*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
*/
#include <sys/dmu.h>
#include <sys/zfs_ioctl.h>
#include <sys/zap.h>
#include <sys/zio_checksum.h>
+#include <sys/zio_compress.h>
#include <sys/sa.h>
#ifdef _KERNEL
#include <sys/vmsystm.h>
#include <sys/zfs_znode.h>
#endif
+/*
+ * Enable/disable nopwrite feature.
+ */
+int zfs_nopwrite_enabled = 1;
+
const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
- { byteswap_uint8_array, TRUE, "unallocated" },
- { zap_byteswap, TRUE, "object directory" },
- { byteswap_uint64_array, TRUE, "object array" },
- { byteswap_uint8_array, TRUE, "packed nvlist" },
- { byteswap_uint64_array, TRUE, "packed nvlist size" },
- { byteswap_uint64_array, TRUE, "bpobj" },
- { byteswap_uint64_array, TRUE, "bpobj header" },
- { byteswap_uint64_array, TRUE, "SPA space map header" },
- { byteswap_uint64_array, TRUE, "SPA space map" },
- { byteswap_uint64_array, TRUE, "ZIL intent log" },
- { dnode_buf_byteswap, TRUE, "DMU dnode" },
- { dmu_objset_byteswap, TRUE, "DMU objset" },
- { byteswap_uint64_array, TRUE, "DSL directory" },
- { zap_byteswap, TRUE, "DSL directory child map"},
- { zap_byteswap, TRUE, "DSL dataset snap map" },
- { zap_byteswap, TRUE, "DSL props" },
- { byteswap_uint64_array, TRUE, "DSL dataset" },
- { zfs_znode_byteswap, TRUE, "ZFS znode" },
- { zfs_oldacl_byteswap, TRUE, "ZFS V0 ACL" },
- { byteswap_uint8_array, FALSE, "ZFS plain file" },
- { zap_byteswap, TRUE, "ZFS directory" },
- { zap_byteswap, TRUE, "ZFS master node" },
- { zap_byteswap, TRUE, "ZFS delete queue" },
- { byteswap_uint8_array, FALSE, "zvol object" },
- { zap_byteswap, TRUE, "zvol prop" },
- { byteswap_uint8_array, FALSE, "other uint8[]" },
- { byteswap_uint64_array, FALSE, "other uint64[]" },
- { zap_byteswap, TRUE, "other ZAP" },
- { zap_byteswap, TRUE, "persistent error log" },
- { byteswap_uint8_array, TRUE, "SPA history" },
- { byteswap_uint64_array, TRUE, "SPA history offsets" },
- { zap_byteswap, TRUE, "Pool properties" },
- { zap_byteswap, TRUE, "DSL permissions" },
- { zfs_acl_byteswap, TRUE, "ZFS ACL" },
- { byteswap_uint8_array, TRUE, "ZFS SYSACL" },
- { byteswap_uint8_array, TRUE, "FUID table" },
- { byteswap_uint64_array, TRUE, "FUID table size" },
- { zap_byteswap, TRUE, "DSL dataset next clones"},
- { zap_byteswap, TRUE, "scan work queue" },
- { zap_byteswap, TRUE, "ZFS user/group used" },
- { zap_byteswap, TRUE, "ZFS user/group quota" },
- { zap_byteswap, TRUE, "snapshot refcount tags"},
- { zap_byteswap, TRUE, "DDT ZAP algorithm" },
- { zap_byteswap, TRUE, "DDT statistics" },
- { byteswap_uint8_array, TRUE, "System attributes" },
- { zap_byteswap, TRUE, "SA master node" },
- { zap_byteswap, TRUE, "SA attr registration" },
- { zap_byteswap, TRUE, "SA attr layouts" },
- { zap_byteswap, TRUE, "scan translations" },
- { byteswap_uint8_array, FALSE, "deduplicated block" },
- { zap_byteswap, TRUE, "DSL deadlist map" },
- { byteswap_uint64_array, TRUE, "DSL deadlist map hdr" },
- { zap_byteswap, TRUE, "DSL dir clones" },
- { byteswap_uint64_array, TRUE, "bpobj subobj" },
+ { DMU_BSWAP_UINT8, TRUE, "unallocated" },
+ { DMU_BSWAP_ZAP, TRUE, "object directory" },
+ { DMU_BSWAP_UINT64, TRUE, "object array" },
+ { DMU_BSWAP_UINT8, TRUE, "packed nvlist" },
+ { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" },
+ { DMU_BSWAP_UINT64, TRUE, "bpobj" },
+ { DMU_BSWAP_UINT64, TRUE, "bpobj header" },
+ { DMU_BSWAP_UINT64, TRUE, "SPA space map header" },
+ { DMU_BSWAP_UINT64, TRUE, "SPA space map" },
+ { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" },
+ { DMU_BSWAP_DNODE, TRUE, "DMU dnode" },
+ { DMU_BSWAP_OBJSET, TRUE, "DMU objset" },
+ { DMU_BSWAP_UINT64, TRUE, "DSL directory" },
+ { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"},
+ { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" },
+ { DMU_BSWAP_ZAP, TRUE, "DSL props" },
+ { DMU_BSWAP_UINT64, TRUE, "DSL dataset" },
+ { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" },
+ { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" },
+ { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" },
+ { DMU_BSWAP_ZAP, TRUE, "ZFS directory" },
+ { DMU_BSWAP_ZAP, TRUE, "ZFS master node" },
+ { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" },
+ { DMU_BSWAP_UINT8, FALSE, "zvol object" },
+ { DMU_BSWAP_ZAP, TRUE, "zvol prop" },
+ { DMU_BSWAP_UINT8, FALSE, "other uint8[]" },
+ { DMU_BSWAP_UINT64, FALSE, "other uint64[]" },
