c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2026-05-22 02:27:36 +03:00
Code Issues Packages Projects Releases Wiki Activity

Illumos #3740

Browse Source 
3740 Poor ZFS send / receive performance due to snapshot
     hold / release processing
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Approved by: Christopher Siden <christopher.siden@delphix.com>

References:
  https://www.illumos.org/issues/3740
  illumos/illumos-gate@a7a845e4bf

Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1775

Porting notes:

1. 13fe019870 introduced a merge conflict
   in dsl_dataset_user_release_tmp where some variables were moved
   outside of the preprocessor directive.

2. dea9dfefdd747534b3846845629d2200f0616dad made the previous merge
   conflict worse by switching KM_SLEEP to KM_PUSHPAGE. This is notable
   because this commit refactors the code, adding a new KM_SLEEP
   allocation. It is not clear to me whether this should be converted
   to KM_PUSHPAGE.

3. We had a merge conflict in libzfs_sendrecv.c because of copyright
   notices.

4. Several small C99 compatibility fixed were made.
This commit is contained in:
Steven Hartland
2013-05-25 02:06:23 +00:00
committed by Brian Behlendorf
parent 7bc7f25040
commit 95fd54a1c5
13 changed files with 672 additions and 505 deletions
Download Patch File Download Diff File Expand all files Collapse all files
module/zfs/dsl_destroy.c
+5
View File
@@ -21,6 +21,7 @@
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2013 by Delphix. All rights reserved.
* Copyright (c) 2013 Steven Hartland. All rights reserved.
*/
#include <sys/zfs_context.h>
@@ -127,6 +128,10 @@ dsl_destroy_snapshot_check(void *arg, dmu_tx_t *tx)
pair = nvlist_next_nvpair(dsda->dsda_errlist, NULL);
if (pair != NULL)
return (fnvpair_value_int32(pair));
if (nvlist_empty(dsda->dsda_successful_snaps))
return (SET_ERROR(ENOENT));
return (0);
}
module/zfs/dsl_pool.c
+15 -3
View File
@@ -21,6 +21,7 @@
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2013 by Delphix. All rights reserved.
* Copyright (c) 2013 Steven Hartland. All rights reserved.
*/
#include <sys/dsl_pool.h>
@@ -840,23 +841,34 @@ dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp)
zap_cursor_t zc;
objset_t *mos = dp->dp_meta_objset;
uint64_t zapobj = dp->dp_tmp_userrefs_obj;
nvlist_t *holds;
if (zapobj == 0)
return;
ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
holds = fnvlist_alloc();
for (zap_cursor_init(&zc, mos, zapobj);
zap_cursor_retrieve(&zc, &za) == 0;
zap_cursor_advance(&zc)) {
char *htag;
uint64_t dsobj;
nvlist_t *tags;
htag = strchr(za.za_name, '-');
*htag = '\0';
++htag;
dsobj = strtonum(za.za_name, NULL);
dsl_dataset_user_release_tmp(dp, dsobj, htag);
if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) {
tags = fnvlist_alloc();
fnvlist_add_boolean(tags, htag);
fnvlist_add_nvlist(holds, za.za_name, tags);
fnvlist_free(tags);
} else {
fnvlist_add_boolean(tags, htag);
}
}
dsl_dataset_user_release_tmp(dp, holds);
fnvlist_free(holds);
zap_cursor_fini(&zc);
}
module/zfs/dsl_userhold.c
+485 -345
View File
@@ -21,6 +21,7 @@
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2013 by Delphix. All rights reserved.
* Copyright (c) 2013 Steven Hartland. All rights reserved.
