int bufoff;
dmu_buf_t *db = dbp[i];
ASSERT(size > 0);
bufoff = offset - db->db_offset;
tocpy = (int)MIN(db->db_size - bufoff, size);
ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
if (tocpy == db->db_size)
dmu_buf_will_fill(db, tx);
else
dmu_buf_will_dirty(db, tx);
bcopy(buf, (char *)db->db_data + bufoff, tocpy);
if (tocpy == db->db_size)
dmu_buf_fill_done(db, tx);
offset += tocpy;
size -= tocpy;
buf = (char *)buf + tocpy;
}
dmu_buf_rele_array(dbp, numbufs, FTAG);
}
#ifdef _KERNEL
int
dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
{
dmu_buf_t **dbp;
int numbufs, i, err;
/*
* NB: we could do this block-at-a-time, but it's nice
* to be reading in parallel.
*/
err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG,
&numbufs, &dbp);
if (err)
return (err);
for (i = 0; i < numbufs; i++) {
int tocpy;
int bufoff;
dmu_buf_t *db = dbp[i];
ASSERT(size > 0);
bufoff = uio->uio_loffset - db->db_offset;
tocpy = (int)MIN(db->db_size - bufoff, size);
err = uiomove((char *)db->db_data + bufoff, tocpy,
UIO_READ, uio);
if (err)
break;
size -= tocpy;
}
dmu_buf_rele_array(dbp, numbufs, FTAG);
return (err);
}
int
dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
dmu_tx_t *tx)
{
dmu_buf_t **dbp;
int numbufs, i;
int err = 0;
if (size == 0)
return (0);
err = dmu_buf_hold_array(os, object, uio->uio_loffset, size,
FALSE, FTAG, &numbufs, &dbp);
if (err)
return (err);
for (i = 0; i < numbufs; i++) {
int tocpy;
int bufoff;
dmu_buf_t *db = dbp[i];
ASSERT(size > 0);
bufoff = uio->uio_loffset - db->db_offset;
tocpy = (int)MIN(db->db_size - bufoff, size);
ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
if (tocpy == db->db_size)
dmu_buf_will_fill(db, tx);
else
dmu_buf_will_dirty(db, tx);
/*
* XXX uiomove could block forever (eg. nfs-backed
* pages). There needs to be a uiolockdown() function
* to lock the pages in memory, so that uiomove won't
* block.
*/
err = uiomove((char *)db->db_data + bufoff, tocpy,
UIO_WRITE, uio);
if (tocpy == db->db_size)
dmu_buf_fill_done(db, tx);
if (err)
break;
size -= tocpy;
}
dmu_buf_rele_array(dbp, numbufs, FTAG);
return (err);
}
int
dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
page_t *pp, dmu_tx_t *tx)
{
dmu_buf_t **dbp;
int numbufs, i;
int err;
if (size == 0)
return (0);
err = dmu_buf_hold_array(os, object, offset, size,
FALSE, FTAG, &numbufs, &dbp);
if (err)
return (err);
for (i = 0; i < numbufs; i++) {
int tocpy, copied, thiscpy;
int bufoff;
dmu_buf_t *db = dbp[i];
caddr_t va;
ASSERT(size > 0);
ASSERT3U(db->db_size, >=, PAGESIZE);
bufoff = offset - db->db_offset;
tocpy = (int)MIN(db->db_size - bufoff, size);
ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
if (tocpy == db->db_size)
dmu_buf_will_fill(db, tx);
else
dmu_buf_will_dirty(db, tx);
for (copied = 0; copied < tocpy; copied += PAGESIZE) {
ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
thiscpy = MIN(PAGESIZE, tocpy - copied);
va = ppmapin(pp, PROT_READ, (caddr_t)-1);
bcopy(va, (char *)db->db_data + bufoff, thiscpy);
ppmapout(va);
pp = pp->p_next;
bufoff += PAGESIZE;
}
if (tocpy == db->db_size)
dmu_buf_fill_done(db, tx);
if (err)
break;
offset += tocpy;
size -= tocpy;
}
dmu_buf_rele_array(dbp, numbufs, FTAG);
return (err);
}
#endif
typedef struct {
dbuf_dirty_record_t *dr;
dmu_sync_cb_t *done;
void *arg;
} dmu_sync_arg_t;
/* ARGSUSED */
static void
dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
{
dmu_sync_arg_t *in = varg;
dbuf_dirty_record_t *dr = in->dr;
dmu_buf_impl_t *db = dr->dr_dbuf;
dmu_sync_cb_t *done = in->done;
if (!BP_IS_HOLE(zio->io_bp)) {
zio->io_bp->blk_fill = 1;
BP_SET_TYPE(zio->io_bp, db->db_dnode->dn_type);
BP_SET_LEVEL(zio->io_bp, 0);
}
mutex_enter(&db->db_mtx);
ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
dr->dt.dl.dr_overridden_by = *zio->io_bp; /* structure assignment */
dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
cv_broadcast(&db->db_changed);
mutex_exit(&db->db_mtx);
if (done)
done(&(db->db), in->arg);
kmem_free(in, sizeof (dmu_sync_arg_t));
}
/*
* Intent log support: sync the block associated with db to disk.