+ { DMU_BSWAP_ZAP, TRUE, "other ZAP" },
+ { DMU_BSWAP_ZAP, TRUE, "persistent error log" },
+ { DMU_BSWAP_UINT8, TRUE, "SPA history" },
+ { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" },
+ { DMU_BSWAP_ZAP, TRUE, "Pool properties" },
+ { DMU_BSWAP_ZAP, TRUE, "DSL permissions" },
+ { DMU_BSWAP_ACL, TRUE, "ZFS ACL" },
+ { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" },
+ { DMU_BSWAP_UINT8, TRUE, "FUID table" },
+ { DMU_BSWAP_UINT64, TRUE, "FUID table size" },
+ { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"},
+ { DMU_BSWAP_ZAP, TRUE, "scan work queue" },
+ { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" },
+ { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" },
+ { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"},
+ { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" },
+ { DMU_BSWAP_ZAP, TRUE, "DDT statistics" },
+ { DMU_BSWAP_UINT8, TRUE, "System attributes" },
+ { DMU_BSWAP_ZAP, TRUE, "SA master node" },
+ { DMU_BSWAP_ZAP, TRUE, "SA attr registration" },
+ { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" },
+ { DMU_BSWAP_ZAP, TRUE, "scan translations" },
+ { DMU_BSWAP_UINT8, FALSE, "deduplicated block" },
+ { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" },
+ { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" },
+ { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" },
+ { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" }
+};
+
+const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
+ { byteswap_uint8_array, "uint8" },
+ { byteswap_uint16_array, "uint16" },
+ { byteswap_uint32_array, "uint32" },
+ { byteswap_uint64_array, "uint64" },
+ { zap_byteswap, "zap" },
+ { dnode_buf_byteswap, "dnode" },
+ { dmu_objset_byteswap, "objset" },
+ { zfs_znode_byteswap, "znode" },
+ { zfs_oldacl_byteswap, "oldacl" },
+ { zfs_acl_byteswap, "acl" }
};
int
db = dbuf_hold(dn, blkid, tag);
rw_exit(&dn->dn_struct_rwlock);
if (db == NULL) {
- err = EIO;
+ err = SET_ERROR(EIO);
} else {
err = dbuf_read(db, NULL, db_flags);
if (err) {
dn = DB_DNODE(db);
if (dn->dn_bonus != db) {
- error = EINVAL;
+ error = SET_ERROR(EINVAL);
} else if (newsize < 0 || newsize > db_fake->db_size) {
- error = EINVAL;
+ error = SET_ERROR(EINVAL);
} else {
dnode_setbonuslen(dn, newsize, tx);
error = 0;
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
- if (type > DMU_OT_NUMTYPES) {
- error = EINVAL;
+ if (!DMU_OT_IS_VALID(type)) {
+ error = SET_ERROR(EINVAL);
} else if (dn->dn_bonus != db) {
- error = EINVAL;
+ error = SET_ERROR(EINVAL);
} else {
dnode_setbonus_type(dn, type, tx);
error = 0;
dn = DB_DNODE(db);
if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
- err = EINVAL;
+ err = SET_ERROR(EINVAL);
} else {
rw_enter(&dn->dn_struct_rwlock, RW_READER);
if (!dn->dn_have_spill) {
- err = ENOENT;
+ err = SET_ERROR(ENOENT);
} else {
err = dmu_spill_hold_by_dnode(dn,
DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
{
- dsl_pool_t *dp = NULL;
dmu_buf_t **dbp;
uint64_t blkid, nblks, i;
uint32_t dbuf_flags;
int err;
zio_t *zio;
- hrtime_t start;
ASSERT(length <= DMU_MAX_ACCESS);
(longlong_t)dn->dn_object, dn->dn_datablksz,
(longlong_t)offset, (longlong_t)length);
rw_exit(&dn->dn_struct_rwlock);
- return (EIO);
+ return (SET_ERROR(EIO));
}
nblks = 1;
}
- dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
+ dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks,
+ KM_PUSHPAGE | KM_NODEBUG);
- if (dn->dn_objset->os_dsl_dataset)
- dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool;
- if (dp && dsl_pool_sync_context(dp))
- start = gethrtime();
zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
blkid = dbuf_whichblock(dn, offset);
for (i = 0; i < nblks; i++) {
rw_exit(&dn->dn_struct_rwlock);
dmu_buf_rele_array(dbp, nblks, tag);
zio_nowait(zio);
- return (EIO);
+ return (SET_ERROR(EIO));
}
/* initiate async i/o */
if (read) {
/* wait for async i/o */
err = zio_wait(zio);
- /* track read overhead when we are in sync context */
- if (dp && dsl_pool_sync_context(dp))
- dp->dp_read_overhead += gethrtime() - start;
if (err) {
dmu_buf_rele_array(dbp, nblks, tag);
return (err);
db->db_state == DB_FILL)
cv_wait(&db->db_changed, &db->db_mtx);
if (db->db_state == DB_UNCACHED)
- err = EIO;
+ err = SET_ERROR(EIO);
mutex_exit(&db->db_mtx);
if (err) {
dmu_buf_rele_array(dbp, nblks, tag);
kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
}
+/*
+ * Issue prefetch i/os for the given blocks.
