freebsd-nq/module/zfs/dsl_dir.c

2021 lines
56 KiB
C
Raw Normal View History

2008-11-20 20:01:55 +00:00
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012, 2016 by Delphix. All rights reserved.
* Copyright (c) 2013 Martin Matuska. All rights reserved.
* Copyright (c) 2014 Joyent, Inc. All rights reserved.
* Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
* Copyright (c) 2016 Actifio, Inc. All rights reserved.
2008-11-20 20:01:55 +00:00
*/
#include <sys/dmu.h>
#include <sys/dmu_objset.h>
#include <sys/dmu_tx.h>
#include <sys/dsl_dataset.h>
#include <sys/dsl_dir.h>
#include <sys/dsl_prop.h>
#include <sys/dsl_synctask.h>
#include <sys/dsl_deleg.h>
#include <sys/dmu_impl.h>
2008-11-20 20:01:55 +00:00
#include <sys/spa.h>
#include <sys/metaslab.h>
2008-11-20 20:01:55 +00:00
#include <sys/zap.h>
#include <sys/zio.h>
#include <sys/arc.h>
#include <sys/sunddi.h>
#include <sys/zfeature.h>
#include <sys/policy.h>
#include <sys/zfs_znode.h>
Remove ZFC_IOC_*_MINOR ioctl()s Early versions of ZFS coordinated the creation and destruction of device minors from userspace. This was inherently racy and in late 2009 these ioctl()s were removed leaving everything up to the kernel. This significantly simplified the code. However, we never picked up these changes in ZoL since we'd already significantly adjusted this code for Linux. This patch aims to rectify that by finally removing ZFC_IOC_*_MINOR ioctl()s and moving all the functionality down in to the kernel. Since this cleanup will change the kernel/user ABI it's being done in the same tag as the previous libzfs_core ABI changes. This will minimize, but not eliminate, the disruption to end users. Once merged ZoL, Illumos, and FreeBSD will basically be back in sync in regards to handling ZVOLs in the common code. While each platform must have its own custom zvol.c implemenation the interfaces provided are consistent. NOTES: 1) This patch introduces one subtle change in behavior which could not be easily avoided. Prior to this change callers of 'zfs create -V ...' were guaranteed that upon exit the /dev/zvol/ block device link would be created or an error returned. That's no longer the case. The utilities will no longer block waiting for the symlink to be created. Callers are now responsible for blocking, this is why a 'udev_wait' call was added to the 'label' function in scripts/common.sh. 2) The read-only behavior of a ZVOL now solely depends on if the ZVOL_RDONLY bit is set in zv->zv_flags. The redundant policy setting in the gendisk structure was removed. This both simplifies the code and allows us to safely leverage set_disk_ro() to issue a KOBJ_CHANGE uevent. See the comment in the code for futher details on this. 3) Because __zvol_create_minor() and zvol_alloc() may now be called in a sync task they must use KM_PUSHPAGE. References: illumos/illumos-gate@681d9761e8516a7dc5ab6589e2dfe717777e1123 Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Tim Chase <tim@chase2k.com> Closes #1969
2013-12-06 22:20:22 +00:00
#include <sys/zvol.h>
2008-11-20 20:01:55 +00:00
#include "zfs_namecheck.h"
#include "zfs_prop.h"
/*
* Filesystem and Snapshot Limits
* ------------------------------
*
* These limits are used to restrict the number of filesystems and/or snapshots
* that can be created at a given level in the tree or below. A typical
* use-case is with a delegated dataset where the administrator wants to ensure
* that a user within the zone is not creating too many additional filesystems
* or snapshots, even though they're not exceeding their space quota.
*
* The filesystem and snapshot counts are stored as extensible properties. This
* capability is controlled by a feature flag and must be enabled to be used.
* Once enabled, the feature is not active until the first limit is set. At
* that point, future operations to create/destroy filesystems or snapshots
* will validate and update the counts.
*
* Because the count properties will not exist before the feature is active,
* the counts are updated when a limit is first set on an uninitialized
* dsl_dir node in the tree (The filesystem/snapshot count on a node includes
* all of the nested filesystems/snapshots. Thus, a new leaf node has a
* filesystem count of 0 and a snapshot count of 0. Non-existent filesystem and
* snapshot count properties on a node indicate uninitialized counts on that
* node.) When first setting a limit on an uninitialized node, the code starts
* at the filesystem with the new limit and descends into all sub-filesystems
* to add the count properties.
*
* In practice this is lightweight since a limit is typically set when the
* filesystem is created and thus has no children. Once valid, changing the
* limit value won't require a re-traversal since the counts are already valid.
* When recursively fixing the counts, if a node with a limit is encountered
* during the descent, the counts are known to be valid and there is no need to
* descend into that filesystem's children. The counts on filesystems above the
* one with the new limit will still be uninitialized, unless a limit is
* eventually set on one of those filesystems. The counts are always recursively
* updated when a limit is set on a dataset, unless there is already a limit.
* When a new limit value is set on a filesystem with an existing limit, it is
* possible for the new limit to be less than the current count at that level
* since a user who can change the limit is also allowed to exceed the limit.
*
* Once the feature is active, then whenever a filesystem or snapshot is
* created, the code recurses up the tree, validating the new count against the
* limit at each initialized level. In practice, most levels will not have a
* limit set. If there is a limit at any initialized level up the tree, the
* check must pass or the creation will fail. Likewise, when a filesystem or
* snapshot is destroyed, the counts are recursively adjusted all the way up
* the initizized nodes in the tree. Renaming a filesystem into different point
* in the tree will first validate, then update the counts on each branch up to
* the common ancestor. A receive will also validate the counts and then update
* them.
*
* An exception to the above behavior is that the limit is not enforced if the
* user has permission to modify the limit. This is primarily so that
* recursive snapshots in the global zone always work. We want to prevent a
* denial-of-service in which a lower level delegated dataset could max out its
* limit and thus block recursive snapshots from being taken in the global zone.
* Because of this, it is possible for the snapshot count to be over the limit
* and snapshots taken in the global zone could cause a lower level dataset to
* hit or exceed its limit. The administrator taking the global zone recursive
* snapshot should be aware of this side-effect and behave accordingly.
* For consistency, the filesystem limit is also not enforced if the user can
* modify the limit.
*
* The filesystem and snapshot limits are validated by dsl_fs_ss_limit_check()
* and updated by dsl_fs_ss_count_adjust(). A new limit value is setup in
* dsl_dir_activate_fs_ss_limit() and the counts are adjusted, if necessary, by
* dsl_dir_init_fs_ss_count().
*
* There is a special case when we receive a filesystem that already exists. In
* this case a temporary clone name of %X is created (see dmu_recv_begin). We
* never update the filesystem counts for temporary clones.
*
* Likewise, we do not update the snapshot counts for temporary snapshots,
* such as those created by zfs diff.
*/
2008-11-20 20:01:55 +00:00
extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
2008-11-20 20:01:55 +00:00
static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
static void
dsl_dir_evict_async(void *dbu)
2008-11-20 20:01:55 +00:00
{
dsl_dir_t *dd = dbu;
2008-11-20 20:01:55 +00:00
int t;
ASSERTV(dsl_pool_t *dp = dd->dd_pool);
2008-11-20 20:01:55 +00:00
dd->dd_dbuf = NULL;
2008-11-20 20:01:55 +00:00
for (t = 0; t < TXG_SIZE; t++) {
ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
ASSERT(dd->dd_tempreserved[t] == 0);
ASSERT(dd->dd_space_towrite[t] == 0);
}
if (dd->dd_parent)
dsl_dir_async_rele(dd->dd_parent, dd);
2008-11-20 20:01:55 +00:00
spa_async_close(dd->dd_pool->dp_spa, dd);
2008-11-20 20:01:55 +00:00
dsl_prop_fini(dd);
2008-11-20 20:01:55 +00:00
mutex_destroy(&dd->dd_lock);
kmem_free(dd, sizeof (dsl_dir_t));
}
int
dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
2008-11-20 20:01:55 +00:00
const char *tail, void *tag, dsl_dir_t **ddp)
{
dmu_buf_t *dbuf;
dsl_dir_t *dd;
int err;
ASSERT(dsl_pool_config_held(dp));
2008-11-20 20:01:55 +00:00
err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
if (err != 0)
2008-11-20 20:01:55 +00:00
return (err);
dd = dmu_buf_get_user(dbuf);
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(dbuf, &doi);
ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
2008-11-20 20:01:55 +00:00
}
#endif
if (dd == NULL) {
dsl_dir_t *winner;
dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
2008-11-20 20:01:55 +00:00
dd->dd_object = ddobj;
dd->dd_dbuf = dbuf;
dd->dd_pool = dp;
mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
dsl_prop_init(dd);
2008-11-20 20:01:55 +00:00
dsl_dir_snap_cmtime_update(dd);
if (dsl_dir_phys(dd)->dd_parent_obj) {
err = dsl_dir_hold_obj(dp,
dsl_dir_phys(dd)->dd_parent_obj, NULL, dd,
&dd->dd_parent);
if (err != 0)
goto errout;
2008-11-20 20:01:55 +00:00
if (tail) {
#ifdef ZFS_DEBUG
uint64_t foundobj;
err = zap_lookup(dp->dp_meta_objset,
dsl_dir_phys(dd->dd_parent)->
dd_child_dir_zapobj, tail,
sizeof (foundobj), 1, &foundobj);
2008-11-20 20:01:55 +00:00
ASSERT(err || foundobj == ddobj);
#endif
(void) strlcpy(dd->dd_myname, tail,
sizeof (dd->dd_myname));
2008-11-20 20:01:55 +00:00
} else {
err = zap_value_search(dp->dp_meta_objset,
dsl_dir_phys(dd->dd_parent)->
dd_child_dir_zapobj,
2008-11-20 20:01:55 +00:00
ddobj, 0, dd->dd_myname);
}
if (err != 0)
goto errout;
2008-11-20 20:01:55 +00:00
} else {
(void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
}
if (dsl_dir_is_clone(dd)) {
dmu_buf_t *origin_bonus;
dsl_dataset_phys_t *origin_phys;
/*
* We can't open the origin dataset, because
* that would require opening this dsl_dir.
* Just look at its phys directly instead.
*/
err = dmu_bonus_hold(dp->dp_meta_objset,
dsl_dir_phys(dd)->dd_origin_obj, FTAG,
&origin_bonus);
if (err != 0)
goto errout;
origin_phys = origin_bonus->db_data;
dd->dd_origin_txg =
origin_phys->ds_creation_txg;
dmu_buf_rele(origin_bonus, FTAG);
}
dmu_buf_init_user(&dd->dd_dbu, NULL, dsl_dir_evict_async,
&dd->dd_dbuf);
winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
if (winner != NULL) {
2008-11-20 20:01:55 +00:00
if (dd->dd_parent)
dsl_dir_rele(dd->dd_parent, dd);
dsl_prop_fini(dd);
2008-11-20 20:01:55 +00:00
mutex_destroy(&dd->dd_lock);
kmem_free(dd, sizeof (dsl_dir_t));
dd = winner;
} else {
spa_open_ref(dp->dp_spa, dd);
}
}
/*
* The dsl_dir_t has both open-to-close and instantiate-to-evict
* holds on the spa. We need the open-to-close holds because
* otherwise the spa_refcnt wouldn't change when we open a
* dir which the spa also has open, so we could incorrectly
* think it was OK to unload/export/destroy the pool. We need
* the instantiate-to-evict hold because the dsl_dir_t has a
* pointer to the dd_pool, which has a pointer to the spa_t.
*/
spa_open_ref(dp->dp_spa, tag);
ASSERT3P(dd->dd_pool, ==, dp);
ASSERT3U(dd->dd_object, ==, ddobj);
ASSERT3P(dd->dd_dbuf, ==, dbuf);
*ddp = dd;
return (0);
errout:
if (dd->dd_parent)
dsl_dir_rele(dd->dd_parent, dd);
dsl_prop_fini(dd);
mutex_destroy(&dd->dd_lock);
kmem_free(dd, sizeof (dsl_dir_t));
dmu_buf_rele(dbuf, tag);
return (err);
2008-11-20 20:01:55 +00:00
}
void
dsl_dir_rele(dsl_dir_t *dd, void *tag)
2008-11-20 20:01:55 +00:00
{
dprintf_dd(dd, "%s\n", "");
spa_close(dd->dd_pool->dp_spa, tag);
dmu_buf_rele(dd->dd_dbuf, tag);
}
/*
* Remove a reference to the given dsl dir that is being asynchronously
* released. Async releases occur from a taskq performing eviction of
* dsl datasets and dirs. This process is identical to a normal release
* with the exception of using the async API for releasing the reference on
* the spa.