*/
#include <sys/zfs_context.h>
@@ -37,6 +38,7 @@
typedef struct dsl_dataset_user_hold_arg {
nvlist_t *dduha_holds;
nvlist_t *dduha_chkholds;
nvlist_t *dduha_errlist;
minor_t dduha_minor;
} dsl_dataset_user_hold_arg_t;
@@ -53,25 +55,24 @@ dsl_dataset_user_hold_check_one(dsl_dataset_t *ds, const char *htag,
objset_t *mos = dp->dp_meta_objset;
int error = 0;
ASSERT(dsl_pool_config_held(dp));
if (strlen(htag) > MAXNAMELEN)
return (E2BIG);
return (SET_ERROR(E2BIG));
/* Tempholds have a more restricted length */
if (temphold && strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
return (E2BIG);
return (SET_ERROR(E2BIG));
/* tags must be unique (if ds already exists) */
if (ds != NULL) {
mutex_enter(&ds->ds_lock);
if (ds->ds_phys->ds_userrefs_obj != 0) {
uint64_t value;
error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj,
htag, 8, 1, &value);
if (error == 0)
error = SET_ERROR(EEXIST);
else if (error == ENOENT)
error = 0;
}
mutex_exit(&ds->ds_lock);
if (ds != NULL && ds->ds_phys->ds_userrefs_obj != 0) {
uint64_t value;
error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj,
htag, 8, 1, &value);
if (error == 0)
error = SET_ERROR(EEXIST);
else if (error == ENOENT)
error = 0;
}
return (error);
@@ -83,51 +84,67 @@ dsl_dataset_user_hold_check(void *arg, dmu_tx_t *tx)
dsl_dataset_user_hold_arg_t *dduha = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
nvpair_t *pair;
int rv = 0;
if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS)
return (SET_ERROR(ENOTSUP));
for (pair = nvlist_next_nvpair(dduha->dduha_holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) {
int error = 0;
if (!dmu_tx_is_syncing(tx))
return (0);
for (pair = nvlist_next_nvpair(dduha->dduha_holds, NULL);
pair != NULL; pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) {
dsl_dataset_t *ds;
char *htag;
int error = 0;
char *htag, *name;
/* must be a snapshot */
if (strchr(nvpair_name(pair), '@') == NULL)
name = nvpair_name(pair);
if (strchr(name, '@') == NULL)
error = SET_ERROR(EINVAL);
if (error == 0)
error = nvpair_value_string(pair, &htag);
if (error == 0) {
error = dsl_dataset_hold(dp,
nvpair_name(pair), FTAG, &ds);
}
if (error == 0)
error = dsl_dataset_hold(dp, name, FTAG, &ds);
if (error == 0) {
error = dsl_dataset_user_hold_check_one(ds, htag,
dduha->dduha_minor != 0, tx);
dsl_dataset_rele(ds, FTAG);
}
if (error != 0) {
rv = error;
fnvlist_add_int32(dduha->dduha_errlist,
nvpair_name(pair), error);
if (error == 0) {
fnvlist_add_string(dduha->dduha_chkholds, name, htag);
} else {
/*
* We register ENOENT errors so they can be correctly
* reported if needed, such as when all holds fail.
*/
fnvlist_add_int32(dduha->dduha_errlist, name, error);
if (error != ENOENT)
return (error);
}
}
return (rv);
/* Return ENOENT if no holds would be created. */
if (nvlist_empty(dduha->dduha_chkholds))
return (SET_ERROR(ENOENT));
return (0);
}
void
dsl_dataset_user_hold_sync_one(dsl_dataset_t *ds, const char *htag,
minor_t minor, uint64_t now, dmu_tx_t *tx)
static void
dsl_dataset_user_hold_sync_one_impl(nvlist_t *tmpholds, dsl_dataset_t *ds,
const char *htag, minor_t minor, uint64_t now, dmu_tx_t *tx)
{
dsl_pool_t *dp = ds->ds_dir->dd_pool;
objset_t *mos = dp->dp_meta_objset;
uint64_t zapobj;
mutex_enter(&ds->ds_lock);
ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
if (ds->ds_phys->ds_userrefs_obj == 0) {
/*
* This is the first user hold for this dataset. Create
@@ -140,14 +157,26 @@ dsl_dataset_user_hold_sync_one(dsl_dataset_t *ds, const char *htag,
zapobj = ds->ds_phys->ds_userrefs_obj;
}
ds->ds_userrefs++;