+ *
+ * Note: The assumption is that we *know* these blocks will be needed
+ * almost immediately. Therefore, the prefetch i/os will be issued at
+ * ZIO_PRIORITY_SYNC_READ
+ *
+ * Note: indirect blocks and other metadata will be read synchronously,
+ * causing this function to block if they are not already cached.
+ */
void
dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
{
dnode_t *dn;
uint64_t blkid;
- int nblks, i, err;
+ int nblks, err;
if (zfs_prefetch_disable)
return;
rw_enter(&dn->dn_struct_rwlock, RW_READER);
blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t));
- dbuf_prefetch(dn, blkid);
+ dbuf_prefetch(dn, blkid, ZIO_PRIORITY_SYNC_READ);
rw_exit(&dn->dn_struct_rwlock);
return;
}
rw_enter(&dn->dn_struct_rwlock, RW_READER);
if (dn->dn_datablkshift) {
int blkshift = dn->dn_datablkshift;
- nblks = (P2ROUNDUP(offset+len, 1<<blkshift) -
- P2ALIGN(offset, 1<<blkshift)) >> blkshift;
+ nblks = (P2ROUNDUP(offset + len, 1 << blkshift) -
+ P2ALIGN(offset, 1 << blkshift)) >> blkshift;
} else {
nblks = (offset < dn->dn_datablksz);
}
if (nblks != 0) {
+ int i;
+
blkid = dbuf_whichblock(dn, offset);
for (i = 0; i < nblks; i++)
- dbuf_prefetch(dn, blkid+i);
+ dbuf_prefetch(dn, blkid + i, ZIO_PRIORITY_SYNC_READ);
}
rw_exit(&dn->dn_struct_rwlock);
* the end so that the file gets shorter over time (if we crashes in the
* middle, this will leave us in a better state). We find allocated file
* data by simply searching the allocated level 1 indirects.
+ *
+ * On input, *start should be the first offset that does not need to be
+ * freed (e.g. "offset + length"). On return, *start will be the first
+ * offset that should be freed.
*/
static int
-get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit)
+get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum)
{
- uint64_t len = *start - limit;
- uint64_t blkcnt = 0;
- uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1));
+ uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1);
+ /* bytes of data covered by a level-1 indirect block */
uint64_t iblkrange =
dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
+ uint64_t blks;
- ASSERT(limit <= *start);
+ ASSERT3U(minimum, <=, *start);
- if (len <= iblkrange * maxblks) {
- *start = limit;
+ if (*start - minimum <= iblkrange * maxblks) {
+ *start = minimum;
return (0);
}
ASSERT(ISP2(iblkrange));
- while (*start > limit && blkcnt < maxblks) {
+ for (blks = 0; *start > minimum && blks < maxblks; blks++) {
int err;
- /* find next allocated L1 indirect */
+ /*
+ * dnode_next_offset(BACKWARDS) will find an allocated L1
+ * indirect block at or before the input offset. We must
+ * decrement *start so that it is at the end of the region
+ * to search.