*/
void
dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
{
dprintf_dd(dd, "%s\n", "");
spa_async_close(dd->dd_pool->dp_spa, tag);
dmu_buf_rele(dd->dd_dbuf, tag);
}
/* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes */
2008-11-20 20:01:55 +00:00
void
dsl_dir_name(dsl_dir_t *dd, char *buf)
{
if (dd->dd_parent) {
dsl_dir_name(dd->dd_parent, buf);
VERIFY3U(strlcat(buf, "/", ZFS_MAX_DATASET_NAME_LEN), <,
ZFS_MAX_DATASET_NAME_LEN);
2008-11-20 20:01:55 +00:00
} else {
buf[0] = '\0';
}
if (!MUTEX_HELD(&dd->dd_lock)) {
/*
* recursive mutex so that we can use
* dprintf_dd() with dd_lock held
*/
mutex_enter(&dd->dd_lock);
VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
<, ZFS_MAX_DATASET_NAME_LEN);
2008-11-20 20:01:55 +00:00
mutex_exit(&dd->dd_lock);
} else {
VERIFY3U(strlcat(buf, dd->dd_myname, ZFS_MAX_DATASET_NAME_LEN),
<, ZFS_MAX_DATASET_NAME_LEN);
2008-11-20 20:01:55 +00:00
}
}
/* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
2008-11-20 20:01:55 +00:00
int
dsl_dir_namelen(dsl_dir_t *dd)
{
int result = 0;
if (dd->dd_parent) {
/* parent's name + 1 for the "/" */
result = dsl_dir_namelen(dd->dd_parent) + 1;
}
if (!MUTEX_HELD(&dd->dd_lock)) {
/* see dsl_dir_name */
mutex_enter(&dd->dd_lock);
result += strlen(dd->dd_myname);
mutex_exit(&dd->dd_lock);
} else {
result += strlen(dd->dd_myname);
}
return (result);
}
static int
getcomponent(const char *path, char *component, const char **nextp)
{
char *p;
2009-07-02 22:44:48 +00:00
if ((path == NULL) || (path[0] == '\0'))
return (SET_ERROR(ENOENT));
2008-11-20 20:01:55 +00:00
/* This would be a good place to reserve some namespace... */
p = strpbrk(path, "/@");
if (p && (p[1] == '/' || p[1] == '@')) {
/* two separators in a row */
return (SET_ERROR(EINVAL));
2008-11-20 20:01:55 +00:00
}
if (p == NULL || p == path) {
/*
* if the first thing is an @ or /, it had better be an
* @ and it had better not have any more ats or slashes,
* and it had better have something after the @.
*/
if (p != NULL &&
(p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
return (SET_ERROR(EINVAL));
if (strlen(path) >= ZFS_MAX_DATASET_NAME_LEN)
return (SET_ERROR(ENAMETOOLONG));
2008-11-20 20:01:55 +00:00
(void) strcpy(component, path);
p = NULL;
} else if (p[0] == '/') {
if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
return (SET_ERROR(ENAMETOOLONG));
2008-11-20 20:01:55 +00:00
(void) strncpy(component, path, p - path);
component[p - path] = '\0';
2008-11-20 20:01:55 +00:00
p++;
} else if (p[0] == '@') {
/*
* if the next separator is an @, there better not be
* any more slashes.
*/
if (strchr(path, '/'))
return (SET_ERROR(EINVAL));
if (p - path >= ZFS_MAX_DATASET_NAME_LEN)
return (SET_ERROR(ENAMETOOLONG));
2008-11-20 20:01:55 +00:00
(void) strncpy(component, path, p - path);
component[p - path] = '\0';
2008-11-20 20:01:55 +00:00
} else {
panic("invalid p=%p", (void *)p);
2008-11-20 20:01:55 +00:00
}
*nextp = p;
return (0);
}
/*
* Return the dsl_dir_t, and possibly the last component which couldn't
* be found in *tail. The name must be in the specified dsl_pool_t. This
* thread must hold the dp_config_rwlock for the pool. Returns NULL if the
* path is bogus, or if tail==NULL and we couldn't parse the whole name.
* (*tail)[0] == '@' means that the last component is a snapshot.
2008-11-20 20:01:55 +00:00
*/
int
dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
2008-11-20 20:01:55 +00:00
dsl_dir_t **ddp, const char **tailp)
{
char *buf;
const char *spaname, *next, *nextnext = NULL;
2008-11-20 20:01:55 +00:00
int err;
dsl_dir_t *dd;
uint64_t ddobj;
buf = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
2008-11-20 20:01:55 +00:00
err = getcomponent(name, buf, &next);
if (err != 0)
goto error;
2008-11-20 20:01:55 +00:00
/* Make sure the name is in the specified pool. */
spaname = spa_name(dp->dp_spa);
if (strcmp(buf, spaname) != 0) {
err = SET_ERROR(EXDEV);
goto error;
2008-11-20 20:01:55 +00:00
}
ASSERT(dsl_pool_config_held(dp));
2008-11-20 20:01:55 +00:00
err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
if (err != 0) {
goto error;
2008-11-20 20:01:55 +00:00
}
while (next != NULL) {
dsl_dir_t *child_dd;
2008-11-20 20:01:55 +00:00
err = getcomponent(next, buf, &nextnext);
if (err != 0)
2008-11-20 20:01:55 +00:00
break;
ASSERT(next[0] != '\0');
if (next[0] == '@')
break;
dprintf("looking up %s in obj%lld\n",
buf, dsl_dir_phys(dd)->dd_child_dir_zapobj);
2008-11-20 20:01:55 +00:00
err = zap_lookup(dp->dp_meta_objset,
dsl_dir_phys(dd)->dd_child_dir_zapobj,
2008-11-20 20:01:55 +00:00
buf, sizeof (ddobj), 1, &ddobj);
if (err != 0) {
2008-11-20 20:01:55 +00:00
if (err == ENOENT)
err = 0;
break;
}
err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
if (err != 0)
2008-11-20 20:01:55 +00:00
break;
dsl_dir_rele(dd, tag);
dd = child_dd;
2008-11-20 20:01:55 +00:00
next = nextnext;
}
if (err != 0) {
dsl_dir_rele(dd, tag);
goto error;
2008-11-20 20:01:55 +00:00
}
/*
* It's an error if there's more than one component left, or
* tailp==NULL and there's any component left.
*/
if (next != NULL &&
(tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
/* bad path name */
dsl_dir_rele(dd, tag);
2008-11-20 20:01:55 +00:00
dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
err = SET_ERROR(ENOENT);
2008-11-20 20:01:55 +00:00
}
if (tailp != NULL)
2008-11-20 20:01:55 +00:00
*tailp = next;
*ddp = dd;
error:
kmem_free(buf, ZFS_MAX_DATASET_NAME_LEN);
2008-11-20 20:01:55 +00:00
return (err);
}
/*
* If the counts are already initialized for this filesystem and its
* descendants then do nothing, otherwise initialize the counts.
*
* The counts on this filesystem, and those below, may be uninitialized due to
* either the use of a pre-existing pool which did not support the
* filesystem/snapshot limit feature, or one in which the feature had not yet
* been enabled.
*
* Recursively descend the filesystem tree and update the filesystem/snapshot
* counts on each filesystem below, then update the cumulative count on the
* current filesystem. If the filesystem already has a count set on it,
* then we know that its counts, and the counts on the filesystems below it,
* are already correct, so we don't have to update this filesystem.
*/
static void
dsl_dir_init_fs_ss_count(dsl_dir_t *dd, dmu_tx_t *tx)
{
uint64_t my_fs_cnt = 0;
uint64_t my_ss_cnt = 0;
dsl_pool_t *dp = dd->dd_pool;
objset_t *os = dp->dp_meta_objset;
zap_cursor_t *zc;
zap_attribute_t *za;
dsl_dataset_t *ds;
ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT));
ASSERT(dsl_pool_config_held(dp));
ASSERT(dmu_tx_is_syncing(tx));
dsl_dir_zapify(dd, tx);
/*
* If the filesystem count has already been initialized then we
* don't need to recurse down any further.
*/
if (zap_contains(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT) == 0)
return;
zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
/* Iterate my child dirs */
for (zap_cursor_init(zc, os, dsl_dir_phys(dd)->dd_child_dir_zapobj);
zap_cursor_retrieve(zc, za) == 0; zap_cursor_advance(zc)) {
dsl_dir_t *chld_dd;
uint64_t count;
VERIFY0(dsl_dir_hold_obj(dp, za->za_first_integer, NULL, FTAG,
&chld_dd));
/*
* Ignore hidden ($FREE, $MOS & $ORIGIN) objsets and
* temporary datasets.
*/
if (chld_dd->dd_myname[0] == '$' ||
chld_dd->dd_myname[0] == '%') {
dsl_dir_rele(chld_dd, FTAG);
continue;
}
my_fs_cnt++; /* count this child */
dsl_dir_init_fs_ss_count(chld_dd, tx);
VERIFY0(zap_lookup(os, chld_dd->dd_object,
DD_FIELD_FILESYSTEM_COUNT, sizeof (count), 1, &count));
my_fs_cnt += count;
VERIFY0(zap_lookup(os, chld_dd->dd_object,
DD_FIELD_SNAPSHOT_COUNT, sizeof (count), 1, &count));
my_ss_cnt += count;
dsl_dir_rele(chld_dd, FTAG);
}
zap_cursor_fini(zc);
/* Count my snapshots (we counted children's snapshots above) */
VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds));
for (zap_cursor_init(zc, os, dsl_dataset_phys(ds)->ds_snapnames_zapobj);
zap_cursor_retrieve(zc, za) == 0;
zap_cursor_advance(zc)) {
/* Don't count temporary snapshots */
if (za->za_name[0] != '%')
my_ss_cnt++;
}
zap_cursor_fini(zc);
dsl_dataset_rele(ds, FTAG);
kmem_free(zc, sizeof (zap_cursor_t));
kmem_free(za, sizeof (zap_attribute_t));
/* we're in a sync task, update counts */
dmu_buf_will_dirty(dd->dd_dbuf, tx);
VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
sizeof (my_fs_cnt), 1, &my_fs_cnt, tx));
VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
sizeof (my_ss_cnt), 1, &my_ss_cnt, tx));
}
static int
dsl_dir_actv_fs_ss_limit_check(void *arg, dmu_tx_t *tx)
{
char *ddname = (char *)arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dataset_t *ds;
dsl_dir_t *dd;
int error;
error = dsl_dataset_hold(dp, ddname, FTAG, &ds);
if (error != 0)
return (error);
if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
dsl_dataset_rele(ds, FTAG);
return (SET_ERROR(ENOTSUP));
}
dd = ds->ds_dir;
if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT) &&
dsl_dir_is_zapified(dd) &&
zap_contains(dp->dp_meta_objset, dd->dd_object,
DD_FIELD_FILESYSTEM_COUNT) == 0) {
dsl_dataset_rele(ds, FTAG);
return (SET_ERROR(EALREADY));
}
dsl_dataset_rele(ds, FTAG);
return (0);
}
static void
dsl_dir_actv_fs_ss_limit_sync(void *arg, dmu_tx_t *tx)
{
char *ddname = (char *)arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dataset_t *ds;
spa_t *spa;
VERIFY0(dsl_dataset_hold(dp, ddname, FTAG, &ds));
spa = dsl_dataset_get_spa(ds);
if (!spa_feature_is_active(spa, SPA_FEATURE_FS_SS_LIMIT)) {
/*
* Since the feature was not active and we're now setting a
* limit, increment the feature-active counter so that the
* feature becomes active for the first time.
*
* We are already in a sync task so we can update the MOS.
*/
spa_feature_incr(spa, SPA_FEATURE_FS_SS_LIMIT, tx);
}
/*
* Since we are now setting a non-UINT64_MAX limit on the filesystem,
* we need to ensure the counts are correct. Descend down the tree from
* this point and update all of the counts to be accurate.
*/
dsl_dir_init_fs_ss_count(ds->ds_dir, tx);
dsl_dataset_rele(ds, FTAG);
}
/*
* Make sure the feature is enabled and activate it if necessary.
* Since we're setting a limit, ensure the on-disk counts are valid.
* This is only called by the ioctl path when setting a limit value.
*
* We do not need to validate the new limit, since users who can change the
* limit are also allowed to exceed the limit.