* N.B. and XXX: the caller is responsible for making sure that the
* data isn't changing while dmu_sync() is writing it.
*
* Return values:
*
* EEXIST: this txg has already been synced, so there's nothing to to.
* The caller should not log the write.
*
* ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
* The caller should not log the write.
*
* EALREADY: this block is already in the process of being synced.
* The caller should track its progress (somehow).
*
* EINPROGRESS: the IO has been initiated.
* The caller should log this blkptr in the callback.
*
* 0: completed. Sets *bp to the blkptr just written.
* The caller should log this blkptr immediately.
*/
int
dmu_sync(zio_t *pio, dmu_buf_t *db_fake,
blkptr_t *bp, uint64_t txg, dmu_sync_cb_t *done, void *arg)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
objset_impl_t *os = db->db_objset;
dsl_pool_t *dp = os->os_dsl_dataset->ds_dir->dd_pool;
tx_state_t *tx = &dp->dp_tx;
dbuf_dirty_record_t *dr;
dmu_sync_arg_t *in;
zbookmark_t zb;
writeprops_t wp = { 0 };
zio_t *zio;
int zio_flags;
int err;
ASSERT(BP_IS_HOLE(bp));
ASSERT(txg != 0);
dprintf("dmu_sync txg=%llu, s,o,q %llu %llu %llu\n",
txg, tx->tx_synced_txg, tx->tx_open_txg, tx->tx_quiesced_txg);
/*
* XXX - would be nice if we could do this without suspending...
*/
txg_suspend(dp);
/*
* If this txg already synced, there's nothing to do.
*/
if (txg <= tx->tx_synced_txg) {
txg_resume(dp);
/*
* If we're running ziltest, we need the blkptr regardless.
*/
if (txg > spa_freeze_txg(dp->dp_spa)) {
/* if db_blkptr == NULL, this was an empty write */
if (db->db_blkptr)
*bp = *db->db_blkptr; /* structure assignment */
return (0);
}
return (EEXIST);
}
mutex_enter(&db->db_mtx);
if (txg == tx->tx_syncing_txg) {
while (db->db_data_pending) {
/*
* IO is in-progress. Wait for it to finish.
* XXX - would be nice to be able to somehow "attach"
* this zio to the parent zio passed in.
*/
cv_wait(&db->db_changed, &db->db_mtx);
if (!db->db_data_pending &&
db->db_blkptr && BP_IS_HOLE(db->db_blkptr)) {
/*
* IO was compressed away
*/
*bp = *db->db_blkptr; /* structure assignment */
mutex_exit(&db->db_mtx);
txg_resume(dp);
return (0);
}
ASSERT(db->db_data_pending ||
(db->db_blkptr && db->db_blkptr->blk_birth == txg));
}
if (db->db_blkptr && db->db_blkptr->blk_birth == txg) {
/*
* IO is already completed.
*/
*bp = *db->db_blkptr; /* structure assignment */
mutex_exit(&db->db_mtx);
txg_resume(dp);
return (0);
}
}
dr = db->db_last_dirty;
while (dr && dr->dr_txg > txg)
dr = dr->dr_next;
if (dr == NULL || dr->dr_txg < txg) {
/*
* This dbuf isn't dirty, must have been free_range'd.
* There's no need to log writes to freed blocks, so we're done.
*/
mutex_exit(&db->db_mtx);
txg_resume(dp);
return (ENOENT);
}
ASSERT(dr->dr_txg == txg);
if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC) {
/*
* We have already issued a sync write for this buffer.
*/
mutex_exit(&db->db_mtx);
txg_resume(dp);
return (EALREADY);
} else if (dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
/*
* This buffer has already been synced. It could not
* have been dirtied since, or we would have cleared the state.