mutex_exit(&ds->ds_lock);
VERIFY0(zap_add(mos, zapobj, htag, 8, 1, &now, tx));
if (minor != 0) {
char name[MAXNAMELEN];
nvlist_t *tags;
VERIFY0(dsl_pool_user_hold(dp, ds->ds_object,
htag, now, tx));
dsl_register_onexit_hold_cleanup(ds, htag, minor);
(void) snprintf(name, sizeof (name), "%llx",
(u_longlong_t)ds->ds_object);
if (nvlist_lookup_nvlist(tmpholds, name, &tags) != 0) {
tags = fnvlist_alloc();
fnvlist_add_boolean(tags, htag);
fnvlist_add_nvlist(tmpholds, name, tags);
fnvlist_free(tags);
} else {
fnvlist_add_boolean(tags, htag);
}
}
spa_history_log_internal_ds(ds, "hold", tx,
@@ -155,306 +184,10 @@ dsl_dataset_user_hold_sync_one(dsl_dataset_t *ds, const char *htag,
htag, minor != 0, ds->ds_userrefs);
}
static void
dsl_dataset_user_hold_sync(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_hold_arg_t *dduha = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
nvpair_t *pair;
uint64_t now = gethrestime_sec();
for (pair = nvlist_next_nvpair(dduha->dduha_holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) {
dsl_dataset_t *ds;
VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
dsl_dataset_user_hold_sync_one(ds, fnvpair_value_string(pair),
dduha->dduha_minor, now, tx);
dsl_dataset_rele(ds, FTAG);
}
}
/*
* holds is nvl of snapname -> holdname
* errlist will be filled in with snapname -> error
* if cleanup_minor is not 0, the holds will be temporary, cleaned up
* when the process exits.
*
* if any fails, all will fail.
*/
int
dsl_dataset_user_hold(nvlist_t *holds, minor_t cleanup_minor, nvlist_t *errlist)
{
dsl_dataset_user_hold_arg_t dduha;
nvpair_t *pair;
pair = nvlist_next_nvpair(holds, NULL);
if (pair == NULL)
return (0);
dduha.dduha_holds = holds;
dduha.dduha_errlist = errlist;
dduha.dduha_minor = cleanup_minor;
return (dsl_sync_task(nvpair_name(pair), dsl_dataset_user_hold_check,
dsl_dataset_user_hold_sync, &dduha, fnvlist_num_pairs(holds)));
}
typedef struct dsl_dataset_user_release_arg {
nvlist_t *ddura_holds;
nvlist_t *ddura_todelete;
nvlist_t *ddura_errlist;
} dsl_dataset_user_release_arg_t;
static int
dsl_dataset_user_release_check_one(dsl_dataset_t *ds,
nvlist_t *holds, boolean_t *todelete)
{
uint64_t zapobj;
nvpair_t *pair;
objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
int error;
int numholds = 0;
*todelete = B_FALSE;
if (!dsl_dataset_is_snapshot(ds))
return (SET_ERROR(EINVAL));
zapobj = ds->ds_phys->ds_userrefs_obj;
if (zapobj == 0)
return (SET_ERROR(ESRCH));
for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(holds, pair)) {
/* Make sure the hold exists */
uint64_t tmp;
error = zap_lookup(mos, zapobj, nvpair_name(pair), 8, 1, &tmp);
if (error == ENOENT)
error = SET_ERROR(ESRCH);
if (error != 0)
return (error);
numholds++;
}
if (DS_IS_DEFER_DESTROY(ds) && ds->ds_phys->ds_num_children == 1 &&
ds->ds_userrefs == numholds) {
/* we need to destroy the snapshot as well */
if (dsl_dataset_long_held(ds))
return (SET_ERROR(EBUSY));
*todelete = B_TRUE;
}
return (0);
}
static int
dsl_dataset_user_release_check(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_release_arg_t *ddura = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
nvpair_t *pair;
int rv = 0;
if (!dmu_tx_is_syncing(tx))
return (0);
for (pair = nvlist_next_nvpair(ddura->ddura_holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) {
const char *name = nvpair_name(pair);
int error;
dsl_dataset_t *ds;
nvlist_t *holds;
error = nvpair_value_nvlist(pair, &holds);
if (error != 0)
return (SET_ERROR(EINVAL));
error = dsl_dataset_hold(dp, name, FTAG, &ds);
if (error == 0) {
boolean_t deleteme;