*/
int
dsl_dir_activate_fs_ss_limit(const char *ddname)
{
int error;
error = dsl_sync_task(ddname, dsl_dir_actv_fs_ss_limit_check,
dsl_dir_actv_fs_ss_limit_sync, (void *)ddname, 0,
ZFS_SPACE_CHECK_RESERVED);
if (error == EALREADY)
error = 0;
return (error);
}
/*
* Used to determine if the filesystem_limit or snapshot_limit should be
* enforced. We allow the limit to be exceeded if the user has permission to
* write the property value. We pass in the creds that we got in the open
* context since we will always be the GZ root in syncing context. We also have
* to handle the case where we are allowed to change the limit on the current
* dataset, but there may be another limit in the tree above.
*
* We can never modify these two properties within a non-global zone. In
* addition, the other checks are modeled on zfs_secpolicy_write_perms. We
* can't use that function since we are already holding the dp_config_rwlock.
* In addition, we already have the dd and dealing with snapshots is simplified
* in this code.
*/
typedef enum {
ENFORCE_ALWAYS,
ENFORCE_NEVER,
ENFORCE_ABOVE
} enforce_res_t;
static enforce_res_t
dsl_enforce_ds_ss_limits(dsl_dir_t *dd, zfs_prop_t prop, cred_t *cr)
{
enforce_res_t enforce = ENFORCE_ALWAYS;
uint64_t obj;
dsl_dataset_t *ds;
uint64_t zoned;
ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
prop == ZFS_PROP_SNAPSHOT_LIMIT);
#ifdef _KERNEL
if (crgetzoneid(cr) != GLOBAL_ZONEID)
return (ENFORCE_ALWAYS);
if (secpolicy_zfs(cr) == 0)
return (ENFORCE_NEVER);
#endif
if ((obj = dsl_dir_phys(dd)->dd_head_dataset_obj) == 0)
return (ENFORCE_ALWAYS);
ASSERT(dsl_pool_config_held(dd->dd_pool));
if (dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds) != 0)
return (ENFORCE_ALWAYS);
if (dsl_prop_get_ds(ds, "zoned", 8, 1, &zoned, NULL) || zoned) {
/* Only root can access zoned fs's from the GZ */
enforce = ENFORCE_ALWAYS;
} else {
if (dsl_deleg_access_impl(ds, zfs_prop_to_name(prop), cr) == 0)
enforce = ENFORCE_ABOVE;
}
dsl_dataset_rele(ds, FTAG);
return (enforce);
}
/*
* Check if adding additional child filesystem(s) would exceed any filesystem
* limits or adding additional snapshot(s) would exceed any snapshot limits.
* The prop argument indicates which limit to check.
*
* Note that all filesystem limits up to the root (or the highest
* initialized) filesystem or the given ancestor must be satisfied.
*/
int
dsl_fs_ss_limit_check(dsl_dir_t *dd, uint64_t delta, zfs_prop_t prop,
dsl_dir_t *ancestor, cred_t *cr)
{
objset_t *os = dd->dd_pool->dp_meta_objset;
uint64_t limit, count;
char *count_prop;
enforce_res_t enforce;
int err = 0;
ASSERT(dsl_pool_config_held(dd->dd_pool));
ASSERT(prop == ZFS_PROP_FILESYSTEM_LIMIT ||
prop == ZFS_PROP_SNAPSHOT_LIMIT);
/*
* If we're allowed to change the limit, don't enforce the limit
* e.g. this can happen if a snapshot is taken by an administrative
* user in the global zone (i.e. a recursive snapshot by root).
* However, we must handle the case of delegated permissions where we
* are allowed to change the limit on the current dataset, but there
* is another limit in the tree above.
*/
enforce = dsl_enforce_ds_ss_limits(dd, prop, cr);
if (enforce == ENFORCE_NEVER)
return (0);
/*
* e.g. if renaming a dataset with no snapshots, count adjustment
* is 0.
*/
if (delta == 0)
return (0);
if (prop == ZFS_PROP_SNAPSHOT_LIMIT) {
/*
* We don't enforce the limit for temporary snapshots. This is
* indicated by a NULL cred_t argument.
*/
if (cr == NULL)
return (0);
count_prop = DD_FIELD_SNAPSHOT_COUNT;
} else {
count_prop = DD_FIELD_FILESYSTEM_COUNT;
}
/*
* If an ancestor has been provided, stop checking the limit once we
* hit that dir. We need this during rename so that we don't overcount
* the check once we recurse up to the common ancestor.
*/
if (ancestor == dd)
return (0);
/*
* If we hit an uninitialized node while recursing up the tree, we can
* stop since we know there is no limit here (or above). The counts are
* not valid on this node and we know we won't touch this node's counts.
*/
if (!dsl_dir_is_zapified(dd) || zap_lookup(os, dd->dd_object,
count_prop, sizeof (count), 1, &count) == ENOENT)
return (0);
err = dsl_prop_get_dd(dd, zfs_prop_to_name(prop), 8, 1, &limit, NULL,
B_FALSE);
if (err != 0)
return (err);
/* Is there a limit which we've hit? */
if (enforce == ENFORCE_ALWAYS && (count + delta) > limit)
return (SET_ERROR(EDQUOT));
if (dd->dd_parent != NULL)
err = dsl_fs_ss_limit_check(dd->dd_parent, delta, prop,
ancestor, cr);
return (err);
}
/*
* Adjust the filesystem or snapshot count for the specified dsl_dir_t and all
* parents. When a new filesystem/snapshot is created, increment the count on
* all parents, and when a filesystem/snapshot is destroyed, decrement the
* count.
*/
void
dsl_fs_ss_count_adjust(dsl_dir_t *dd, int64_t delta, const char *prop,
dmu_tx_t *tx)
{
int err;
objset_t *os = dd->dd_pool->dp_meta_objset;
uint64_t count;
ASSERT(dsl_pool_config_held(dd->dd_pool));
ASSERT(dmu_tx_is_syncing(tx));
ASSERT(strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0 ||
strcmp(prop, DD_FIELD_SNAPSHOT_COUNT) == 0);
/*
* When we receive an incremental stream into a filesystem that already
* exists, a temporary clone is created. We don't count this temporary
* clone, whose name begins with a '%'. We also ignore hidden ($FREE,
* $MOS & $ORIGIN) objsets.
*/
if ((dd->dd_myname[0] == '%' || dd->dd_myname[0] == '$') &&
strcmp(prop, DD_FIELD_FILESYSTEM_COUNT) == 0)
return;
/*
* e.g. if renaming a dataset with no snapshots, count adjustment is 0
*/
if (delta == 0)
return;
/*
* If we hit an uninitialized node while recursing up the tree, we can
* stop since we know the counts are not valid on this node and we
* know we shouldn't touch this node's counts. An uninitialized count
* on the node indicates that either the feature has not yet been
* activated or there are no limits on this part of the tree.
*/
if (!dsl_dir_is_zapified(dd) || (err = zap_lookup(os, dd->dd_object,
prop, sizeof (count), 1, &count)) == ENOENT)
return;
VERIFY0(err);
count += delta;
/* Use a signed verify to make sure we're not neg. */
VERIFY3S(count, >=, 0);
VERIFY0(zap_update(os, dd->dd_object, prop, sizeof (count), 1, &count,
tx));
/* Roll up this additional count into our ancestors */
if (dd->dd_parent != NULL)
dsl_fs_ss_count_adjust(dd->dd_parent, delta, prop, tx);
}
2008-11-20 20:01:55 +00:00
uint64_t
dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
dmu_tx_t *tx)
2008-11-20 20:01:55 +00:00
{
objset_t *mos = dp->dp_meta_objset;
2008-11-20 20:01:55 +00:00
uint64_t ddobj;
dsl_dir_phys_t *ddphys;
2008-11-20 20:01:55 +00:00
dmu_buf_t *dbuf;
ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
if (pds) {
VERIFY(0 == zap_add(mos, dsl_dir_phys(pds)->dd_child_dir_zapobj,
name, sizeof (uint64_t), 1, &ddobj, tx));
} else {
/* it's the root dir */
VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
}
2008-11-20 20:01:55 +00:00
VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
dmu_buf_will_dirty(dbuf, tx);
ddphys = dbuf->db_data;
2008-11-20 20:01:55 +00:00
ddphys->dd_creation_time = gethrestime_sec();
if (pds) {
ddphys->dd_parent_obj = pds->dd_object;
/* update the filesystem counts */
dsl_fs_ss_count_adjust(pds, 1, DD_FIELD_FILESYSTEM_COUNT, tx);
}
ddphys->dd_props_zapobj = zap_create(mos,
2008-11-20 20:01:55 +00:00
DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
ddphys->dd_child_dir_zapobj = zap_create(mos,
2008-11-20 20:01:55 +00:00
DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
2008-11-20 20:01:55 +00:00
dmu_buf_rele(dbuf, FTAG);
return (ddobj);
}
boolean_t
dsl_dir_is_clone(dsl_dir_t *dd)
2008-11-20 20:01:55 +00:00
{
return (dsl_dir_phys(dd)->dd_origin_obj &&
(dd->dd_pool->dp_origin_snap == NULL ||
dsl_dir_phys(dd)->dd_origin_obj !=
dd->dd_pool->dp_origin_snap->ds_object));
2008-11-20 20:01:55 +00:00
}
void
dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
{
mutex_enter(&dd->dd_lock);
dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
dsl_dir_phys(dd)->dd_used_bytes);
dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA,
dsl_dir_phys(dd)->dd_quota);
2008-11-20 20:01:55 +00:00
dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
dsl_dir_phys(dd)->dd_reserved);
2008-11-20 20:01:55 +00:00
dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
dsl_dir_phys(dd)->dd_compressed_bytes == 0 ? 100 :
(dsl_dir_phys(dd)->dd_uncompressed_bytes * 100 /
dsl_dir_phys(dd)->dd_compressed_bytes));
dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALUSED,
dsl_dir_phys(dd)->dd_uncompressed_bytes);
if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]);
dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_HEAD]);
dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_REFRSRV]);
dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD] +
dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_CHILD_RSRV]);
}
2008-11-20 20:01:55 +00:00
mutex_exit(&dd->dd_lock);
if (dsl_dir_is_zapified(dd)) {
uint64_t count;
objset_t *os = dd->dd_pool->dp_meta_objset;
if (zap_lookup(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
sizeof (count), 1, &count) == 0) {
dsl_prop_nvlist_add_uint64(nv,
ZFS_PROP_FILESYSTEM_COUNT, count);
}
if (zap_lookup(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
sizeof (count), 1, &count) == 0) {
dsl_prop_nvlist_add_uint64(nv,
ZFS_PROP_SNAPSHOT_COUNT, count);
}
}
if (dsl_dir_is_clone(dd)) {
2008-11-20 20:01:55 +00:00
dsl_dataset_t *ds;
char buf[ZFS_MAX_DATASET_NAME_LEN];
2008-11-20 20:01:55 +00:00
VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
dsl_dir_phys(dd)->dd_origin_obj, FTAG, &ds));
2008-11-20 20:01:55 +00:00
dsl_dataset_name(ds, buf);
dsl_dataset_rele(ds, FTAG);
2008-11-20 20:01:55 +00:00
dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
}
}
void
dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
{
dsl_pool_t *dp = dd->dd_pool;
ASSERT(dsl_dir_phys(dd));
2008-11-20 20:01:55 +00:00
if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
2008-11-20 20:01:55 +00:00
/* up the hold count until we can be written out */
dmu_buf_add_ref(dd->dd_dbuf, dd);
}
}
static int64_t
parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
{
uint64_t old_accounted = MAX(used, dsl_dir_phys(dd)->dd_reserved);
uint64_t new_accounted =
MAX(used + delta, dsl_dir_phys(dd)->dd_reserved);
2008-11-20 20:01:55 +00:00
return (new_accounted - old_accounted);
}
void
dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
{
ASSERT(dmu_tx_is_syncing(tx));
mutex_enter(&dd->dd_lock);
ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
2008-11-20 20:01:55 +00:00
dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
mutex_exit(&dd->dd_lock);
/* release the hold from dsl_dir_dirty */
dmu_buf_rele(dd->dd_dbuf, dd);
}
static uint64_t
dsl_dir_space_towrite(dsl_dir_t *dd)
{
uint64_t space = 0;
ASSERT(MUTEX_HELD(&dd->dd_lock));
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
for (int i = 0; i < TXG_SIZE; i++) {
space += dd->dd_space_towrite[i & TXG_MASK];
ASSERT3U(dd->dd_space_towrite[i & TXG_MASK], >=, 0);
2008-11-20 20:01:55 +00:00
}
return (space);
}
/*
* How much space would dd have available if ancestor had delta applied
* to it? If ondiskonly is set, we're only interested in what's
* on-disk, not estimated pending changes.
*/
uint64_t
dsl_dir_space_available(dsl_dir_t *dd,
dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
{
uint64_t parentspace, myspace, quota, used;
/*
* If there are no restrictions otherwise, assume we have
* unlimited space available.