*/
*bp = dr->dt.dl.dr_overridden_by; /* structure assignment */
mutex_exit(&db->db_mtx);
txg_resume(dp);
return (0);
}
dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
in = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
in->dr = dr;
in->done = done;
in->arg = arg;
mutex_exit(&db->db_mtx);
txg_resume(dp);
zb.zb_objset = os->os_dsl_dataset->ds_object;
zb.zb_object = db->db.db_object;
zb.zb_level = db->db_level;
zb.zb_blkid = db->db_blkid;
zio_flags = ZIO_FLAG_MUSTSUCCEED;
if (dmu_ot[db->db_dnode->dn_type].ot_metadata || zb.zb_level != 0)
zio_flags |= ZIO_FLAG_METADATA;
wp.wp_type = db->db_dnode->dn_type;
wp.wp_copies = os->os_copies;
wp.wp_level = db->db_level;
wp.wp_dnchecksum = db->db_dnode->dn_checksum;
wp.wp_oschecksum = os->os_checksum;
wp.wp_dncompress = db->db_dnode->dn_compress;
wp.wp_oscompress = os->os_compress;
zio = arc_write(pio, os->os_spa, &wp,
DBUF_IS_L2CACHEABLE(db), txg, bp, dr->dt.dl.dr_data, NULL,
dmu_sync_done, in, ZIO_PRIORITY_SYNC_WRITE, zio_flags, &zb);
if (pio) {
zio_nowait(zio);
err = EINPROGRESS;
} else {
err = zio_wait(zio);
ASSERT(err == 0);
}
return (err);
}
int
dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
dmu_tx_t *tx)
{
dnode_t *dn;
int err;
err = dnode_hold(os->os, object, FTAG, &dn);
if (err)
return (err);
err = dnode_set_blksz(dn, size, ibs, tx);
dnode_rele(dn, FTAG);
return (err);
}
void
dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
dmu_tx_t *tx)
{
dnode_t *dn;
/* XXX assumes dnode_hold will not get an i/o error */
(void) dnode_hold(os->os, object, FTAG, &dn);
ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS);
dn->dn_checksum = checksum;
dnode_setdirty(dn, tx);
dnode_rele(dn, FTAG);
}
void
dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
dmu_tx_t *tx)
{
dnode_t *dn;
/* XXX assumes dnode_hold will not get an i/o error */
(void) dnode_hold(os->os, object, FTAG, &dn);
ASSERT(compress < ZIO_COMPRESS_FUNCTIONS);
dn->dn_compress = compress;
dnode_setdirty(dn, tx);
dnode_rele(dn, FTAG);
}
int
dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
{
dnode_t *dn;
int i, err;
err = dnode_hold(os->os, object, FTAG, &dn);
if (err)
return (err);
/*
* Sync any current changes before
* we go trundling through the block pointers.
*/
for (i = 0; i < TXG_SIZE; i++) {
if (list_link_active(&dn->dn_dirty_link[i]))
break;
}
if (i != TXG_SIZE) {
dnode_rele(dn, FTAG);
txg_wait_synced(dmu_objset_pool(os), 0);
err = dnode_hold(os->os, object, FTAG, &dn);
if (err)
return (err);
}
err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
dnode_rele(dn, FTAG);
return (err);
}
void
dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
{
rw_enter(&dn->dn_struct_rwlock, RW_READER);
mutex_enter(&dn->dn_mtx);
doi->doi_data_block_size = dn->dn_datablksz;
doi->doi_metadata_block_size = dn->dn_indblkshift ?
1ULL << dn->dn_indblkshift : 0;
doi->doi_indirection = dn->dn_nlevels;
doi->doi_checksum = dn->dn_checksum;
doi->doi_compress = dn->dn_compress;
doi->doi_physical_blks = (DN_USED_BYTES(dn->dn_phys) +
SPA_MINBLOCKSIZE/2) >> SPA_MINBLOCKSHIFT;
doi->doi_max_block_offset = dn->dn_phys->dn_maxblkid;
doi->doi_type = dn->dn_type;
doi->doi_bonus_size = dn->dn_bonuslen;
doi->doi_bonus_type = dn->dn_bonustype;
mutex_exit(&dn->dn_mtx);
rw_exit(&dn->dn_struct_rwlock);
}
/*
* Get information on a DMU object.
* If doi is NULL, just indicates whether the object exists.
*/
int
dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
{
dnode_t *dn;
int err = dnode_hold(os->os, object, FTAG, &dn);
if (err)
return (err);
if (doi != NULL)
dmu_object_info_from_dnode(dn, doi);
dnode_rele(dn, FTAG);
return (0);
}
/*
* As above, but faster; can be used when you have a held dbuf in hand.
*/
void
dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi)
{
dmu_object_info_from_dnode(((dmu_buf_impl_t *)db)->db_dnode, doi);
}
/*
* Faster still when you only care about the size.
* This is specifically optimized for zfs_getattr().
*/
void
dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, u_longlong_t *nblk512)
{
dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode;
*blksize = dn->dn_datablksz;
/* add 1 for dnode space */
*nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
SPA_MINBLOCKSHIFT) + 1;
}
void
byteswap_uint64_array(void *vbuf, size_t size)
{
uint64_t *buf = vbuf;
size_t count = size >> 3;
int i;
ASSERT((size & 7) == 0);
for (i = 0; i < count; i++)
buf[i] = BSWAP_64(buf[i]);
}
void
byteswap_uint32_array(void *vbuf, size_t size)
{
uint32_t *buf = vbuf;
size_t count = size >> 2;
int i;
ASSERT((size & 3) == 0);
for (i = 0; i < count; i++)
buf[i] = BSWAP_32(buf[i]);
}
void
byteswap_uint16_array(void *vbuf, size_t size)
{
uint16_t *buf = vbuf;
size_t count = size >> 1;
int i;
ASSERT((size & 1) == 0);
for (i = 0; i < count; i++)
buf[i] = BSWAP_16(buf[i]);
}
/* ARGSUSED */
void
byteswap_uint8_array(void *vbuf, size_t size)
{
}
void
dmu_init(void)
{
dbuf_init();
dnode_init();
arc_init();
l2arc_init();
}
void
dmu_fini(void)
{
arc_fini();
dnode_fini();
dbuf_fini();
l2arc_fini();
}