error = dsl_dataset_user_release_check_one(ds,
holds, &deleteme);
if (error == 0 && deleteme) {
fnvlist_add_boolean(ddura->ddura_todelete,
name);
}
dsl_dataset_rele(ds, FTAG);
}
if (error != 0) {
if (ddura->ddura_errlist != NULL) {
fnvlist_add_int32(ddura->ddura_errlist,
name, error);
}
rv = error;
}
}
return (rv);
}
static void
dsl_dataset_user_release_sync_one(dsl_dataset_t *ds, nvlist_t *holds,
dmu_tx_t *tx)
{
dsl_pool_t *dp = ds->ds_dir->dd_pool;
objset_t *mos = dp->dp_meta_objset;
uint64_t zapobj;
int error;
nvpair_t *pair;
for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(holds, pair)) {
ds->ds_userrefs--;
error = dsl_pool_user_release(dp, ds->ds_object,
nvpair_name(pair), tx);
VERIFY(error == 0 || error == ENOENT);
zapobj = ds->ds_phys->ds_userrefs_obj;
VERIFY0(zap_remove(mos, zapobj, nvpair_name(pair), tx));
spa_history_log_internal_ds(ds, "release", tx,
"tag=%s refs=%lld", nvpair_name(pair),
(longlong_t)ds->ds_userrefs);
}
}
static void
dsl_dataset_user_release_sync(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_release_arg_t *ddura = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
nvpair_t *pair;
for (pair = nvlist_next_nvpair(ddura->ddura_holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) {
dsl_dataset_t *ds;
VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
dsl_dataset_user_release_sync_one(ds,
fnvpair_value_nvlist(pair), tx);
if (nvlist_exists(ddura->ddura_todelete,
nvpair_name(pair))) {
ASSERT(ds->ds_userrefs == 0 &&
ds->ds_phys->ds_num_children == 1 &&
DS_IS_DEFER_DESTROY(ds));
dsl_destroy_snapshot_sync_impl(ds, B_FALSE, tx);
}
dsl_dataset_rele(ds, FTAG);
}
}
/*
* holds is nvl of snapname -> { holdname, ... }
* errlist will be filled in with snapname -> error
*
* if any fails, all will fail.
*/
int
dsl_dataset_user_release(nvlist_t *holds, nvlist_t *errlist)
{
dsl_dataset_user_release_arg_t ddura;
nvpair_t *pair;
int error;
pair = nvlist_next_nvpair(holds, NULL);
if (pair == NULL)
return (0);
ddura.ddura_holds = holds;
ddura.ddura_errlist = errlist;
ddura.ddura_todelete = fnvlist_alloc();
error = dsl_sync_task(nvpair_name(pair), dsl_dataset_user_release_check,
dsl_dataset_user_release_sync, &ddura, fnvlist_num_pairs(holds));
fnvlist_free(ddura.ddura_todelete);
return (error);
}
typedef struct dsl_dataset_user_release_tmp_arg {
uint64_t ddurta_dsobj;
nvlist_t *ddurta_holds;
boolean_t ddurta_deleteme;
} dsl_dataset_user_release_tmp_arg_t;
static int
dsl_dataset_user_release_tmp_check(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_release_tmp_arg_t *ddurta = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dataset_t *ds;
int error;
if (!dmu_tx_is_syncing(tx))
return (0);
error = dsl_dataset_hold_obj(dp, ddurta->ddurta_dsobj, FTAG, &ds);
if (error)
return (error);
error = dsl_dataset_user_release_check_one(ds,
ddurta->ddurta_holds, &ddurta->ddurta_deleteme);
dsl_dataset_rele(ds, FTAG);
return (error);
}
static void
dsl_dataset_user_release_tmp_sync(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_release_tmp_arg_t *ddurta = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dataset_t *ds;
VERIFY0(dsl_dataset_hold_obj(dp, ddurta->ddurta_dsobj, FTAG, &ds));
dsl_dataset_user_release_sync_one(ds, ddurta->ddurta_holds, tx);
if (ddurta->ddurta_deleteme) {
ASSERT(ds->ds_userrefs == 0 &&
ds->ds_phys->ds_num_children == 1 &&
DS_IS_DEFER_DESTROY(ds));
dsl_destroy_snapshot_sync_impl(ds, B_FALSE, tx);
}
dsl_dataset_rele(ds, FTAG);
}
/*
* Called at spa_load time to release a stale temporary user hold.
* Also called by the onexit code.