*/
quota = UINT64_MAX;
parentspace = UINT64_MAX;
if (dd->dd_parent != NULL) {
parentspace = dsl_dir_space_available(dd->dd_parent,
ancestor, delta, ondiskonly);
}
mutex_enter(&dd->dd_lock);
if (dsl_dir_phys(dd)->dd_quota != 0)
quota = dsl_dir_phys(dd)->dd_quota;
used = dsl_dir_phys(dd)->dd_used_bytes;
2008-11-20 20:01:55 +00:00
if (!ondiskonly)
used += dsl_dir_space_towrite(dd);
if (dd->dd_parent == NULL) {
uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
quota = MIN(quota, poolsize);
}
if (dsl_dir_phys(dd)->dd_reserved > used && parentspace != UINT64_MAX) {
2008-11-20 20:01:55 +00:00
/*
* We have some space reserved, in addition to what our
* parent gave us.
*/
parentspace += dsl_dir_phys(dd)->dd_reserved - used;
2008-11-20 20:01:55 +00:00
}
if (dd == ancestor) {
ASSERT(delta <= 0);
ASSERT(used >= -delta);
used += delta;
if (parentspace != UINT64_MAX)
parentspace -= delta;
}
2008-11-20 20:01:55 +00:00
if (used > quota) {
/* over quota */
myspace = 0;
} else {
/*
* the lesser of the space provided by our parent and
* the space left in our quota
*/
myspace = MIN(parentspace, quota - used);
}
mutex_exit(&dd->dd_lock);
return (myspace);
}
struct tempreserve {
list_node_t tr_node;
dsl_dir_t *tr_ds;
uint64_t tr_size;
};
static int
dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
boolean_t ignorequota, list_t *tr_list,
2008-11-20 20:01:55 +00:00
dmu_tx_t *tx, boolean_t first)
{
uint64_t txg;
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
uint64_t quota;
2008-11-20 20:01:55 +00:00
struct tempreserve *tr;
int retval;
uint64_t ref_rsrv;
top_of_function:
txg = tx->tx_txg;
retval = EDQUOT;
ref_rsrv = 0;
2008-11-20 20:01:55 +00:00
ASSERT3U(txg, !=, 0);
ASSERT3S(asize, >, 0);
mutex_enter(&dd->dd_lock);
/*
* Check against the dsl_dir's quota. We don't add in the delta
* when checking for over-quota because they get one free hit.
*/
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
uint64_t est_inflight = dsl_dir_space_towrite(dd);
for (int i = 0; i < TXG_SIZE; i++)
2008-11-20 20:01:55 +00:00
est_inflight += dd->dd_tempreserved[i];
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
uint64_t used_on_disk = dsl_dir_phys(dd)->dd_used_bytes;
2008-11-20 20:01:55 +00:00
/*
* On the first iteration, fetch the dataset's used-on-disk and
* refreservation values. Also, if checkrefquota is set, test if
* allocating this space would exceed the dataset's refquota.
*/
if (first && tx->tx_objset) {
int error;
dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;
2008-11-20 20:01:55 +00:00
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
error = dsl_dataset_check_quota(ds, !netfree,
2008-11-20 20:01:55 +00:00
asize, est_inflight, &used_on_disk, &ref_rsrv);
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
if (error != 0) {
2008-11-20 20:01:55 +00:00
mutex_exit(&dd->dd_lock);
DMU_TX_STAT_BUMP(dmu_tx_quota);
2008-11-20 20:01:55 +00:00
return (error);
}
}
/*
* If this transaction will result in a net free of space,
* we want to let it through.
*/
if (ignorequota || netfree || dsl_dir_phys(dd)->dd_quota == 0)
2008-11-20 20:01:55 +00:00
quota = UINT64_MAX;
else
quota = dsl_dir_phys(dd)->dd_quota;
2008-11-20 20:01:55 +00:00
/*
* Adjust the quota against the actual pool size at the root
* minus any outstanding deferred frees.
2008-11-20 20:01:55 +00:00
* To ensure that it's possible to remove files from a full
* pool without inducing transient overcommits, we throttle
* netfree transactions against a quota that is slightly larger,
* but still within the pool's allocation slop. In cases where
* we're very close to full, this will allow a steady trickle of
* removes to get through.
*/
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
uint64_t deferred = 0;
2008-11-20 20:01:55 +00:00
if (dd->dd_parent == NULL) {
spa_t *spa = dd->dd_pool->dp_spa;
2008-11-20 20:01:55 +00:00
uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
deferred = metaslab_class_get_deferred(spa_normal_class(spa));
if (poolsize - deferred < quota) {
quota = poolsize - deferred;
retval = ENOSPC;
2008-11-20 20:01:55 +00:00
}
}
/*
* If they are requesting more space, and our current estimate
* is over quota, they get to try again unless the actual
* on-disk is over quota and there are no pending changes (which
* may free up space for us).
*/
if (used_on_disk + est_inflight >= quota) {
if (est_inflight > 0 || used_on_disk < quota ||
(retval == ENOSPC && used_on_disk < quota + deferred))
retval = ERESTART;
2008-11-20 20:01:55 +00:00
dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
"quota=%lluK tr=%lluK err=%d\n",
used_on_disk>>10, est_inflight>>10,
quota>>10, asize>>10, retval);
2008-11-20 20:01:55 +00:00
mutex_exit(&dd->dd_lock);
DMU_TX_STAT_BUMP(dmu_tx_quota);
return (SET_ERROR(retval));
2008-11-20 20:01:55 +00:00
}
/* We need to up our estimated delta before dropping dd_lock */
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
dd->dd_tempreserved[txg & TXG_MASK] += asize;
2008-11-20 20:01:55 +00:00
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
uint64_t parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
2008-11-20 20:01:55 +00:00
asize - ref_rsrv);
mutex_exit(&dd->dd_lock);
tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
2008-11-20 20:01:55 +00:00
tr->tr_ds = dd;
tr->tr_size = asize;
list_insert_tail(tr_list, tr);
/* see if it's OK with our parent */
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
if (dd->dd_parent != NULL && parent_rsrv != 0) {
/*
* Recurse on our parent without recursion. This has been
* observed to be potentially large stack usage even within
* the test suite. Largest seen stack was 7632 bytes on linux.
*/
dd = dd->dd_parent;
asize = parent_rsrv;
ignorequota = (dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
first = B_FALSE;
goto top_of_function;
2008-11-20 20:01:55 +00:00
} else {
return (0);
}
}
/*
* Reserve space in this dsl_dir, to be used in this tx's txg.
* After the space has been dirtied (and dsl_dir_willuse_space()
* has been called), the reservation should be canceled, using
* dsl_dir_tempreserve_clear().
*/
int
dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
boolean_t netfree, void **tr_cookiep, dmu_tx_t *tx)
2008-11-20 20:01:55 +00:00
{
int err;
list_t *tr_list;
if (asize == 0) {
*tr_cookiep = NULL;
return (0);
}
tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
2008-11-20 20:01:55 +00:00
list_create(tr_list, sizeof (struct tempreserve),
offsetof(struct tempreserve, tr_node));
ASSERT3S(asize, >, 0);
err = arc_tempreserve_space(lsize, tx->tx_txg);
if (err == 0) {
struct tempreserve *tr;
tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
2008-11-20 20:01:55 +00:00
tr->tr_size = lsize;
list_insert_tail(tr_list, tr);
} else {
if (err == EAGAIN) {
Illumos #4045 write throttle & i/o scheduler performance work 4045 zfs write throttle & i/o scheduler performance work 1. The ZFS i/o scheduler (vdev_queue.c) now divides i/os into 5 classes: sync read, sync write, async read, async write, and scrub/resilver. The scheduler issues a number of concurrent i/os from each class to the device. Once a class has been selected, an i/o is selected from this class using either an elevator algorithem (async, scrub classes) or FIFO (sync classes). The number of concurrent async write i/os is tuned dynamically based on i/o load, to achieve good sync i/o latency when there is not a high load of writes, and good write throughput when there is. See the block comment in vdev_queue.c (reproduced below) for more details. 2. The write throttle (dsl_pool_tempreserve_space() and txg_constrain_throughput()) is rewritten to produce much more consistent delays when under constant load. The new write throttle is based on the amount of dirty data, rather than guesses about future performance of the system. When there is a lot of dirty data, each transaction (e.g. write() syscall) will be delayed by the same small amount. This eliminates the "brick wall of wait" that the old write throttle could hit, causing all transactions to wait several seconds until the next txg opens. One of the keys to the new write throttle is decrementing the amount of dirty data as i/o completes, rather than at the end of spa_sync(). Note that the write throttle is only applied once the i/o scheduler is issuing the maximum number of outstanding async writes. See the block comments in dsl_pool.c and above dmu_tx_delay() (reproduced below) for more details. This diff has several other effects, including: * the commonly-tuned global variable zfs_vdev_max_pending has been removed; use per-class zfs_vdev_*_max_active values or zfs_vdev_max_active instead. * the size of each txg (meaning the amount of dirty data written, and thus the time it takes to write out) is now controlled differently. There is no longer an explicit time goal; the primary determinant is amount of dirty data. Systems that are under light or medium load will now often see that a txg is always syncing, but the impact to performance (e.g. read latency) is minimal. Tune zfs_dirty_data_max and zfs_dirty_data_sync to control this. * zio_taskq_batch_pct = 75 -- Only use 75% of all CPUs for compression, checksum, etc. This improves latency by not allowing these CPU-intensive tasks to consume all CPU (on machines with at least 4 CPU's; the percentage is rounded up). --matt APPENDIX: problems with the current i/o scheduler The current ZFS i/o scheduler (vdev_queue.c) is deadline based. The problem with this is that if there are always i/os pending, then certain classes of i/os can see very long delays. For example, if there are always synchronous reads outstanding, then no async writes will be serviced until they become "past due". One symptom of this situation is that each pass of the txg sync takes at least several seconds (typically 3 seconds). If many i/os become "past due" (their deadline is in the past), then we must service all of these overdue i/os before any new i/os. This happens when we enqueue a batch of async writes for the txg sync, with deadlines 2.5 seconds in the future. If we can't complete all the i/os in 2.5 seconds (e.g. because there were always reads pending), then these i/os will become past due. Now we must service all the "async" writes (which could be hundreds of megabytes) before we service any reads, introducing considerable latency to synchronous i/os (reads or ZIL writes). Notes on porting to ZFS on Linux: - zio_t gained new members io_physdone and io_phys_children. Because object caches in the Linux port call the constructor only once at allocation time, objects may contain residual data when retrieved from the cache. Therefore zio_create() was updated to zero out the two new fields. - vdev_mirror_pending() relied on the depth of the per-vdev pending queue (vq->vq_pending_tree) to select the least-busy leaf vdev to read from. This tree has been replaced by vq->vq_active_tree which is now used for the same purpose. - vdev_queue_init() used the value of zfs_vdev_max_pending to determine the number of vdev I/O buffers to pre-allocate. That global no longer exists, so we instead use the sum of the *_max_active values for each of the five I/O classes described above. - The Illumos implementation of dmu_tx_delay() delays a transaction by sleeping in condition variable embedded in the thread (curthread->t_delay_cv). We do not have an equivalent CV to use in Linux, so this change replaced the delay logic with a wrapper called zfs_sleep_until(). This wrapper could be adopted upstream and in other downstream ports to abstract away operating system-specific delay logic. - These tunables are added as module parameters, and descriptions added to the zfs-module-parameters.5 man page. spa_asize_inflation zfs_deadman_synctime_ms zfs_vdev_max_active zfs_vdev_async_write_active_min_dirty_percent zfs_vdev_async_write_active_max_dirty_percent zfs_vdev_async_read_max_active zfs_vdev_async_read_min_active zfs_vdev_async_write_max_active zfs_vdev_async_write_min_active zfs_vdev_scrub_max_active zfs_vdev_scrub_min_active zfs_vdev_sync_read_max_active zfs_vdev_sync_read_min_active zfs_vdev_sync_write_max_active zfs_vdev_sync_write_min_active zfs_dirty_data_max_percent zfs_delay_min_dirty_percent zfs_dirty_data_max_max_percent zfs_dirty_data_max zfs_dirty_data_max_max zfs_dirty_data_sync zfs_delay_scale The latter four have type unsigned long, whereas they are uint64_t in Illumos. This accommodates Linux's module_param() supported types, but means they may overflow on 32-bit architectures. The values zfs_dirty_data_max and zfs_dirty_data_max_max are the most likely to overflow on 32-bit systems, since they express physical RAM sizes in bytes. In fact, Illumos initializes zfs_dirty_data_max_max to 2^32 which does overflow. To resolve that, this port instead initializes it in arc_init() to 25% of physical RAM, and adds the tunable zfs_dirty_data_max_max_percent to override that percentage. While this solution doesn't completely avoid the overflow issue, it should be a reasonable default for most systems, and the minority of affected systems can work around the issue by overriding the defaults. - Fixed reversed logic in comment above zfs_delay_scale declaration. - Clarified comments in vdev_queue.c regarding when per-queue minimums take effect. - Replaced dmu_tx_write_limit in the dmu_tx kstat file with dmu_tx_dirty_delay and dmu_tx_dirty_over_max. The first counts how many times a transaction has been delayed because the pool dirty data has exceeded zfs_delay_min_dirty_percent. The latter counts how many times the pool dirty data has exceeded zfs_dirty_data_max (which we expect to never happen). - The original patch would have regressed the bug fixed in zfsonlinux/zfs@c418410, which prevented users from setting the zfs_vdev_aggregation_limit tuning larger than SPA_MAXBLOCKSIZE. A similar fix is added to vdev_queue_aggregate(). - In vdev_queue_io_to_issue(), dynamically allocate 'zio_t search' on the heap instead of the stack. In Linux we can't afford such large structures on the stack. Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Ned Bass <bass6@llnl.gov> Reviewed by: Brendan Gregg <brendan.gregg@joyent.com> Approved by: Robert Mustacchi <rm@joyent.com> References: http://www.illumos.org/issues/4045 illumos/illumos-gate@69962b5647e4a8b9b14998733b765925381b727e Ported-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1913
2013-08-29 03:01:20 +00:00
/*
* If arc_memory_throttle() detected that pageout
* is running and we are low on memory, we delay new
* non-pageout transactions to give pageout an
* advantage.