+ */
+ (*start)--;
err = dnode_next_offset(dn,
DNODE_FIND_BACKWARDS, start, 2, 1, 0);
- /* if there are no more, then we are done */
+ /* if there are no indirect blocks before start, we are done */
if (err == ESRCH) {
- *start = limit;
- return (0);
- } else if (err) {
+ *start = minimum;
+ break;
+ } else if (err != 0) {
return (err);
}
- blkcnt += 1;
- /* reset offset to end of "next" block back */
+ /* set start to the beginning of this L1 indirect */
*start = P2ALIGN(*start, iblkrange);
- if (*start <= limit)
- *start = limit;
- else
- *start -= 1;
}
+ if (*start < minimum)
+ *start = minimum;
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)
+ uint64_t length)
{
- 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;
-
- end = offset + length;
- if (trunc || end > object_size)
- end = object_size;
- if (end <= offset)
+ uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
+ int err;
+
+ if (offset >= object_size)
return (0);
- length = end - offset;
- while (length) {
- start = end;
- /* assert(offset <= start) */
- err = get_next_chunk(dn, &start, offset);
+ if (length == DMU_OBJECT_END || offset + length > object_size)
+ length = object_size - offset;
+
+ while (length != 0) {
+ uint64_t chunk_end, chunk_begin;
+ dmu_tx_t *tx;
+
+ chunk_end = chunk_begin = offset + length;
+
+ /* move chunk_begin backwards to the beginning of this chunk */
+ err = get_next_chunk(dn, &chunk_begin, offset);
if (err)
return (err);
- len = trunc ? DMU_OBJECT_END : end - start;
+ ASSERT3U(chunk_begin, >=, offset);
+ ASSERT3U(chunk_begin, <=, chunk_end);
tx = dmu_tx_create(os);
- dmu_tx_hold_free(tx, dn->dn_object, start, len);
+ dmu_tx_hold_free(tx, dn->dn_object,
+ chunk_begin, chunk_end - chunk_begin);
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 && free_dnode) {
- ASSERT(trunc);
- dnode_free(dn, tx);
- }
-
- length -= end - start;
-
+ dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx);
dmu_tx_commit(tx);
- end = start;
+
+ length -= chunk_end - chunk_begin;
}
return (0);
}
err = dnode_hold(os, object, FTAG, &dn);
if (err != 0)
return (err);
- err = dmu_free_long_range_impl(os, dn, offset, length, FALSE);
+ err = dmu_free_long_range_impl(os, dn, offset, length);
+
+ /*
+ * It is important to zero out the maxblkid when freeing the entire
+ * file, so that (a) subsequent calls to dmu_free_long_range_impl()
+ * will take the fast path, and (b) dnode_reallocate() can verify
+ * that the entire file has been freed.
+ */
+ if (offset == 0 && length == DMU_OBJECT_END)
+ dn->dn_maxblkid = 0;
+
dnode_rele(dn, FTAG);
return (err);
}
int
-dmu_free_object(objset_t *os, uint64_t object)
+dmu_free_long_object(objset_t *os, uint64_t object)
{
- dnode_t *dn;
dmu_tx_t *tx;
int err;
- err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED,
- FTAG, &dn);
+ err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END);
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);
- }
+
+ tx = dmu_tx_create(os);
+ dmu_tx_hold_bonus(tx, object);
+ dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
+ err = dmu_tx_assign(tx, TXG_WAIT);
+ if (err == 0) {
+ err = dmu_object_free(os, object, tx);
+ dmu_tx_commit(tx);
} else {
- err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE);
+ dmu_tx_abort(tx);
}
- dnode_rele(dn, FTAG);
+
return (err);
}
else
dmu_buf_will_dirty(db, tx);
- bcopy(buf, (char *)db->db_data + bufoff, tocpy);
+ (void) memcpy((char *)db->db_data + bufoff, buf, tocpy);
if (tocpy == db->db_size)
dmu_buf_fill_done(db, tx);
*/
kstat_t *xuio_ksp = NULL;
+typedef struct xuio_stats {
+ /* loaned yet not returned arc_buf */
+ kstat_named_t xuiostat_onloan_rbuf;
+ kstat_named_t xuiostat_onloan_wbuf;
+ /* whether a copy is made when loaning out a read buffer */
+ kstat_named_t xuiostat_rbuf_copied;
+ kstat_named_t xuiostat_rbuf_nocopy;
+ /* whether a copy is made when assigning a write buffer */
+ kstat_named_t xuiostat_wbuf_copied;
+ kstat_named_t xuiostat_wbuf_nocopy;
+} xuio_stats_t;
+
+static xuio_stats_t xuio_stats = {
+ { "onloan_read_buf", KSTAT_DATA_UINT64 },
+ { "onloan_write_buf", KSTAT_DATA_UINT64 },
+ { "read_buf_copied", KSTAT_DATA_UINT64 },
+ { "read_buf_nocopy", KSTAT_DATA_UINT64 },
+ { "write_buf_copied", KSTAT_DATA_UINT64 },
+ { "write_buf_nocopy", KSTAT_DATA_UINT64 }
+};
+
+#define XUIOSTAT_INCR(stat, val) \
+ atomic_add_64(&xuio_stats.stat.value.ui64, (val))
+#define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1)
+
int
dmu_xuio_init(xuio_t *xuio, int nblk)
{
uio_t *uio = &xuio->xu_uio;
uio->uio_iovcnt = nblk;
- uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
+ uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_PUSHPAGE);
- priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
+ priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_PUSHPAGE);
priv->cnt = nblk;
- priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
+ priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_PUSHPAGE);
priv->iovp = uio->uio_iov;
XUIO_XUZC_PRIV(xuio) = priv;
}
#ifdef _KERNEL
+
+/*
+ * Copy up to size bytes between arg_buf and req based on the data direction
+ * described by the req. If an entire req's data cannot be transfered the
+ * req's is updated such that it's current index and bv offsets correctly
+ * reference any residual data which could not be copied. The return value
+ * is the number of bytes successfully copied to arg_buf.