*/
void
dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, const char *htag)
{
dsl_dataset_user_release_tmp_arg_t ddurta;
#ifdef _KERNEL
dsl_dataset_t *ds;
int error;
/* Make sure it is not mounted. */
dsl_pool_config_enter(dp, FTAG);
error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
if (error == 0) {
char name[MAXNAMELEN];
dsl_dataset_name(ds, name);
dsl_dataset_rele(ds, FTAG);
dsl_pool_config_exit(dp, FTAG);
(void) zfs_unmount_snap(name);
} else {
dsl_pool_config_exit(dp, FTAG);
}
#endif
ddurta.ddurta_dsobj = dsobj;
ddurta.ddurta_holds = fnvlist_alloc();
fnvlist_add_boolean(ddurta.ddurta_holds, htag);
(void) dsl_sync_task(spa_name(dp->dp_spa),
dsl_dataset_user_release_tmp_check,
dsl_dataset_user_release_tmp_sync, &ddurta, 1);
fnvlist_free(ddurta.ddurta_holds);
}
typedef struct zfs_hold_cleanup_arg {
char zhca_spaname[MAXNAMELEN];
uint64_t zhca_spa_load_guid;
uint64_t zhca_dsobj;
char zhca_htag[MAXNAMELEN];
nvlist_t *zhca_holds;
} zfs_hold_cleanup_arg_t;
static void
@@ -466,40 +199,447 @@ dsl_dataset_user_release_onexit(void *arg)
error = spa_open(ca->zhca_spaname, &spa, FTAG);
if (error != 0) {
zfs_dbgmsg("couldn't release hold on pool=%s ds=%llu tag=%s "
zfs_dbgmsg("couldn't release holds on pool=%s "
"because pool is no longer loaded",
ca->zhca_spaname, ca->zhca_dsobj, ca->zhca_htag);
ca->zhca_spaname);
return;
}
if (spa_load_guid(spa) != ca->zhca_spa_load_guid) {
zfs_dbgmsg("couldn't release hold on pool=%s ds=%llu tag=%s "
zfs_dbgmsg("couldn't release holds on pool=%s "
"because pool is no longer loaded (guid doesn't match)",
ca->zhca_spaname, ca->zhca_dsobj, ca->zhca_htag);
ca->zhca_spaname);
spa_close(spa, FTAG);
return;
}
dsl_dataset_user_release_tmp(spa_get_dsl(spa),
ca->zhca_dsobj, ca->zhca_htag);
(void) dsl_dataset_user_release_tmp(spa_get_dsl(spa), ca->zhca_holds);
fnvlist_free(ca->zhca_holds);
kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
spa_close(spa, FTAG);
}
void
dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
minor_t minor)
static void
dsl_onexit_hold_cleanup(spa_t *spa, nvlist_t *holds, minor_t minor)
{
zfs_hold_cleanup_arg_t *ca = kmem_alloc(sizeof (*ca), KM_PUSHPAGE);
spa_t *spa = dsl_dataset_get_spa(ds);
zfs_hold_cleanup_arg_t *ca;
if (minor == 0 || nvlist_empty(holds)) {
fnvlist_free(holds);
return;
}
ASSERT(spa != NULL);
ca = kmem_alloc(sizeof (*ca), KM_SLEEP);
(void) strlcpy(ca->zhca_spaname, spa_name(spa),
sizeof (ca->zhca_spaname));
ca->zhca_spa_load_guid = spa_load_guid(spa);
ca->zhca_dsobj = ds->ds_object;
(void) strlcpy(ca->zhca_htag, htag, sizeof (ca->zhca_htag));
ca->zhca_holds = holds;
VERIFY0(zfs_onexit_add_cb(minor,
dsl_dataset_user_release_onexit, ca, NULL));
}
void
dsl_dataset_user_hold_sync_one(dsl_dataset_t *ds, const char *htag,
minor_t minor, uint64_t now, dmu_tx_t *tx)
{
nvlist_t *tmpholds;
if (minor != 0)
tmpholds = fnvlist_alloc();
else
tmpholds = NULL;
dsl_dataset_user_hold_sync_one_impl(tmpholds, ds, htag, minor, now, tx);
dsl_onexit_hold_cleanup(dsl_dataset_get_spa(ds), tmpholds, minor);
}
static void
dsl_dataset_user_hold_sync(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_hold_arg_t *dduha = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
nvlist_t *tmpholds;
nvpair_t *pair;
uint64_t now = gethrestime_sec();
if (dduha->dduha_minor != 0)
tmpholds = fnvlist_alloc();
else
tmpholds = NULL;
for (pair = nvlist_next_nvpair(dduha->dduha_chkholds, NULL);
pair != NULL;
pair = nvlist_next_nvpair(dduha->dduha_chkholds, pair)) {
dsl_dataset_t *ds;
VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
dsl_dataset_user_hold_sync_one_impl(tmpholds, ds,
fnvpair_value_string(pair), dduha->dduha_minor, now, tx);
dsl_dataset_rele(ds, FTAG);
}
dsl_onexit_hold_cleanup(dp->dp_spa, tmpholds, dduha->dduha_minor);
}
/*
* The full semantics of this function are described in the comment above
* lzc_hold().