*
* It is unfortunate to be delaying while the caller's
* locks are held.
*/
txg_delay(dd->dd_pool, tx->tx_txg,
MSEC2NSEC(10), MSEC2NSEC(10));
err = SET_ERROR(ERESTART);
2008-11-20 20:01:55 +00:00
}
}
if (err == 0) {
OpenZFS 7793 - ztest fails assertion in dmu_tx_willuse_space Reviewed by: Steve Gonczi <steve.gonczi@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Background information: This assertion about tx_space_* verifies that we are not dirtying more stuff than we thought we would. We “need” to know how much we will dirty so that we can check if we should fail this transaction with ENOSPC/EDQUOT, in dmu_tx_assign(). While the transaction is open (i.e. between dmu_tx_assign() and dmu_tx_commit() — typically less than a millisecond), we call dbuf_dirty() on the exact blocks that will be modified. Once this happens, the temporary accounting in tx_space_* is unnecessary, because we know exactly what blocks are newly dirtied; we call dnode_willuse_space() to track this more exact accounting. The fundamental problem causing this bug is that dmu_tx_hold_*() relies on the current state in the DMU (e.g. dn_nlevels) to predict how much will be dirtied by this transaction, but this state can change before we actually perform the transaction (i.e. call dbuf_dirty()). This bug will be fixed by removing the assertion that the tx_space_* accounting is perfectly accurate (i.e. we never dirty more than was predicted by dmu_tx_hold_*()). By removing the requirement that this accounting be perfectly accurate, we can also vastly simplify it, e.g. removing most of the logic in dmu_tx_count_*(). The new tx space accounting will be very approximate, and may be more or less than what is actually dirtied. It will still be used to determine if this transaction will put us over quota. Transactions that are marked by dmu_tx_mark_netfree() will be excepted from this check. We won’t make an attempt to determine how much space will be freed by the transaction — this was rarely accurate enough to determine if a transaction should be permitted when we are over quota, which is why dmu_tx_mark_netfree() was introduced in 2014. We also won’t attempt to give “credit” when overwriting existing blocks, if those blocks may be freed. This allows us to remove the do_free_accounting logic in dbuf_dirty(), and associated routines. This logic attempted to predict what will be on disk when this txg syncs, to know if the overwritten block will be freed (i.e. exists, and has no snapshots). OpenZFS-issue: https://www.illumos.org/issues/7793 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3704e0a Upstream bugs: DLPX-32883a Closes #5804 Porting notes: - DNODE_SIZE replaced with DNODE_MIN_SIZE in dmu_tx_count_dnode(), Using the default dnode size would be slightly better. - DEBUG_DMU_TX wrappers and configure option removed. - Resolved _by_dnode() conflicts these changes have not yet been applied to OpenZFS.
2017-03-07 17:51:59 +00:00
err = dsl_dir_tempreserve_impl(dd, asize, netfree,
B_FALSE, tr_list, tx, B_TRUE);
2008-11-20 20:01:55 +00:00
}
if (err != 0)
2008-11-20 20:01:55 +00:00
dsl_dir_tempreserve_clear(tr_list, tx);
else
*tr_cookiep = tr_list;
return (err);
}
/*
* Clear a temporary reservation that we previously made with
* dsl_dir_tempreserve_space().
*/
void
dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
{
int txgidx = tx->tx_txg & TXG_MASK;
list_t *tr_list = tr_cookie;
struct tempreserve *tr;
ASSERT3U(tx->tx_txg, !=, 0);
if (tr_cookie == NULL)
return;
Illumos #4045 write throttle & i/o scheduler performance work 4045 zfs write throttle & i/o scheduler performance work 1. The ZFS i/o scheduler (vdev_queue.c) now divides i/os into 5 classes: sync read, sync write, async read, async write, and scrub/resilver. The scheduler issues a number of concurrent i/os from each class to the device. Once a class has been selected, an i/o is selected from this class using either an elevator algorithem (async, scrub classes) or FIFO (sync classes). The number of concurrent async write i/os is tuned dynamically based on i/o load, to achieve good sync i/o latency when there is not a high load of writes, and good write throughput when there is. See the block comment in vdev_queue.c (reproduced below) for more details. 2. The write throttle (dsl_pool_tempreserve_space() and txg_constrain_throughput()) is rewritten to produce much more consistent delays when under constant load. The new write throttle is based on the amount of dirty data, rather than guesses about future performance of the system. When there is a lot of dirty data, each transaction (e.g. write() syscall) will be delayed by the same small amount. This eliminates the "brick wall of wait" that the old write throttle could hit, causing all transactions to wait several seconds until the next txg opens. One of the keys to the new write throttle is decrementing the amount of dirty data as i/o completes, rather than at the end of spa_sync(). Note that the write throttle is only applied once the i/o scheduler is issuing the maximum number of outstanding async writes. See the block comments in dsl_pool.c and above dmu_tx_delay() (reproduced below) for more details. This diff has several other effects, including: * the commonly-tuned global variable zfs_vdev_max_pending has been removed; use per-class zfs_vdev_*_max_active values or zfs_vdev_max_active instead. * the size of each txg (meaning the amount of dirty data written, and thus the time it takes to write out) is now controlled differently. There is no longer an explicit time goal; the primary determinant is amount of dirty data. Systems that are under light or medium load will now often see that a txg is always syncing, but the impact to performance (e.g. read latency) is minimal. Tune zfs_dirty_data_max and zfs_dirty_data_sync to control this. * zio_taskq_batch_pct = 75 -- Only use 75% of all CPUs for compression, checksum, etc. This improves latency by not allowing these CPU-intensive tasks to consume all CPU (on machines with at least 4 CPU's; the percentage is rounded up). --matt APPENDIX: problems with the current i/o scheduler The current ZFS i/o scheduler (vdev_queue.c) is deadline based. The problem with this is that if there are always i/os pending, then certain classes of i/os can see very long delays. For example, if there are always synchronous reads outstanding, then no async writes will be serviced until they become "past due". One symptom of this situation is that each pass of the txg sync takes at least several seconds (typically 3 seconds). If many i/os become "past due" (their deadline is in the past), then we must service all of these overdue i/os before any new i/os. This happens when we enqueue a batch of async writes for the txg sync, with deadlines 2.5 seconds in the future. If we can't complete all the i/os in 2.5 seconds (e.g. because there were always reads pending), then these i/os will become past due. Now we must service all the "async" writes (which could be hundreds of megabytes) before we service any reads, introducing considerable latency to synchronous i/os (reads or ZIL writes). Notes on porting to ZFS on Linux: - zio_t gained new members io_physdone and io_phys_children. Because object caches in the Linux port call the constructor only once at allocation time, objects may contain residual data when retrieved from the cache. Therefore zio_create() was updated to zero out the two new fields. - vdev_mirror_pending() relied on the depth of the per-vdev pending queue (vq->vq_pending_tree) to select the least-busy leaf vdev to read from. This tree has been replaced by vq->vq_active_tree which is now used for the same purpose. - vdev_queue_init() used the value of zfs_vdev_max_pending to determine the number of vdev I/O buffers to pre-allocate. That global no longer exists, so we instead use the sum of the *_max_active values for each of the five I/O classes described above. - The Illumos implementation of dmu_tx_delay() delays a transaction by sleeping in condition variable embedded in the thread (curthread->t_delay_cv). We do not have an equivalent CV to use in Linux, so this change replaced the delay logic with a wrapper called zfs_sleep_until(). This wrapper could be adopted upstream and in other downstream ports to abstract away operating system-specific delay logic. - These tunables are added as module parameters, and descriptions added to the zfs-module-parameters.5 man page. spa_asize_inflation zfs_deadman_synctime_ms zfs_vdev_max_active zfs_vdev_async_write_active_min_dirty_percent zfs_vdev_async_write_active_max_dirty_percent zfs_vdev_async_read_max_active zfs_vdev_async_read_min_active zfs_vdev_async_write_max_active zfs_vdev_async_write_min_active zfs_vdev_scrub_max_active zfs_vdev_scrub_min_active zfs_vdev_sync_read_max_active zfs_vdev_sync_read_min_active zfs_vdev_sync_write_max_active zfs_vdev_sync_write_min_active zfs_dirty_data_max_percent zfs_delay_min_dirty_percent zfs_dirty_data_max_max_percent zfs_dirty_data_max zfs_dirty_data_max_max zfs_dirty_data_sync zfs_delay_scale The latter four have type unsigned long, whereas they are uint64_t in Illumos. This accommodates Linux's module_param() supported types, but means they may overflow on 32-bit architectures. The values zfs_dirty_data_max and zfs_dirty_data_max_max are the most likely to overflow on 32-bit systems, since they express physical RAM sizes in bytes. In fact, Illumos initializes zfs_dirty_data_max_max to 2^32 which does overflow. To resolve that, this port instead initializes it in arc_init() to 25% of physical RAM, and adds the tunable zfs_dirty_data_max_max_percent to override that percentage. While this solution doesn't completely avoid the overflow issue, it should be a reasonable default for most systems, and the minority of affected systems can work around the issue by overriding the defaults. - Fixed reversed logic in comment above zfs_delay_scale declaration. - Clarified comments in vdev_queue.c regarding when per-queue minimums take effect. - Replaced dmu_tx_write_limit in the dmu_tx kstat file with dmu_tx_dirty_delay and dmu_tx_dirty_over_max. The first counts how many times a transaction has been delayed because the pool dirty data has exceeded zfs_delay_min_dirty_percent. The latter counts how many times the pool dirty data has exceeded zfs_dirty_data_max (which we expect to never happen). - The original patch would have regressed the bug fixed in zfsonlinux/zfs@c418410, which prevented users from setting the zfs_vdev_aggregation_limit tuning larger than SPA_MAXBLOCKSIZE. A similar fix is added to vdev_queue_aggregate(). - In vdev_queue_io_to_issue(), dynamically allocate 'zio_t search' on the heap instead of the stack. In Linux we can't afford such large structures on the stack. Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Ned Bass <bass6@llnl.gov> Reviewed by: Brendan Gregg <brendan.gregg@joyent.com> Approved by: Robert Mustacchi <rm@joyent.com> References: http://www.illumos.org/issues/4045 illumos/illumos-gate@69962b5647e4a8b9b14998733b765925381b727e Ported-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1913
2013-08-29 03:01:20 +00:00
while ((tr = list_head(tr_list)) != NULL) {
if (tr->tr_ds) {
2008-11-20 20:01:55 +00:00
mutex_enter(&tr->tr_ds->dd_lock);
ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
tr->tr_size);
tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
mutex_exit(&tr->tr_ds->dd_lock);
} else {
arc_tempreserve_clear(tr->tr_size);
}
list_remove(tr_list, tr);
kmem_free(tr, sizeof (struct tempreserve));
}
kmem_free(tr_list, sizeof (list_t));
}
Illumos #4045 write throttle & i/o scheduler performance work 4045 zfs write throttle & i/o scheduler performance work 1. The ZFS i/o scheduler (vdev_queue.c) now divides i/os into 5 classes: sync read, sync write, async read, async write, and scrub/resilver. The scheduler issues a number of concurrent i/os from each class to the device. Once a class has been selected, an i/o is selected from this class using either an elevator algorithem (async, scrub classes) or FIFO (sync classes). The number of concurrent async write i/os is tuned dynamically based on i/o load, to achieve good sync i/o latency when there is not a high load of writes, and good write throughput when there is. See the block comment in vdev_queue.c (reproduced below) for more details. 