+ */
+static int
+dmu_req_copy(void *arg_buf, int size, int *offset, struct request *req)
+{
+ struct bio_vec *bv;
+ struct req_iterator iter;
+ char *bv_buf;
+ int tocpy;
+
+ *offset = 0;
+ rq_for_each_segment(bv, req, iter) {
+
+ /* Fully consumed the passed arg_buf */
+ ASSERT3S(*offset, <=, size);
+ if (size == *offset)
+ break;
+
+ /* Skip fully consumed bv's */
+ if (bv->bv_len == 0)
+ continue;
+
+ tocpy = MIN(bv->bv_len, size - *offset);
+ ASSERT3S(tocpy, >=, 0);
+
+ bv_buf = page_address(bv->bv_page) + bv->bv_offset;
+ ASSERT3P(bv_buf, !=, NULL);
+
+ if (rq_data_dir(req) == WRITE)
+ memcpy(arg_buf + *offset, bv_buf, tocpy);
+ else
+ memcpy(bv_buf, arg_buf + *offset, tocpy);
+
+ *offset += tocpy;
+ bv->bv_offset += tocpy;
+ bv->bv_len -= tocpy;
+ }
+
+ return (0);
+}
+
+static void
+dmu_bio_put(struct bio *bio)
+{
+ struct bio *bio_next;
+
+ while (bio) {
+ bio_next = bio->bi_next;
+ bio_put(bio);
+ bio = bio_next;
+ }
+}
+
+static int
+dmu_bio_clone(struct bio *bio, struct bio **bio_copy)
+{
+ struct bio *bio_root = NULL;
+ struct bio *bio_last = NULL;
+ struct bio *bio_new;
+
+ if (bio == NULL)
+ return (EINVAL);
+
+ while (bio) {
+ bio_new = bio_clone(bio, GFP_NOIO);
+ if (bio_new == NULL) {
+ dmu_bio_put(bio_root);
+ return (ENOMEM);
+ }
+
+ if (bio_last) {
+ bio_last->bi_next = bio_new;
+ bio_last = bio_new;
+ } else {
+ bio_root = bio_new;
+ bio_last = bio_new;
+ }
+
+ bio = bio->bi_next;
+ }
+
+ *bio_copy = bio_root;
+
+ return (0);
+}
+
+int
+dmu_read_req(objset_t *os, uint64_t object, struct request *req)
+{
+ uint64_t size = blk_rq_bytes(req);
+ uint64_t offset = blk_rq_pos(req) << 9;
+ struct bio *bio_saved = req->bio;
+ 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, offset, size, TRUE, FTAG,
+ &numbufs, &dbp);
+ if (err)
+ return (err);
+
+ /*
+ * Clone the bio list so the bv->bv_offset and bv->bv_len members
+ * can be safely modified. The original bio list is relinked in to
+ * the request when the function exits. This is required because
+ * some file systems blindly assume that these values will remain
+ * constant between bio_submit() and the IO completion callback.
+ */
+ err = dmu_bio_clone(bio_saved, &req->bio);
+ if (err)
+ goto error;
+
+ for (i = 0; i < numbufs; i++) {
+ int tocpy, didcpy, bufoff;
+ dmu_buf_t *db = dbp[i];
+
+ bufoff = offset - db->db_offset;
+ ASSERT3S(bufoff, >=, 0);
+
+ tocpy = (int)MIN(db->db_size - bufoff, size);
+ if (tocpy == 0)
+ break;
+
+ err = dmu_req_copy(db->db_data + bufoff, tocpy, &didcpy, req);
+
+ if (didcpy < tocpy)
+ err = EIO;
+
+ if (err)
+ break;
+
+ size -= tocpy;
+ offset += didcpy;
+ err = 0;
+ }
+
+ dmu_bio_put(req->bio);
+ req->bio = bio_saved;
+error:
+ dmu_buf_rele_array(dbp, numbufs, FTAG);
+
+ return (err);
+}
+
+int
+dmu_write_req(objset_t *os, uint64_t object, struct request *req, dmu_tx_t *tx)
+{
+ uint64_t size = blk_rq_bytes(req);
+ uint64_t offset = blk_rq_pos(req) << 9;
+ struct bio *bio_saved = req->bio;
+ dmu_buf_t **dbp;
+ int numbufs;
+ int err = 0;
+ int i;
+
+ if (size == 0)
+ return (0);
+
+ err = dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG,
+ &numbufs, &dbp);
+ if (err)
+ return (err);
+
+ /*
+ * Clone the bio list so the bv->bv_offset and bv->bv_len members
+ * can be safely modified. The original bio list is relinked in to
+ * the request when the function exits. This is required because
+ * some file systems blindly assume that these values will remain
+ * constant between bio_submit() and the IO completion callback.