*
* To summarize:
* holds is nvl of snapname -> holdname
* errlist will be filled in with snapname -> error
*
* The snaphosts must all be in the same pool.
*
* Holds for snapshots that don't exist will be skipped.
*
* If none of the snapshots for requested holds exist then ENOENT will be
* returned.
*
* If cleanup_minor is not 0, the holds will be temporary, which will be cleaned
* up when the process exits.
*
* On success all the holds, for snapshots that existed, will be created and 0
* will be returned.
*
* On failure no holds will be created, the errlist will be filled in,
* and an errno will returned.
*
* In all cases the errlist will contain entries for holds where the snapshot
* didn't exist.
*/
int
dsl_dataset_user_hold(nvlist_t *holds, minor_t cleanup_minor, nvlist_t *errlist)
{
dsl_dataset_user_hold_arg_t dduha;
nvpair_t *pair;
int ret;
pair = nvlist_next_nvpair(holds, NULL);
if (pair == NULL)
return (0);
dduha.dduha_holds = holds;
dduha.dduha_chkholds = fnvlist_alloc();
dduha.dduha_errlist = errlist;
dduha.dduha_minor = cleanup_minor;
ret = dsl_sync_task(nvpair_name(pair), dsl_dataset_user_hold_check,
dsl_dataset_user_hold_sync, &dduha, fnvlist_num_pairs(holds));
fnvlist_free(dduha.dduha_chkholds);
return (ret);
}
typedef int (dsl_holdfunc_t)(dsl_pool_t *dp, const char *name, void *tag,
dsl_dataset_t **dsp);
typedef struct dsl_dataset_user_release_arg {
dsl_holdfunc_t *ddura_holdfunc;
nvlist_t *ddura_holds;
nvlist_t *ddura_todelete;
nvlist_t *ddura_errlist;
nvlist_t *ddura_chkholds;
} dsl_dataset_user_release_arg_t;
/* Place a dataset hold on the snapshot identified by passed dsobj string */
static int
dsl_dataset_hold_obj_string(dsl_pool_t *dp, const char *dsobj, void *tag,
dsl_dataset_t **dsp)
{
return (dsl_dataset_hold_obj(dp, strtonum(dsobj, NULL), tag, dsp));
}
static int
dsl_dataset_user_release_check_one(dsl_dataset_user_release_arg_t *ddura,
dsl_dataset_t *ds, nvlist_t *holds, const char *snapname)
{
uint64_t zapobj;
nvlist_t *holds_found;
nvpair_t *pair;
objset_t *mos;
int numholds;
if (!dsl_dataset_is_snapshot(ds))
return (SET_ERROR(EINVAL));
if (nvlist_empty(holds))
return (0);
numholds = 0;
mos = ds->ds_dir->dd_pool->dp_meta_objset;
zapobj = ds->ds_phys->ds_userrefs_obj;
holds_found = fnvlist_alloc();
for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(holds, pair)) {
uint64_t tmp;
int error;
const char *holdname = nvpair_name(pair);
if (zapobj != 0)
error = zap_lookup(mos, zapobj, holdname, 8, 1, &tmp);
else
error = SET_ERROR(ENOENT);
/*
* Non-existent holds are put on the errlist, but don't
* cause an overall failure.