2. The write throttle (dsl_pool_tempreserve_space() and txg_constrain_throughput()) is rewritten to produce much more consistent delays when under constant load. The new write throttle is based on the amount of dirty data, rather than guesses about future performance of the system. When there is a lot of dirty data, each transaction (e.g. write() syscall) will be delayed by the same small amount. This eliminates the "brick wall of wait" that the old write throttle could hit, causing all transactions to wait several seconds until the next txg opens. One of the keys to the new write throttle is decrementing the amount of dirty data as i/o completes, rather than at the end of spa_sync(). Note that the write throttle is only applied once the i/o scheduler is issuing the maximum number of outstanding async writes. See the block comments in dsl_pool.c and above dmu_tx_delay() (reproduced below) for more details. This diff has several other effects, including: * the commonly-tuned global variable zfs_vdev_max_pending has been removed; use per-class zfs_vdev_*_max_active values or zfs_vdev_max_active instead. * the size of each txg (meaning the amount of dirty data written, and thus the time it takes to write out) is now controlled differently. There is no longer an explicit time goal; the primary determinant is amount of dirty data. Systems that are under light or medium load will now often see that a txg is always syncing, but the impact to performance (e.g. read latency) is minimal. Tune zfs_dirty_data_max and zfs_dirty_data_sync to control this. * zio_taskq_batch_pct = 75 -- Only use 75% of all CPUs for compression, checksum, etc. This improves latency by not allowing these CPU-intensive tasks to consume all CPU (on machines with at least 4 CPU's; the percentage is rounded up). --matt APPENDIX: problems with the current i/o scheduler The current ZFS i/o scheduler (vdev_queue.c) is deadline based. The problem with this is that if there are always i/os pending, then certain classes of i/os can see very long delays. For example, if there are always synchronous reads outstanding, then no async writes will be serviced until they become "past due". One symptom of this situation is that each pass of the txg sync takes at least several seconds (typically 3 seconds). If many i/os become "past due" (their deadline is in the past), then we must service all of these overdue i/os before any new i/os. This happens when we enqueue a batch of async writes for the txg sync, with deadlines 2.5 seconds in the future. If we can't complete all the i/os in 2.5 seconds (e.g. because there were always reads pending), then these i/os will become past due. Now we must service all the "async" writes (which could be hundreds of megabytes) before we service any reads, introducing considerable latency to synchronous i/os (reads or ZIL writes). Notes on porting to ZFS on Linux: - zio_t gained new members io_physdone and io_phys_children. Because object caches in the Linux port call the constructor only once at allocation time, objects may contain residual data when retrieved from the cache. Therefore zio_create() was updated to zero out the two new fields. - vdev_mirror_pending() relied on the depth of the per-vdev pending queue (vq->vq_pending_tree) to select the least-busy leaf vdev to read from. This tree has been replaced by vq->vq_active_tree which is now used for the same purpose. - vdev_queue_init() used the value of zfs_vdev_max_pending to determine the number of vdev I/O buffers to pre-allocate. That global no longer exists, so we instead use the sum of the *_max_active values for each of the five I/O classes described above. - The Illumos implementation of dmu_tx_delay() delays a transaction by sleeping in condition variable embedded in the thread (curthread->t_delay_cv). We do not have an equivalent CV to use in Linux, so this change replaced the delay logic with a wrapper called zfs_sleep_until(). This wrapper could be adopted upstream and in other downstream ports to abstract away operating system-specific delay logic. - These tunables are added as module parameters, and descriptions added to the zfs-module-parameters.5 man page. spa_asize_inflation zfs_deadman_synctime_ms zfs_vdev_max_active zfs_vdev_async_write_active_min_dirty_percent zfs_vdev_async_write_active_max_dirty_percent zfs_vdev_async_read_max_active zfs_vdev_async_read_min_active zfs_vdev_async_write_max_active zfs_vdev_async_write_min_active zfs_vdev_scrub_max_active zfs_vdev_scrub_min_active zfs_vdev_sync_read_max_active zfs_vdev_sync_read_min_active zfs_vdev_sync_write_max_active zfs_vdev_sync_write_min_active zfs_dirty_data_max_percent zfs_delay_min_dirty_percent zfs_dirty_data_max_max_percent zfs_dirty_data_max zfs_dirty_data_max_max zfs_dirty_data_sync zfs_delay_scale The latter four have type unsigned long, whereas they are uint64_t in Illumos. This accommodates Linux's module_param() supported types, but means they may overflow on 32-bit architectures. The values zfs_dirty_data_max and zfs_dirty_data_max_max are the most likely to overflow on 32-bit systems, since they express physical RAM sizes in bytes. In fact, Illumos initializes zfs_dirty_data_max_max to 2^32 which does overflow. To resolve that, this port instead initializes it in arc_init() to 25% of physical RAM, and adds the tunable zfs_dirty_data_max_max_percent to override that percentage. While this solution doesn't completely avoid the overflow issue, it should be a reasonable default for most systems, and the minority of affected systems can work around the issue by overriding the defaults. - Fixed reversed logic in comment above zfs_delay_scale declaration. - Clarified comments in vdev_queue.c regarding when per-queue minimums take effect. - Replaced dmu_tx_write_limit in the dmu_tx kstat file with dmu_tx_dirty_delay and dmu_tx_dirty_over_max. The first counts how many times a transaction has been delayed because the pool dirty data has exceeded zfs_delay_min_dirty_percent. The latter counts how many times the pool dirty data has exceeded zfs_dirty_data_max (which we expect to never happen). - The original patch would have regressed the bug fixed in zfsonlinux/zfs@c418410, which prevented users from setting the zfs_vdev_aggregation_limit tuning larger than SPA_MAXBLOCKSIZE. A similar fix is added to vdev_queue_aggregate(). - In vdev_queue_io_to_issue(), dynamically allocate 'zio_t search' on the heap instead of the stack. In Linux we can't afford such large structures on the stack. Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Ned Bass <bass6@llnl.gov> Reviewed by: Brendan Gregg <brendan.gregg@joyent.com> Approved by: Robert Mustacchi <rm@joyent.com> References: http://www.illumos.org/issues/4045 illumos/illumos-gate@69962b5647e4a8b9b14998733b765925381b727e Ported-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1913
2013-08-29 03:01:20 +00:00
/*
* This should be called from open context when we think we're going to write
* or free space, for example when dirtying data. Be conservative; it's okay
* to write less space or free more, but we don't want to write more or free
* less than the amount specified.
*
* NOTE: The behavior of this function is identical to the Illumos / FreeBSD
* version however it has been adjusted to use an iterative rather then
* recursive algorithm to minimize stack usage.
Illumos #4045 write throttle & i/o scheduler performance work 4045 zfs write throttle & i/o scheduler performance work 1. The ZFS i/o scheduler (vdev_queue.c) now divides i/os into 5 classes: sync read, sync write, async read, async write, and scrub/resilver. The scheduler issues a number of concurrent i/os from each class to the device. Once a class has been selected, an i/o is selected from this class using either an elevator algorithem (async, scrub classes) or FIFO (sync classes). The number of concurrent async write i/os is tuned dynamically based on i/o load, to achieve good sync i/o latency when there is not a high load of writes, and good write throughput when there is. See the block comment in vdev_queue.c (reproduced below) for more details. 2. The write throttle (dsl_pool_tempreserve_space() and txg_constrain_throughput()) is rewritten to produce much more consistent delays when under constant load. The new write throttle is based on the amount of dirty data, rather than guesses about future performance of the system. When there is a lot of dirty data, each transaction (e.g. write() syscall) will be delayed by the same small amount. This eliminates the "brick wall of wait" that the old write throttle could hit, causing all transactions to wait several seconds until the next txg opens. One of the keys to the new write throttle is decrementing the amount of dirty data as i/o completes, rather than at the end of spa_sync(). Note that the write throttle is only applied once the i/o scheduler is issuing the maximum number of outstanding async writes. See the block comments in dsl_pool.c and above dmu_tx_delay() (reproduced below) for more details. This diff has several other effects, including: * the commonly-tuned global variable zfs_vdev_max_pending has been removed; use per-class zfs_vdev_*_max_active values or zfs_vdev_max_active instead. * the size of each txg (meaning the amount of dirty data written, and thus the time it takes to write out) is now controlled differently. There is no longer an explicit time goal; the primary determinant is amount of dirty data. Systems that are under light or medium load will now often see that a txg is always syncing, but the impact to performance (e.g. read latency) is minimal. Tune zfs_dirty_data_max and zfs_dirty_data_sync to control this. * zio_taskq_batch_pct = 75 -- Only use 75% of all CPUs for compression, checksum, etc. This improves latency by not allowing these CPU-intensive tasks to consume all CPU (on machines with at least 4 CPU's; the percentage is rounded up). --matt APPENDIX: problems with the current i/o scheduler The current ZFS i/o scheduler (vdev_queue.c) is deadline based. The problem with this is that if there are always i/os pending, then certain classes of i/os can see very long delays. For example, if there are always synchronous reads outstanding, then no async writes will be serviced until they become "past due". One symptom of this situation is that each pass of the txg sync takes at least several seconds (typically 3 seconds). If many i/os become "past due" (their deadline is in the past), then we must service all of these overdue i/os before any new i/os. This happens when we enqueue a batch of async writes for the txg sync, with deadlines 2.5 seconds in the future. If we can't complete all the i/os in 2.5 seconds (e.g. because there were always reads pending), then these i/os will become past due. Now we must service all the "async" writes (which could be hundreds of megabytes) before we service any reads, introducing considerable latency to synchronous i/os (reads or ZIL writes). Notes on porting to ZFS on Linux: - zio_t gained new members io_physdone and io_phys_children. Because object caches in the Linux port call the constructor only once at allocation time, objects may contain residual data when retrieved from the cache. Therefore zio_create() was updated to zero out the two new fields. - vdev_mirror_pending() relied on the depth of the per-vdev pending queue (vq->vq_pending_tree) to select the least-busy leaf vdev to read from. This tree has been replaced by vq->vq_active_tree which is now used for the same purpose. - vdev_queue_init() used the value of zfs_vdev_max_pending to determine the number of vdev I/O buffers to pre-allocate. That global no longer exists, so we instead use the sum of the *_max_active values for each of the five I/O classes described above. - The Illumos implementation of dmu_tx_delay() delays a transaction by sleeping in condition variable embedded in the thread (curthread->t_delay_cv). We do not have an equivalent CV to use in Linux, so this change replaced the delay logic with a wrapper called zfs_sleep_until(). This wrapper could be adopted upstream and in other downstream ports to abstract away operating system-specific delay logic. - These tunables are added as module parameters, and descriptions added to the zfs-module-parameters.5 man page. spa_asize_inflation zfs_deadman_synctime_ms zfs_vdev_max_active zfs_vdev_async_write_active_min_dirty_percent zfs_vdev_async_write_active_max_dirty_percent zfs_vdev_async_read_max_active zfs_vdev_async_read_min_active zfs_vdev_async_write_max_active zfs_vdev_async_write_min_active zfs_vdev_scrub_max_active zfs_vdev_scrub_min_active zfs_vdev_sync_read_max_active zfs_vdev_sync_read_min_active zfs_vdev_sync_write_max_active zfs_vdev_sync_write_min_active zfs_dirty_data_max_percent zfs_delay_min_dirty_percent zfs_dirty_data_max_max_percent zfs_dirty_data_max zfs_dirty_data_max_max zfs_dirty_data_sync zfs_delay_scale The latter four have type unsigned long, whereas they are uint64_t in Illumos. This accommodates Linux's module_param() supported types, but means they may overflow on 32-bit architectures. The values zfs_dirty_data_max and zfs_dirty_data_max_max are the most likely to overflow on 32-bit systems, since they express physical RAM sizes in bytes. In fact, Illumos initializes zfs_dirty_data_max_max to 2^32 which does overflow. To resolve that, this port instead initializes it in arc_init() to 25% of physical RAM, and adds the tunable zfs_dirty_data_max_max_percent to override that percentage. While this solution doesn't completely avoid the overflow issue, it should be a reasonable default for most systems, and the minority of affected systems can work around the issue by overriding the defaults. - Fixed reversed logic in comment above zfs_delay_scale declaration. - Clarified comments in vdev_queue.c regarding when per-queue minimums take effect. - Replaced dmu_tx_write_limit in the dmu_tx kstat file with dmu_tx_dirty_delay and dmu_tx_dirty_over_max. The first counts how many times a transaction has been delayed because the pool dirty data has exceeded zfs_delay_min_dirty_percent. The latter counts how many times the pool dirty data has exceeded zfs_dirty_data_max (which we expect to never happen). - The original patch would have regressed the bug fixed in zfsonlinux/zfs@c418410, which prevented users from setting the zfs_vdev_aggregation_limit tuning larger than SPA_MAXBLOCKSIZE. A similar fix is added to vdev_queue_aggregate(). - In vdev_queue_io_to_issue(), dynamically allocate 'zio_t search' on the heap instead of the stack. In Linux we can't afford such large structures on the stack. Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Ned Bass <bass6@llnl.gov> Reviewed by: Brendan Gregg <brendan.gregg@joyent.com> Approved by: Robert Mustacchi <rm@joyent.com> References: http://www.illumos.org/issues/4045 illumos/illumos-gate@69962b5647e4a8b9b14998733b765925381b727e Ported-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1913
2013-08-29 03:01:20 +00:00
*/
void
dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
2008-11-20 20:01:55 +00:00
{
int64_t parent_space;
uint64_t est_used;
do {
mutex_enter(&dd->dd_lock);
if (space > 0)
dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;
2008-11-20 20:01:55 +00:00
est_used = dsl_dir_space_towrite(dd) +
dsl_dir_phys(dd)->dd_used_bytes;
parent_space = parent_delta(dd, est_used, space);
mutex_exit(&dd->dd_lock);
2008-11-20 20:01:55 +00:00
/* Make sure that we clean up dd_space_to* */
dsl_dir_dirty(dd, tx);
2008-11-20 20:01:55 +00:00
dd = dd->dd_parent;
space = parent_space;
} while (space && dd);
2008-11-20 20:01:55 +00:00
}
/* call from syncing context when we actually write/free space for this dd */
void
dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
2008-11-20 20:01:55 +00:00
int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
{
int64_t accounted_delta;
/*
* dsl_dataset_set_refreservation_sync_impl() calls this with
* dd_lock held, so that it can atomically update
* ds->ds_reserved and the dsl_dir accounting, so that
* dsl_dataset_check_quota() can see dataset and dir accounting
* consistently.