+ */
+ err = dmu_bio_clone(bio_saved, &req->bio);
+ if (err)
+ goto error;
+
+ for (i = 0; i < numbufs; i++) {
+ int tocpy, didcpy, bufoff;
+ dmu_buf_t *db = dbp[i];
+
+ bufoff = offset - db->db_offset;
+ ASSERT3S(bufoff, >=, 0);
+
+ tocpy = (int)MIN(db->db_size - bufoff, size);
+ if (tocpy == 0)
+ break;
+
+ 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);
+
+ err = dmu_req_copy(db->db_data + bufoff, tocpy, &didcpy, req);
+
+ if (tocpy == db->db_size)
+ dmu_buf_fill_done(db, tx);
+
+ if (didcpy < tocpy)
+ err = EIO;
+
+ if (err)
+ break;
+
+ size -= tocpy;
+ offset += didcpy;
+ err = 0;
+ }
+
+ dmu_bio_put(req->bio);
+ req->bio = bio_saved;
+error:
+ dmu_buf_rele_array(dbp, numbufs, FTAG);
+
+ return (err);
+}
+
int
dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
{
if (err)
return (err);
- if (uio->uio_extflg == UIO_XUIO)
- xuio = (xuio_t *)uio;
-
for (i = 0; i < numbufs; i++) {
int tocpy;
int bufoff;
dmu_buf_will_dirty(db, tx);
/*
- * XXX uiomove could block forever (eg. nfs-backed
+ * 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.
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 = zfs_map_page(pp, S_READ);
- bcopy(va, (char *)db->db_data + bufoff, thiscpy);
- zfs_unmap_page(pp, va);
- pp = pp->p_next;
- bufoff += PAGESIZE;
- }
-
- if (tocpy == db->db_size)
- dmu_buf_fill_done(db, tx);
-
- offset += tocpy;
- size -= tocpy;
- }
- dmu_buf_rele_array(dbp, numbufs, FTAG);
- return (err);
-}
-#endif
+#endif /* _KERNEL */
/*
* Allocate a loaned anonymous arc buffer.
dmu_return_arcbuf(arc_buf_t *buf)
{
arc_return_buf(buf, FTAG);
- VERIFY(arc_buf_remove_ref(buf, FTAG) == 1);
+ VERIFY(arc_buf_remove_ref(buf, FTAG));
}
/*
mutex_enter(&db->db_mtx);
ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
if (zio->io_error == 0) {
+ dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE);
+ if (dr->dt.dl.dr_nopwrite) {
+ ASSERTV(blkptr_t *bp = zio->io_bp);
+ ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig);
+ ASSERTV(uint8_t chksum = BP_GET_CHECKSUM(bp_orig));
+
+ ASSERT(BP_EQUAL(bp, bp_orig));
+ ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF);
+ ASSERT(zio_checksum_table[chksum].ci_dedup);
+ }
dr->dt.dl.dr_overridden_by = *zio->io_bp;
dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
{
blkptr_t *bp = zio->io_bp;
dmu_sync_arg_t *dsa = zio->io_private;
+ ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig);
if (zio->io_error == 0 && !BP_IS_HOLE(bp)) {
- ASSERT(zio->io_bp->blk_birth == zio->io_txg);
- ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
- zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
+ /*
+ * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE)
+ * then there is nothing to do here. Otherwise, free the
+ * newly allocated block in this txg.
+ */
+ if (zio->io_flags & ZIO_FLAG_NOPWRITE) {
+ ASSERT(BP_EQUAL(bp, bp_orig));
+ } else {
+ ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig));
+ ASSERT(zio->io_bp->blk_birth == zio->io_txg);
+ ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
+ zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
+ }
}
dmu_tx_commit(dsa->dsa_tx);
dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
dmu_tx_abort(tx);
- return (EIO); /* Make zl_get_data do txg_waited_synced() */
+ /* Make zl_get_data do txg_waited_synced() */
+ return (SET_ERROR(EIO));
}
- dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
+ dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_PUSHPAGE);
dsa->dsa_dr = NULL;
dsa->dsa_done = done;
dsa->dsa_zgd = zgd;
zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp,
- dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa,
- ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb));
+ dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, dsa,
+ ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL|ZIO_FLAG_FASTWRITE, zb));
return (0);
}
*
* Return values:
*
- * EEXIST: this txg has already been synced, so there's nothing to to.