*/
if (error == ENOENT) {
if (ddura->ddura_errlist != NULL) {
char *errtag = kmem_asprintf("%s#%s",
snapname, holdname);
fnvlist_add_int32(ddura->ddura_errlist, errtag,
ENOENT);
strfree(errtag);
}
continue;
}
if (error != 0) {
fnvlist_free(holds_found);
return (error);
}
fnvlist_add_boolean(holds_found, holdname);
numholds++;
}
if (DS_IS_DEFER_DESTROY(ds) && ds->ds_phys->ds_num_children == 1 &&
ds->ds_userrefs == numholds) {
/* we need to destroy the snapshot as well */
if (dsl_dataset_long_held(ds)) {
fnvlist_free(holds_found);
return (SET_ERROR(EBUSY));
}
fnvlist_add_boolean(ddura->ddura_todelete, snapname);
}
if (numholds != 0) {
fnvlist_add_nvlist(ddura->ddura_chkholds, snapname,
holds_found);
}
fnvlist_free(holds_found);
return (0);
}
static int
dsl_dataset_user_release_check(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_release_arg_t *ddura;
dsl_holdfunc_t *holdfunc;
dsl_pool_t *dp;
nvpair_t *pair;
if (!dmu_tx_is_syncing(tx))
return (0);
dp = dmu_tx_pool(tx);
ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
ddura = arg;
holdfunc = ddura->ddura_holdfunc;
for (pair = nvlist_next_nvpair(ddura->ddura_holds, NULL);
pair != NULL; pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) {
int error;
dsl_dataset_t *ds;
nvlist_t *holds;
const char *snapname = nvpair_name(pair);
error = nvpair_value_nvlist(pair, &holds);
if (error != 0)
error = (SET_ERROR(EINVAL));
else
error = holdfunc(dp, snapname, FTAG, &ds);
if (error == 0) {
error = dsl_dataset_user_release_check_one(ddura, ds,
holds, snapname);
dsl_dataset_rele(ds, FTAG);
}
if (error != 0) {
if (ddura->ddura_errlist != NULL) {
fnvlist_add_int32(ddura->ddura_errlist,
snapname, error);
}
/*
* Non-existent snapshots are put on the errlist,
* but don't cause an overall failure.
*/
if (error != ENOENT)
return (error);
}
}
/* Return ENOENT if none of the holds existed. */
if (nvlist_empty(ddura->ddura_chkholds))
return (SET_ERROR(ENOENT));
return (0);
}
static void
dsl_dataset_user_release_sync_one(dsl_dataset_t *ds, nvlist_t *holds,
dmu_tx_t *tx)
{
dsl_pool_t *dp = ds->ds_dir->dd_pool;
objset_t *mos = dp->dp_meta_objset;
nvpair_t *pair;
for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(holds, pair)) {
int error;
const char *holdname = nvpair_name(pair);
/* Remove temporary hold if one exists. */
error = dsl_pool_user_release(dp, ds->ds_object, holdname, tx);
VERIFY(error == 0 || error == ENOENT);
VERIFY0(zap_remove(mos, ds->ds_phys->ds_userrefs_obj, holdname,
tx));
ds->ds_userrefs--;
spa_history_log_internal_ds(ds, "release", tx,
"tag=%s refs=%lld", holdname, (longlong_t)ds->ds_userrefs);
}
}
static void
dsl_dataset_user_release_sync(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_release_arg_t *ddura = arg;
dsl_holdfunc_t *holdfunc = ddura->ddura_holdfunc;
dsl_pool_t *dp = dmu_tx_pool(tx);
nvpair_t *pair;
ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
for (pair = nvlist_next_nvpair(ddura->ddura_chkholds, NULL);
pair != NULL; pair = nvlist_next_nvpair(ddura->ddura_chkholds,
pair)) {
dsl_dataset_t *ds;
const char *name = nvpair_name(pair);
VERIFY0(holdfunc(dp, name, FTAG, &ds));
dsl_dataset_user_release_sync_one(ds,
fnvpair_value_nvlist(pair), tx);
if (nvlist_exists(ddura->ddura_todelete, name)) {
ASSERT(ds->ds_userrefs == 0 &&
ds->ds_phys->ds_num_children == 1 &&
DS_IS_DEFER_DESTROY(ds));
dsl_destroy_snapshot_sync_impl(ds, B_FALSE, tx);
}
dsl_dataset_rele(ds, FTAG);
}
}
/*
* The full semantics of this function are described in the comment above
* lzc_release().