*/
boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
2008-11-20 20:01:55 +00:00
ASSERT(dmu_tx_is_syncing(tx));
ASSERT(type < DD_USED_NUM);
2008-11-20 20:01:55 +00:00
dmu_buf_will_dirty(dd->dd_dbuf, tx);
if (needlock)
mutex_enter(&dd->dd_lock);
accounted_delta =
parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, used);
ASSERT(used >= 0 || dsl_dir_phys(dd)->dd_used_bytes >= -used);
2008-11-20 20:01:55 +00:00
ASSERT(compressed >= 0 ||
dsl_dir_phys(dd)->dd_compressed_bytes >= -compressed);
2008-11-20 20:01:55 +00:00
ASSERT(uncompressed >= 0 ||
dsl_dir_phys(dd)->dd_uncompressed_bytes >= -uncompressed);
dsl_dir_phys(dd)->dd_used_bytes += used;
dsl_dir_phys(dd)->dd_uncompressed_bytes += uncompressed;
dsl_dir_phys(dd)->dd_compressed_bytes += compressed;
if (dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
ASSERT(used > 0 ||
dsl_dir_phys(dd)->dd_used_breakdown[type] >= -used);
dsl_dir_phys(dd)->dd_used_breakdown[type] += used;
#ifdef DEBUG
{
dd_used_t t;
uint64_t u = 0;
for (t = 0; t < DD_USED_NUM; t++)
u += dsl_dir_phys(dd)->dd_used_breakdown[t];
ASSERT3U(u, ==, dsl_dir_phys(dd)->dd_used_bytes);
}
#endif
}
if (needlock)
mutex_exit(&dd->dd_lock);
2008-11-20 20:01:55 +00:00
if (dd->dd_parent != NULL) {
dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
2008-11-20 20:01:55 +00:00
accounted_delta, compressed, uncompressed, tx);
dsl_dir_transfer_space(dd->dd_parent,
used - accounted_delta,
DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
2008-11-20 20:01:55 +00:00
}
}
void
dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
{
ASSERT(dmu_tx_is_syncing(tx));
ASSERT(oldtype < DD_USED_NUM);
ASSERT(newtype < DD_USED_NUM);
if (delta == 0 ||
!(dsl_dir_phys(dd)->dd_flags & DD_FLAG_USED_BREAKDOWN))
return;
dmu_buf_will_dirty(dd->dd_dbuf, tx);
mutex_enter(&dd->dd_lock);
ASSERT(delta > 0 ?
dsl_dir_phys(dd)->dd_used_breakdown[oldtype] >= delta :
dsl_dir_phys(dd)->dd_used_breakdown[newtype] >= -delta);
ASSERT(dsl_dir_phys(dd)->dd_used_bytes >= ABS(delta));
dsl_dir_phys(dd)->dd_used_breakdown[oldtype] -= delta;
dsl_dir_phys(dd)->dd_used_breakdown[newtype] += delta;
mutex_exit(&dd->dd_lock);
}
typedef struct dsl_dir_set_qr_arg {
const char *ddsqra_name;
zprop_source_t ddsqra_source;
uint64_t ddsqra_value;
} dsl_dir_set_qr_arg_t;
2008-11-20 20:01:55 +00:00
static int
dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
2008-11-20 20:01:55 +00:00
{
dsl_dir_set_qr_arg_t *ddsqra = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dataset_t *ds;
int error;
uint64_t towrite, newval;
2008-11-20 20:01:55 +00:00
error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
if (error != 0)
return (error);
error = dsl_prop_predict(ds->ds_dir, "quota",
ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
if (error != 0) {
dsl_dataset_rele(ds, FTAG);
return (error);
}
if (newval == 0) {
dsl_dataset_rele(ds, FTAG);
2008-11-20 20:01:55 +00:00
return (0);
}
2008-11-20 20:01:55 +00:00
mutex_enter(&ds->ds_dir->dd_lock);
2008-11-20 20:01:55 +00:00
/*
* If we are doing the preliminary check in open context, and
* there are pending changes, then don't fail it, since the
* pending changes could under-estimate the amount of space to be
* freed up.
*/
towrite = dsl_dir_space_towrite(ds->ds_dir);
2008-11-20 20:01:55 +00:00
if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
(newval < dsl_dir_phys(ds->ds_dir)->dd_reserved ||
newval < dsl_dir_phys(ds->ds_dir)->dd_used_bytes + towrite)) {
error = SET_ERROR(ENOSPC);
2008-11-20 20:01:55 +00:00
}
mutex_exit(&ds->ds_dir->dd_lock);
dsl_dataset_rele(ds, FTAG);
return (error);
2008-11-20 20:01:55 +00:00
}
static void
dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
2008-11-20 20:01:55 +00:00
{
dsl_dir_set_qr_arg_t *ddsqra = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dataset_t *ds;
uint64_t newval;
VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2008-11-20 20:01:55 +00:00
if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
&ddsqra->ddsqra_value, tx);
2008-11-20 20:01:55 +00:00
VERIFY0(dsl_prop_get_int_ds(ds,
zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
} else {
newval = ddsqra->ddsqra_value;
spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
zfs_prop_to_name(ZFS_PROP_QUOTA), (longlong_t)newval);
}
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 11:45:09 +00:00
dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
mutex_enter(&ds->ds_dir->dd_lock);
dsl_dir_phys(ds->ds_dir)->dd_quota = newval;
mutex_exit(&ds->ds_dir->dd_lock);
dsl_dataset_rele(ds, FTAG);
2008-11-20 20:01:55 +00:00
}
int
dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
2008-11-20 20:01:55 +00:00
{
dsl_dir_set_qr_arg_t ddsqra;
ddsqra.ddsqra_name = ddname;
ddsqra.ddsqra_source = source;
ddsqra.ddsqra_value = quota;
return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
dsl_dir_set_quota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
2008-11-20 20:01:55 +00:00
}
int
dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
2008-11-20 20:01:55 +00:00
{
dsl_dir_set_qr_arg_t *ddsqra = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dataset_t *ds;
dsl_dir_t *dd;
uint64_t newval, used, avail;
int error;
error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
if (error != 0)
return (error);
dd = ds->ds_dir;
2008-11-20 20:01:55 +00:00
/*
* If we are doing the preliminary check in open context, the
* space estimates may be inaccurate.
*/
if (!dmu_tx_is_syncing(tx)) {
dsl_dataset_rele(ds, FTAG);
2008-11-20 20:01:55 +00:00
return (0);
}
error = dsl_prop_predict(ds->ds_dir,
zfs_prop_to_name(ZFS_PROP_RESERVATION),
ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
if (error != 0) {
dsl_dataset_rele(ds, FTAG);
return (error);
}
2008-11-20 20:01:55 +00:00
mutex_enter(&dd->dd_lock);
used = dsl_dir_phys(dd)->dd_used_bytes;
2008-11-20 20:01:55 +00:00
mutex_exit(&dd->dd_lock);
if (dd->dd_parent) {
avail = dsl_dir_space_available(dd->dd_parent,
NULL, 0, FALSE);
} else {
avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
}
if (MAX(used, newval) > MAX(used, dsl_dir_phys(dd)->dd_reserved)) {
uint64_t delta = MAX(used, newval) -
MAX(used, dsl_dir_phys(dd)->dd_reserved);
2009-02-18 20:51:31 +00:00
if (delta > avail ||
(dsl_dir_phys(dd)->dd_quota > 0 &&
newval > dsl_dir_phys(dd)->dd_quota))
error = SET_ERROR(ENOSPC);
2009-02-18 20:51:31 +00:00
}
dsl_dataset_rele(ds, FTAG);
return (error);
2008-11-20 20:01:55 +00:00
}
void
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 11:45:09 +00:00
dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
2008-11-20 20:01:55 +00:00
{
uint64_t used;
int64_t delta;
dmu_buf_will_dirty(dd->dd_dbuf, tx);
mutex_enter(&dd->dd_lock);
used = dsl_dir_phys(dd)->dd_used_bytes;
delta = MAX(used, value) - MAX(used, dsl_dir_phys(dd)->dd_reserved);
dsl_dir_phys(dd)->dd_reserved = value;
2008-11-20 20:01:55 +00:00
if (dd->dd_parent != NULL) {
/* Roll up this additional usage into our ancestors */
dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
delta, 0, 0, tx);
2008-11-20 20:01:55 +00:00
}
mutex_exit(&dd->dd_lock);
2008-11-20 20:01:55 +00:00
}
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 11:45:09 +00:00
static void
dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 11:45:09 +00:00
{
dsl_dir_set_qr_arg_t *ddsqra = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dataset_t *ds;
uint64_t newval;
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 11:45:09 +00:00
VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
if (spa_version(dp->dp_spa) >= SPA_VERSION_RECVD_PROPS) {
dsl_prop_set_sync_impl(ds,
zfs_prop_to_name(ZFS_PROP_RESERVATION),
ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
&ddsqra->ddsqra_value, tx);
VERIFY0(dsl_prop_get_int_ds(ds,
zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
} else {
newval = ddsqra->ddsqra_value;
spa_history_log_internal_ds(ds, "set", tx, "%s=%lld",
zfs_prop_to_name(ZFS_PROP_RESERVATION),
(longlong_t)newval);
}
dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
dsl_dataset_rele(ds, FTAG);
}
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 11:45:09 +00:00
2008-11-20 20:01:55 +00:00
int
dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
uint64_t reservation)
2008-11-20 20:01:55 +00:00
{
dsl_dir_set_qr_arg_t ddsqra;
ddsqra.ddsqra_name = ddname;
ddsqra.ddsqra_source = source;
ddsqra.ddsqra_value = reservation;
return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
dsl_dir_set_reservation_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
2008-11-20 20:01:55 +00:00
}
static dsl_dir_t *
closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
{
for (; ds1; ds1 = ds1->dd_parent) {
dsl_dir_t *dd;
for (dd = ds2; dd; dd = dd->dd_parent) {
if (ds1 == dd)
return (dd);
}
}
return (NULL);
}
/*
* If delta is applied to dd, how much of that delta would be applied to
* ancestor? Syncing context only.