+ * EEXIST: this txg has already been synced, so there's nothing to do.
* The caller should not log the write.
*
* ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
dnode_t *dn;
ASSERT(pio != NULL);
- ASSERT(BP_IS_HOLE(bp));
ASSERT(txg != 0);
SET_BOOKMARK(&zb, ds->ds_object,
* This txg has already synced. There's nothing to do.
*/
mutex_exit(&db->db_mtx);
- return (EEXIST);
+ return (SET_ERROR(EEXIST));
}
if (txg <= spa_syncing_txg(os->os_spa)) {
* There's no need to log writes to freed blocks, so we're done.
*/
mutex_exit(&db->db_mtx);
- return (ENOENT);
+ return (SET_ERROR(ENOENT));
}
+ ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg);
+
+ /*
+ * Assume the on-disk data is X, the current syncing data is Y,
+ * and the current in-memory data is Z (currently in dmu_sync).
+ * X and Z are identical but Y is has been modified. Normally,
+ * when X and Z are the same we will perform a nopwrite but if Y
+ * is different we must disable nopwrite since the resulting write
+ * of Y to disk can free the block containing X. If we allowed a
+ * nopwrite to occur the block pointing to Z would reference a freed
+ * block. Since this is a rare case we simplify this by disabling
+ * nopwrite if the current dmu_sync-ing dbuf has been modified in
+ * a previous transaction.
+ */
+ if (dr->dr_next)
+ zp.zp_nopwrite = B_FALSE;
+
ASSERT(dr->dr_txg == txg);
if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
* have been dirtied since, or we would have cleared the state.
*/
mutex_exit(&db->db_mtx);
- return (EALREADY);
+ return (SET_ERROR(EALREADY));
}
ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
mutex_exit(&db->db_mtx);
- dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
+ dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_PUSHPAGE);
dsa->dsa_dr = dr;
dsa->dsa_done = done;
dsa->dsa_zgd = zgd;
dsa->dsa_tx = NULL;
zio_nowait(arc_write(pio, os->os_spa, txg,
- bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), &zp,
- dmu_sync_ready, dmu_sync_done, dsa,
- ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb));
+ bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db),
+ DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready,
+ NULL, dmu_sync_done, dsa, ZIO_PRIORITY_SYNC_WRITE,
+ ZIO_FLAG_CANFAIL, &zb));
return (0);
}
dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
{
dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
- boolean_t ismd = (level > 0 || dmu_ot[type].ot_metadata ||
+ boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
(wp & WP_SPILL));
enum zio_checksum checksum = os->os_checksum;
enum zio_compress compress = os->os_compress;
enum zio_checksum dedup_checksum = os->os_dedup_checksum;
- boolean_t dedup;
+ boolean_t dedup = B_FALSE;
+ boolean_t nopwrite = B_FALSE;
boolean_t dedup_verify = os->os_dedup_verify;
int copies = os->os_copies;
/*
- * Determine checksum setting.
+ * We maintain different write policies for each of the following
+ * types of data:
+ * 1. metadata
+ * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
+ * 3. all other level 0 blocks
*/
if (ismd) {
+ /*
+ * XXX -- we should design a compression algorithm
+ * that specializes in arrays of bps.
+ */
+ compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY :
+ ZIO_COMPRESS_LZJB;
+
/*
* Metadata always gets checksummed. If the data
* checksum is multi-bit correctable, and it's not a
if (zio_checksum_table[checksum].ci_correctable < 1 ||
zio_checksum_table[checksum].ci_eck)
checksum = ZIO_CHECKSUM_FLETCHER_4;
- } else {
- checksum = zio_checksum_select(dn->dn_checksum, checksum);
- }
+ } else if (wp & WP_NOFILL) {
+ ASSERT(level == 0);
- /*
- * Determine compression setting.
- */
- if (ismd) {
/*
- * XXX -- we should design a compression algorithm
- * that specializes in arrays of bps.
+ * If we're writing preallocated blocks, we aren't actually
+ * writing them so don't set any policy properties. These
+ * blocks are currently only used by an external subsystem
+ * outside of zfs (i.e. dump) and not written by the zio
+ * pipeline.
*/
- compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY :
- ZIO_COMPRESS_LZJB;
+ compress = ZIO_COMPRESS_OFF;
+ checksum = ZIO_CHECKSUM_OFF;
} else {
compress = zio_compress_select(dn->dn_compress, compress);
- }
- /*
- * Determine dedup setting. If we are in dmu_sync(), we won't
- * actually dedup now because that's all done in syncing context;
- * but we do want to use the dedup checkum. If the checksum is not
- * strong enough to ensure unique signatures, force dedup_verify.