*
* To summarize:
* Releases holds specified in the nvl holds.
*
* holds is nvl of snapname -> { holdname, ... }
* errlist will be filled in with snapname -> error
*
* If tmpdp is not NULL the names for holds should be the dsobj's of snapshots,
* otherwise they should be the names of shapshots.
*
* As a release may cause snapshots to be destroyed this trys to ensure they
* aren't mounted.
*
* The release of non-existent holds are skipped.
*
* At least one hold must have been released for the this function to succeed
* and return 0.
*/
static int
dsl_dataset_user_release_impl(nvlist_t *holds, nvlist_t *errlist,
dsl_pool_t *tmpdp)
{
dsl_dataset_user_release_arg_t ddura;
nvpair_t *pair;
char *pool;
int error;
pair = nvlist_next_nvpair(holds, NULL);
if (pair == NULL)
return (0);
/*
* The release may cause snapshots to be destroyed; make sure they
* are not mounted.
*/
if (tmpdp != NULL) {
/* Temporary holds are specified by dsobj string. */
ddura.ddura_holdfunc = dsl_dataset_hold_obj_string;
pool = spa_name(tmpdp->dp_spa);
#ifdef _KERNEL
dsl_pool_config_enter(tmpdp, FTAG);
for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(holds, pair)) {
dsl_dataset_t *ds;
error = dsl_dataset_hold_obj_string(tmpdp,
nvpair_name(pair), FTAG, &ds);
if (error == 0) {
char name[MAXNAMELEN];
dsl_dataset_name(ds, name);
dsl_dataset_rele(ds, FTAG);
(void) zfs_unmount_snap(name);
}
}
dsl_pool_config_exit(tmpdp, FTAG);
#endif
} else {
/* Non-temporary holds are specified by name. */
ddura.ddura_holdfunc = dsl_dataset_hold;
pool = nvpair_name(pair);
#ifdef _KERNEL
for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(holds, pair)) {
(void) zfs_unmount_snap(nvpair_name(pair));
}
#endif
}
ddura.ddura_holds = holds;
ddura.ddura_errlist = errlist;
ddura.ddura_todelete = fnvlist_alloc();
ddura.ddura_chkholds = fnvlist_alloc();
error = dsl_sync_task(pool, dsl_dataset_user_release_check,
dsl_dataset_user_release_sync, &ddura,
fnvlist_num_pairs(holds));
fnvlist_free(ddura.ddura_todelete);
fnvlist_free(ddura.ddura_chkholds);
return (error);
}
/*
* holds is nvl of snapname -> { holdname, ... }
* errlist will be filled in with snapname -> error
*/
int
dsl_dataset_user_release(nvlist_t *holds, nvlist_t *errlist)
{
return (dsl_dataset_user_release_impl(holds, errlist, NULL));
}
/*
* holds is nvl of snapdsobj -> { holdname, ... }
*/
void
dsl_dataset_user_release_tmp(struct dsl_pool *dp, nvlist_t *holds)
{
ASSERT(dp != NULL);
(void) dsl_dataset_user_release_impl(holds, NULL, dp);
}
int
dsl_dataset_get_holds(const char *dsname, nvlist_t *nvl)
{
module/zfs/zfs_ioctl.c
+1 -14
View File
@@ -28,6 +28,7 @@
* Copyright (c) 2012, Joyent, Inc. All rights reserved.
* Copyright (c) 201i3 by Delphix. All rights reserved.
* Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
* Copyright (c) 2013 Steven Hartland. All rights reserved.
*/
/*
@@ -4867,20 +4868,6 @@ zfs_ioc_get_holds(const char *snapname, nvlist_t *args, nvlist_t *outnvl)
static int
zfs_ioc_release(const char *pool, nvlist_t *holds, nvlist_t *errlist)
{
nvpair_t *pair;
int err;
/*
* The release may cause the snapshot to be destroyed; make sure it
* is not mounted.
*/
for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(holds, pair)) {
err = zfs_unmount_snap(nvpair_name(pair));
if (err != 0)
return (err);
}
return (dsl_dataset_user_release(holds, errlist));
}