d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

9539 Make zvol operations use _by_dnode routines

Browse Source 
Continues what was started in 7801 add more by-dnode routines by fully
converting zvols to avoid unnecessary dnode_hold() calls. This saves a
small amount of CPU time and slightly improves latencies of operations
on zvols.

illumos/illumos-gate@8dfe5547fb

Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Rick McNeal <rick.mcneal@nexenta.com>
Approved by: Dan McDonald <danmcd@joyent.com>
Author:     Richard Yao <richard.yao@prophetstor.com>
This commit is contained in:
Alexander Motin 2018-08-02 20:49:08 +00:00
parent 903252254b
commit 8398b8e966
3 changed files with 35 additions and 26 deletions
uts/common/fs/zfs
dmu.c
sys
dmu.h
zvol.c

27
uts/common/fs/zfs/dmu.c

@ -443,7 +443,7 @@ dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
* and can induce severe lock contention when writing to several files * and can induce severe lock contention when writing to several files
* whose dnodes are in the same block. * whose dnodes are in the same block.
*/ */
static int int
dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
{ {
@ -1302,7 +1302,7 @@ xuio_stat_wbuf_nocopy(void)
} }
#ifdef _KERNEL #ifdef _KERNEL
static int int
dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size)
{ {
dmu_buf_t **dbp; dmu_buf_t **dbp;
@ -1411,7 +1411,7 @@ dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
return (err); return (err);
} }
static int int
dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
{ {
dmu_buf_t **dbp; dmu_buf_t **dbp;
@ -1600,22 +1600,17 @@ dmu_return_arcbuf(arc_buf_t *buf)
* dmu_write(). * dmu_write().
*/ */
void void
dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, dmu_assign_arcbuf_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf,
dmu_tx_t *tx) dmu_tx_t *tx)
{ {
dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
dnode_t *dn;
dmu_buf_impl_t *db; dmu_buf_impl_t *db;
uint32_t blksz = (uint32_t)arc_buf_lsize(buf); uint32_t blksz = (uint32_t)arc_buf_lsize(buf);
uint64_t blkid; uint64_t blkid;
DB_DNODE_ENTER(dbuf);
dn = DB_DNODE(dbuf);
rw_enter(&dn->dn_struct_rwlock, RW_READER); rw_enter(&dn->dn_struct_rwlock, RW_READER);
blkid = dbuf_whichblock(dn, 0, offset); blkid = dbuf_whichblock(dn, 0, offset);
VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
rw_exit(&dn->dn_struct_rwlock); rw_exit(&dn->dn_struct_rwlock);
DB_DNODE_EXIT(dbuf);
/* /*
* We can only assign if the offset is aligned, the arc buf is the * We can only assign if the offset is aligned, the arc buf is the
@ -1632,11 +1627,8 @@ dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF);
ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED));
DB_DNODE_ENTER(dbuf);
dn = DB_DNODE(dbuf);
os = dn->dn_objset; os = dn->dn_objset;
object = dn->dn_object; object = dn->dn_object;
DB_DNODE_EXIT(dbuf);
dbuf_rele(db, FTAG); dbuf_rele(db, FTAG);
dmu_write(os, object, offset, blksz, buf->b_data, tx); dmu_write(os, object, offset, blksz, buf->b_data, tx);
@ -1645,6 +1637,17 @@ dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
} }
} }
void
dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
dmu_tx_t *tx)
{
dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
DB_DNODE_ENTER(dbuf);
dmu_assign_arcbuf_dnode(DB_DNODE(dbuf), offset, buf, tx);
DB_DNODE_EXIT(dbuf);
}
typedef struct { typedef struct {
dbuf_dirty_record_t *dsa_dr; dbuf_dirty_record_t *dsa_dr;
dmu_sync_cb_t *dsa_done; dmu_sync_cb_t *dsa_done;

8
uts/common/fs/zfs/sys/dmu.h

@ -517,6 +517,9 @@ uint64_t dmu_buf_refcount(dmu_buf_t *db);
int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset, int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
uint64_t length, boolean_t read, void *tag, uint64_t length, boolean_t read, void *tag,
int *numbufsp, dmu_buf_t ***dbpp); int *numbufsp, dmu_buf_t ***dbpp);
int dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp,
uint32_t flags);
void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag); void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
typedef void dmu_buf_evict_func_t(void *user_ptr); typedef void dmu_buf_evict_func_t(void *user_ptr);
@ -755,14 +758,19 @@ void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
dmu_tx_t *tx); dmu_tx_t *tx);
int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size); int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size); int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
int dmu_read_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size);
int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size, int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
dmu_tx_t *tx); dmu_tx_t *tx);
int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size, int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
dmu_tx_t *tx); dmu_tx_t *tx);
int dmu_write_uio_dnode(dnode_t *dn, struct uio *uio, uint64_t size,
dmu_tx_t *tx);
int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset,
uint64_t size, struct page *pp, dmu_tx_t *tx); uint64_t size, struct page *pp, dmu_tx_t *tx);
struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size); struct arc_buf *dmu_request_arcbuf(dmu_buf_t *handle, int size);
void dmu_return_arcbuf(struct arc_buf *buf); void dmu_return_arcbuf(struct arc_buf *buf);
void dmu_assign_arcbuf_dnode(dnode_t *handle, uint64_t offset,
struct arc_buf *buf, dmu_tx_t *tx);
void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf, void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, struct arc_buf *buf,
dmu_tx_t *tx); dmu_tx_t *tx);
int dmu_xuio_init(struct xuio *uio, int niov); int dmu_xuio_init(struct xuio *uio, int niov);