- */
- dedup = (!ismd && dedup_checksum != ZIO_CHECKSUM_OFF);
- if (dedup) {
- checksum = dedup_checksum;
- if (!zio_checksum_table[checksum].ci_dedup)
- dedup_verify = 1;
- }
+ checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ?
+ zio_checksum_select(dn->dn_checksum, checksum) :
+ dedup_checksum;
- if (wp & WP_DMU_SYNC)
- dedup = 0;
+ /*
+ * Determine dedup setting. If we are in dmu_sync(),
+ * we won't actually dedup now because that's all
+ * done in syncing context; but we do want to use the
+ * dedup checkum. If the checksum is not strong
+ * enough to ensure unique signatures, force
+ * dedup_verify.
+ */
+ if (dedup_checksum != ZIO_CHECKSUM_OFF) {
+ dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE;
+ if (!zio_checksum_table[checksum].ci_dedup)
+ dedup_verify = B_TRUE;
+ }
- if (wp & WP_NOFILL) {
- ASSERT(!ismd && level == 0);
- checksum = ZIO_CHECKSUM_OFF;
- compress = ZIO_COMPRESS_OFF;
- dedup = B_FALSE;
+ /*
+ * Enable nopwrite if we have a cryptographically secure
+ * checksum that has no known collisions (i.e. SHA-256)
+ * and compression is enabled. We don't enable nopwrite if
+ * dedup is enabled as the two features are mutually exclusive.
+ */
+ nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup &&
+ compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
}
zp->zp_checksum = checksum;
zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa));
zp->zp_dedup = dedup;
zp->zp_dedup_verify = dedup && dedup_verify;
+ zp->zp_nopwrite = nopwrite;
}
int
}
void
-dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
+__dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
{
- dnode_phys_t *dnp;
+ dnode_phys_t *dnp = dn->dn_phys;
int i;
- rw_enter(&dn->dn_struct_rwlock, RW_READER);
- mutex_enter(&dn->dn_mtx);
-
- dnp = dn->dn_phys;
-
doi->doi_data_block_size = dn->dn_datablksz;
doi->doi_metadata_block_size = dn->dn_indblkshift ?
1ULL << dn->dn_indblkshift : 0;
doi->doi_checksum = dn->dn_checksum;
doi->doi_compress = dn->dn_compress;
doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
- doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz;
+ doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
doi->doi_fill_count = 0;
for (i = 0; i < dnp->dn_nblkptr; i++)
doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill;
+}
+
+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);
+
+ __dmu_object_info_from_dnode(dn, doi);
mutex_exit(&dn->dn_mtx);
rw_exit(&dn->dn_struct_rwlock);
dnode_init();
dbuf_init();
zfetch_init();
- arc_init();
+ dmu_tx_init();
l2arc_init();
+ arc_init();
}
void
dmu_fini(void)
{
+ arc_fini(); /* arc depends on l2arc, so arc must go first */
l2arc_fini();
- arc_fini();
+ dmu_tx_fini();
zfetch_fini();
dbuf_fini();
dnode_fini();
sa_cache_fini();
zfs_dbgmsg_fini();
}
+
+#if defined(_KERNEL) && defined(HAVE_SPL)
+EXPORT_SYMBOL(dmu_bonus_hold);
+EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus);
+EXPORT_SYMBOL(dmu_buf_rele_array);
+EXPORT_SYMBOL(dmu_prefetch);
+EXPORT_SYMBOL(dmu_free_range);
+EXPORT_SYMBOL(dmu_free_long_range);
+EXPORT_SYMBOL(dmu_free_long_object);
+EXPORT_SYMBOL(dmu_read);
+EXPORT_SYMBOL(dmu_write);
+EXPORT_SYMBOL(dmu_prealloc);
+EXPORT_SYMBOL(dmu_object_info);
+EXPORT_SYMBOL(dmu_object_info_from_dnode);
+EXPORT_SYMBOL(dmu_object_info_from_db);
+EXPORT_SYMBOL(dmu_object_size_from_db);
+EXPORT_SYMBOL(dmu_object_set_blocksize);
+EXPORT_SYMBOL(dmu_object_set_checksum);
+EXPORT_SYMBOL(dmu_object_set_compress);
+EXPORT_SYMBOL(dmu_write_policy);
+EXPORT_SYMBOL(dmu_sync);
+EXPORT_SYMBOL(dmu_request_arcbuf);
+EXPORT_SYMBOL(dmu_return_arcbuf);
+EXPORT_SYMBOL(dmu_assign_arcbuf);
+EXPORT_SYMBOL(dmu_buf_hold);
+EXPORT_SYMBOL(dmu_ot);
+
+module_param(zfs_mdcomp_disable, int, 0644);
+MODULE_PARM_DESC(zfs_mdcomp_disable, "Disable meta data compression");
+
+module_param(zfs_nopwrite_enabled, int, 0644);
+MODULE_PARM_DESC(zfs_nopwrite_enabled, "Enable NOP writes");
+
+#endif