*/
static int64_t
would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
{
if (dd == ancestor)
return (delta);
mutex_enter(&dd->dd_lock);
delta = parent_delta(dd, dsl_dir_phys(dd)->dd_used_bytes, delta);
2008-11-20 20:01:55 +00:00
mutex_exit(&dd->dd_lock);
return (would_change(dd->dd_parent, delta, ancestor));
}
typedef struct dsl_dir_rename_arg {
const char *ddra_oldname;
const char *ddra_newname;
cred_t *ddra_cred;
} dsl_dir_rename_arg_t;
2008-11-20 20:01:55 +00:00
/* ARGSUSED */
2008-11-20 20:01:55 +00:00
static int
dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2008-11-20 20:01:55 +00:00
{
int *deltap = arg;
char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2008-11-20 20:01:55 +00:00
dsl_dataset_name(ds, namebuf);
if (strlen(namebuf) + *deltap >= ZFS_MAX_DATASET_NAME_LEN)
return (SET_ERROR(ENAMETOOLONG));
return (0);
}
static int
dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
{
dsl_dir_rename_arg_t *ddra = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dir_t *dd, *newparent;
const char *mynewname;
int error;
int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
2008-11-20 20:01:55 +00:00
/* target dir should exist */
error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
if (error != 0)
return (error);
/* new parent should exist */
error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
&newparent, &mynewname);
if (error != 0) {
dsl_dir_rele(dd, FTAG);
return (error);
}
/* can't rename to different pool */
if (dd->dd_pool != newparent->dd_pool) {
dsl_dir_rele(newparent, FTAG);
dsl_dir_rele(dd, FTAG);
return (SET_ERROR(EXDEV));
}
/* new name should not already exist */
if (mynewname == NULL) {
dsl_dir_rele(newparent, FTAG);
dsl_dir_rele(dd, FTAG);
return (SET_ERROR(EEXIST));
}
/* if the name length is growing, validate child name lengths */
if (delta > 0) {
error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
&delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
if (error != 0) {
dsl_dir_rele(newparent, FTAG);
dsl_dir_rele(dd, FTAG);
return (error);
}
}
2008-11-20 20:01:55 +00:00
if (dmu_tx_is_syncing(tx)) {
if (spa_feature_is_active(dp->dp_spa,
SPA_FEATURE_FS_SS_LIMIT)) {
/*
* Although this is the check function and we don't
* normally make on-disk changes in check functions,
* we need to do that here.
*
* Ensure this portion of the tree's counts have been
* initialized in case the new parent has limits set.
*/
dsl_dir_init_fs_ss_count(dd, tx);
}
}
if (newparent != dd->dd_parent) {
2008-11-20 20:01:55 +00:00
/* is there enough space? */
uint64_t myspace =
MAX(dsl_dir_phys(dd)->dd_used_bytes,
dsl_dir_phys(dd)->dd_reserved);
objset_t *os = dd->dd_pool->dp_meta_objset;
uint64_t fs_cnt = 0;
uint64_t ss_cnt = 0;
if (dsl_dir_is_zapified(dd)) {
int err;
err = zap_lookup(os, dd->dd_object,
DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
&fs_cnt);
if (err != ENOENT && err != 0) {
dsl_dir_rele(newparent, FTAG);
dsl_dir_rele(dd, FTAG);
return (err);
}
/*
* have to add 1 for the filesystem itself that we're
* moving
*/
fs_cnt++;
err = zap_lookup(os, dd->dd_object,
DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
&ss_cnt);
if (err != ENOENT && err != 0) {
dsl_dir_rele(newparent, FTAG);
dsl_dir_rele(dd, FTAG);
return (err);
}
}
2008-11-20 20:01:55 +00:00
/* no rename into our descendant */
if (closest_common_ancestor(dd, newparent) == dd) {
dsl_dir_rele(newparent, FTAG);
dsl_dir_rele(dd, FTAG);
return (SET_ERROR(EINVAL));
}
2008-11-20 20:01:55 +00:00
error = dsl_dir_transfer_possible(dd->dd_parent,
newparent, fs_cnt, ss_cnt, myspace, ddra->ddra_cred);
if (error != 0) {
dsl_dir_rele(newparent, FTAG);
dsl_dir_rele(dd, FTAG);
return (error);
}
2008-11-20 20:01:55 +00:00
}
dsl_dir_rele(newparent, FTAG);
dsl_dir_rele(dd, FTAG);
2008-11-20 20:01:55 +00:00
return (0);
}
static void
dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
2008-11-20 20:01:55 +00:00
{
dsl_dir_rename_arg_t *ddra = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dir_t *dd, *newparent;
const char *mynewname;
int error;
2008-11-20 20:01:55 +00:00
objset_t *mos = dp->dp_meta_objset;
VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
&mynewname));
2008-11-20 20:01:55 +00:00
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 11:45:09 +00:00
/* Log this before we change the name. */
spa_history_log_internal_dd(dd, "rename", tx,
"-> %s", ddra->ddra_newname);
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 11:45:09 +00:00
if (newparent != dd->dd_parent) {
objset_t *os = dd->dd_pool->dp_meta_objset;
uint64_t fs_cnt = 0;
uint64_t ss_cnt = 0;
/*
* We already made sure the dd counts were initialized in the
* check function.
*/
if (spa_feature_is_active(dp->dp_spa,
SPA_FEATURE_FS_SS_LIMIT)) {
VERIFY0(zap_lookup(os, dd->dd_object,
DD_FIELD_FILESYSTEM_COUNT, sizeof (fs_cnt), 1,
&fs_cnt));
/* add 1 for the filesystem itself that we're moving */
fs_cnt++;
VERIFY0(zap_lookup(os, dd->dd_object,
DD_FIELD_SNAPSHOT_COUNT, sizeof (ss_cnt), 1,
&ss_cnt));
}
dsl_fs_ss_count_adjust(dd->dd_parent, -fs_cnt,
DD_FIELD_FILESYSTEM_COUNT, tx);
dsl_fs_ss_count_adjust(newparent, fs_cnt,
DD_FIELD_FILESYSTEM_COUNT, tx);
dsl_fs_ss_count_adjust(dd->dd_parent, -ss_cnt,
DD_FIELD_SNAPSHOT_COUNT, tx);
dsl_fs_ss_count_adjust(newparent, ss_cnt,
DD_FIELD_SNAPSHOT_COUNT, tx);
dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
-dsl_dir_phys(dd)->dd_used_bytes,
-dsl_dir_phys(dd)->dd_compressed_bytes,
-dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
dsl_dir_diduse_space(newparent, DD_USED_CHILD,
dsl_dir_phys(dd)->dd_used_bytes,
dsl_dir_phys(dd)->dd_compressed_bytes,
dsl_dir_phys(dd)->dd_uncompressed_bytes, tx);
if (dsl_dir_phys(dd)->dd_reserved >
dsl_dir_phys(dd)->dd_used_bytes) {
uint64_t unused_rsrv = dsl_dir_phys(dd)->dd_reserved -
dsl_dir_phys(dd)->dd_used_bytes;
dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
-unused_rsrv, 0, 0, tx);
dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
unused_rsrv, 0, 0, tx);
}
2008-11-20 20:01:55 +00:00
}
dmu_buf_will_dirty(dd->dd_dbuf, tx);
/* remove from old parent zapobj */
error = zap_remove(mos,
dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
2008-11-20 20:01:55 +00:00
dd->dd_myname, tx);
ASSERT0(error);
2008-11-20 20:01:55 +00:00
(void) strlcpy(dd->dd_myname, mynewname,
sizeof (dd->dd_myname));
dsl_dir_rele(dd->dd_parent, dd);
dsl_dir_phys(dd)->dd_parent_obj = newparent->dd_object;
VERIFY0(dsl_dir_hold_obj(dp,
newparent->dd_object, NULL, dd, &dd->dd_parent));
2008-11-20 20:01:55 +00:00
/* add to new parent zapobj */
VERIFY0(zap_add(mos, dsl_dir_phys(newparent)->dd_child_dir_zapobj,
dd->dd_myname, 8, 1, &dd->dd_object, tx));
zvol_rename_minors(dp->dp_spa, ddra->ddra_oldname,
ddra->ddra_newname, B_TRUE);
Remove ZFC_IOC_*_MINOR ioctl()s Early versions of ZFS coordinated the creation and destruction of device minors from userspace. This was inherently racy and in late 2009 these ioctl()s were removed leaving everything up to the kernel. This significantly simplified the code. However, we never picked up these changes in ZoL since we'd already significantly adjusted this code for Linux. This patch aims to rectify that by finally removing ZFC_IOC_*_MINOR ioctl()s and moving all the functionality down in to the kernel. Since this cleanup will change the kernel/user ABI it's being done in the same tag as the previous libzfs_core ABI changes. This will minimize, but not eliminate, the disruption to end users. Once merged ZoL, Illumos, and FreeBSD will basically be back in sync in regards to handling ZVOLs in the common code. While each platform must have its own custom zvol.c implemenation the interfaces provided are consistent. NOTES: 1) This patch introduces one subtle change in behavior which could not be easily avoided. Prior to this change callers of 'zfs create -V ...' were guaranteed that upon exit the /dev/zvol/ block device link would be created or an error returned. That's no longer the case. The utilities will no longer block waiting for the symlink to be created. Callers are now responsible for blocking, this is why a 'udev_wait' call was added to the 'label' function in scripts/common.sh. 2) The read-only behavior of a ZVOL now solely depends on if the ZVOL_RDONLY bit is set in zv->zv_flags. The redundant policy setting in the gendisk structure was removed. This both simplifies the code and allows us to safely leverage set_disk_ro() to issue a KOBJ_CHANGE uevent. See the comment in the code for futher details on this. 3) Because __zvol_create_minor() and zvol_alloc() may now be called in a sync task they must use KM_PUSHPAGE. References: illumos/illumos-gate@681d9761e8516a7dc5ab6589e2dfe717777e1123 Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Tim Chase <tim@chase2k.com> Closes #1969
2013-12-06 22:20:22 +00:00
dsl_prop_notify_all(dd);
2008-11-20 20:01:55 +00:00
dsl_dir_rele(newparent, FTAG);
dsl_dir_rele(dd, FTAG);
2008-11-20 20:01:55 +00:00
}
int
dsl_dir_rename(const char *oldname, const char *newname)
2008-11-20 20:01:55 +00:00
{
dsl_dir_rename_arg_t ddra;
2008-11-20 20:01:55 +00:00
ddra.ddra_oldname = oldname;
ddra.ddra_newname = newname;
ddra.ddra_cred = CRED();
2008-11-20 20:01:55 +00:00
return (dsl_sync_task(oldname,
dsl_dir_rename_check, dsl_dir_rename_sync, &ddra,
3, ZFS_SPACE_CHECK_RESERVED));
2008-11-20 20:01:55 +00:00
}
int
dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd,
uint64_t fs_cnt, uint64_t ss_cnt, uint64_t space, cred_t *cr)
2008-11-20 20:01:55 +00:00
{
dsl_dir_t *ancestor;
int64_t adelta;
uint64_t avail;
int err;
2008-11-20 20:01:55 +00:00
ancestor = closest_common_ancestor(sdd, tdd);
adelta = would_change(sdd, -space, ancestor);
avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
if (avail < space)
return (SET_ERROR(ENOSPC));
2008-11-20 20:01:55 +00:00
err = dsl_fs_ss_limit_check(tdd, fs_cnt, ZFS_PROP_FILESYSTEM_LIMIT,
ancestor, cr);
if (err != 0)
return (err);
err = dsl_fs_ss_limit_check(tdd, ss_cnt, ZFS_PROP_SNAPSHOT_LIMIT,
ancestor, cr);
if (err != 0)
return (err);
2008-11-20 20:01:55 +00:00
return (0);
}
timestruc_t
dsl_dir_snap_cmtime(dsl_dir_t *dd)
{
timestruc_t t;
mutex_enter(&dd->dd_lock);
t = dd->dd_snap_cmtime;
mutex_exit(&dd->dd_lock);
return (t);
}
void
dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
{
timestruc_t t;
gethrestime(&t);
mutex_enter(&dd->dd_lock);
dd->dd_snap_cmtime = t;
mutex_exit(&dd->dd_lock);
}
void
dsl_dir_zapify(dsl_dir_t *dd, dmu_tx_t *tx)
{
objset_t *mos = dd->dd_pool->dp_meta_objset;
dmu_object_zapify(mos, dd->dd_object, DMU_OT_DSL_DIR, tx);
}
boolean_t
dsl_dir_is_zapified(dsl_dir_t *dd)
{
dmu_object_info_t doi;
dmu_object_info_from_db(dd->dd_dbuf, &doi);
return (doi.doi_type == DMU_OTN_ZAP_METADATA);
}
#if defined(_KERNEL) && defined(HAVE_SPL)
EXPORT_SYMBOL(dsl_dir_set_quota);
EXPORT_SYMBOL(dsl_dir_set_reservation);
#endif