26
uts/common/fs/zfs/zvol.c

@ -128,7 +128,7 @@ typedef struct zvol_state {
zilog_t *zv_zilog; /* ZIL handle */ zilog_t *zv_zilog; /* ZIL handle */
list_t zv_extents; /* List of extents for dump */ list_t zv_extents; /* List of extents for dump */
znode_t zv_znode; /* for range locking */ znode_t zv_znode; /* for range locking */
dmu_buf_t *zv_dbuf; /* bonus handle */ dnode_t *zv_dn; /* dnode hold */
} zvol_state_t; } zvol_state_t;
/* /*
@ -646,7 +646,7 @@ zvol_first_open(zvol_state_t *zv)
return (error); return (error);
} }
error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); error = dnode_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dn);
if (error) { if (error) {
dmu_objset_disown(os, zvol_tag); dmu_objset_disown(os, zvol_tag);
return (error); return (error);
@ -671,8 +671,8 @@ zvol_last_close(zvol_state_t *zv)
zil_close(zv->zv_zilog); zil_close(zv->zv_zilog);
zv->zv_zilog = NULL; zv->zv_zilog = NULL;
dmu_buf_rele(zv->zv_dbuf, zvol_tag); dnode_rele(zv->zv_dn, zvol_tag);
zv->zv_dbuf = NULL; zv->zv_dn = NULL;
/* /*
* Evict cached data * Evict cached data
@ -993,8 +993,6 @@ static int
zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio) zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
{ {
zvol_state_t *zv = arg; zvol_state_t *zv = arg;
objset_t *os = zv->zv_objset;
uint64_t object = ZVOL_OBJ;
uint64_t offset = lr->lr_offset; uint64_t offset = lr->lr_offset;
uint64_t size = lr->lr_length; /* length of user data */ uint64_t size = lr->lr_length; /* length of user data */
dmu_buf_t *db; dmu_buf_t *db;
@ -1018,7 +1016,7 @@ zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
if (buf != NULL) { /* immediate write */ if (buf != NULL) { /* immediate write */
zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
RL_READER); RL_READER);
error = dmu_read(os, object, offset, size, buf, error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
DMU_READ_NO_PREFETCH); DMU_READ_NO_PREFETCH);
} else { /* indirect write */ } else { /* indirect write */
/* /*
@ -1031,7 +1029,7 @@ zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
offset = P2ALIGN(offset, size); offset = P2ALIGN(offset, size);
zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
RL_READER); RL_READER);
error = dmu_buf_hold(os, object, offset, zgd, &db, error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
DMU_READ_NO_PREFETCH); DMU_READ_NO_PREFETCH);
if (error == 0) { if (error == 0) {
blkptr_t *bp = &lr->lr_blkptr; blkptr_t *bp = &lr->lr_blkptr;
@ -1098,8 +1096,8 @@ zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
itx = zil_itx_create(TX_WRITE, sizeof (*lr) + itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
(wr_state == WR_COPIED ? len : 0)); (wr_state == WR_COPIED ? len : 0));
lr = (lr_write_t *)&itx->itx_lr; lr = (lr_write_t *)&itx->itx_lr;
if (wr_state == WR_COPIED && dmu_read(zv->zv_objset, if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
zil_itx_destroy(itx); zil_itx_destroy(itx);
itx = zil_itx_create(TX_WRITE, sizeof (*lr)); itx = zil_itx_create(TX_WRITE, sizeof (*lr));
lr = (lr_write_t *)&itx->itx_lr; lr = (lr_write_t *)&itx->itx_lr;
@ -1468,7 +1466,7 @@ zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
dmu_tx_abort(tx); dmu_tx_abort(tx);
break; break;
} }
error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); error = dmu_write_uio_dnode(zv->zv_dn, uio, bytes, tx);
if (error == 0) if (error == 0)
zvol_log_write(zv, tx, off, bytes, sync); zvol_log_write(zv, tx, off, bytes, sync);
dmu_tx_commit(tx); dmu_tx_commit(tx);
@ -1549,7 +1547,7 @@ zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
int int
zvol_get_volume_params(minor_t minor, uint64_t *blksize, zvol_get_volume_params(minor_t minor, uint64_t *blksize,
uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
void **rl_hdl, void **bonus_hdl) void **rl_hdl, void **dnode_hdl)
{ {
zvol_state_t *zv; zvol_state_t *zv;
@ -1560,7 +1558,7 @@ zvol_get_volume_params(minor_t minor, uint64_t *blksize,
return (SET_ERROR(ENXIO)); return (SET_ERROR(ENXIO));
ASSERT(blksize && max_xfer_len && minor_hdl && ASSERT(blksize && max_xfer_len && minor_hdl &&
objset_hdl && zil_hdl && rl_hdl && bonus_hdl); objset_hdl && zil_hdl && rl_hdl && dnode_hdl);
*blksize = zv->zv_volblocksize; *blksize = zv->zv_volblocksize;
*max_xfer_len = (uint64_t)zvol_maxphys; *max_xfer_len = (uint64_t)zvol_maxphys;
@ -1568,7 +1566,7 @@ zvol_get_volume_params(minor_t minor, uint64_t *blksize,
*objset_hdl = zv->zv_objset; *objset_hdl = zv->zv_objset;
*zil_hdl = zv->zv_zilog; *zil_hdl = zv->zv_zilog;
*rl_hdl = &zv->zv_znode; *rl_hdl = &zv->zv_znode;
*bonus_hdl = zv->zv_dbuf; *dnode_hdl = zv->zv_dn;
return (0); return (0);
} }