diff --git a/sys/contrib/openzfs/META b/sys/contrib/openzfs/META
index 2ea3a7300f23..9776223b88d8 100644
--- a/sys/contrib/openzfs/META
+++ b/sys/contrib/openzfs/META
@@ -1,10 +1,10 @@
Meta: 1
Name: zfs
Branch: 1.0
Version: 2.1.99
Release: 1
Release-Tags: relext
License: CDDL
Author: OpenZFS
-Linux-Maximum: 5.12
+Linux-Maximum: 5.13
Linux-Minimum: 3.10
diff --git a/sys/contrib/openzfs/cmd/zed/zed.d/zed-functions.sh b/sys/contrib/openzfs/cmd/zed/zed.d/zed-functions.sh
index 9f8531d737a6..2ec0ea6948d8 100644
--- a/sys/contrib/openzfs/cmd/zed/zed.d/zed-functions.sh
+++ b/sys/contrib/openzfs/cmd/zed/zed.d/zed-functions.sh
@@ -1,614 +1,614 @@
#!/bin/sh
# shellcheck disable=SC2039
# zed-functions.sh
#
# ZED helper functions for use in ZEDLETs
# Variable Defaults
#
: "${ZED_LOCKDIR:="/var/lock"}"
: "${ZED_NOTIFY_INTERVAL_SECS:=3600}"
: "${ZED_NOTIFY_VERBOSE:=0}"
: "${ZED_RUNDIR:="/var/run"}"
: "${ZED_SYSLOG_PRIORITY:="daemon.notice"}"
: "${ZED_SYSLOG_TAG:="zed"}"
ZED_FLOCK_FD=8
# zed_check_cmd (cmd, ...)
#
# For each argument given, search PATH for the executable command [cmd].
# Log a message if [cmd] is not found.
#
# Arguments
# cmd: name of executable command for which to search
#
# Return
# 0 if all commands are found in PATH and are executable
# n for a count of the command executables that are not found
#
zed_check_cmd()
{
local cmd
local rv=0
for cmd; do
if ! command -v "${cmd}" >/dev/null 2>&1; then
zed_log_err "\"${cmd}\" not installed"
rv=$((rv + 1))
fi
done
return "${rv}"
}
# zed_log_msg (msg, ...)
#
# Write all argument strings to the system log.
#
# Globals
# ZED_SYSLOG_PRIORITY
# ZED_SYSLOG_TAG
#
# Return
# nothing
#
zed_log_msg()
{
logger -p "${ZED_SYSLOG_PRIORITY}" -t "${ZED_SYSLOG_TAG}" -- "$@"
}
# zed_log_err (msg, ...)
#
# Write an error message to the system log. This message will contain the
# script name, EID, and all argument strings.
#
# Globals
# ZED_SYSLOG_PRIORITY
# ZED_SYSLOG_TAG
# ZEVENT_EID
#
# Return
# nothing
#
zed_log_err()
{
logger -p "${ZED_SYSLOG_PRIORITY}" -t "${ZED_SYSLOG_TAG}" -- "error:" \
"$(basename -- "$0"):""${ZEVENT_EID:+" eid=${ZEVENT_EID}:"}" "$@"
}
# zed_lock (lockfile, [fd])
#
# Obtain an exclusive (write) lock on [lockfile]. If the lock cannot be
# immediately acquired, wait until it becomes available.
#
# Every zed_lock() must be paired with a corresponding zed_unlock().
#
# By default, flock-style locks associate the lockfile with file descriptor 8.
# The bash manpage warns that file descriptors >9 should be used with care as
# they may conflict with file descriptors used internally by the shell. File
# descriptor 9 is reserved for zed_rate_limit(). If concurrent locks are held
# within the same process, they must use different file descriptors (preferably
# decrementing from 8); otherwise, obtaining a new lock with a given file
# descriptor will release the previous lock associated with that descriptor.
#
# Arguments
# lockfile: pathname of the lock file; the lock will be stored in
# ZED_LOCKDIR unless the pathname contains a "/".
# fd: integer for the file descriptor used by flock (OPTIONAL unless holding
# concurrent locks)
#
# Globals
# ZED_FLOCK_FD
# ZED_LOCKDIR
#
# Return
# nothing
#
zed_lock()
{
local lockfile="$1"
local fd="${2:-${ZED_FLOCK_FD}}"
local umask_bak
local err
[ -n "${lockfile}" ] || return
if ! expr "${lockfile}" : '.*/' >/dev/null 2>&1; then
lockfile="${ZED_LOCKDIR}/${lockfile}"
fi
umask_bak="$(umask)"
umask 077
# Obtain a lock on the file bound to the given file descriptor.
#
eval "exec ${fd}>> '${lockfile}'"
if ! err="$(flock --exclusive "${fd}" 2>&1)"; then
zed_log_err "failed to lock \"${lockfile}\": ${err}"
fi
umask "${umask_bak}"
}
# zed_unlock (lockfile, [fd])
#
# Release the lock on [lockfile].
#
# Arguments
# lockfile: pathname of the lock file
# fd: integer for the file descriptor used by flock (must match the file
# descriptor passed to the zed_lock function call)
#
# Globals
# ZED_FLOCK_FD
# ZED_LOCKDIR
#
# Return
# nothing
#
zed_unlock()
{
local lockfile="$1"
local fd="${2:-${ZED_FLOCK_FD}}"
local err
[ -n "${lockfile}" ] || return
if ! expr "${lockfile}" : '.*/' >/dev/null 2>&1; then
lockfile="${ZED_LOCKDIR}/${lockfile}"
fi
# Release the lock and close the file descriptor.
if ! err="$(flock --unlock "${fd}" 2>&1)"; then
zed_log_err "failed to unlock \"${lockfile}\": ${err}"
fi
eval "exec ${fd}>&-"
}
# zed_notify (subject, pathname)
#
# Send a notification via all available methods.
#
# Arguments
# subject: notification subject
# pathname: pathname containing the notification message (OPTIONAL)
#
# Return
# 0: notification succeeded via at least one method
# 1: notification failed
# 2: no notification methods configured
#
zed_notify()
{
local subject="$1"
local pathname="$2"
local num_success=0
local num_failure=0
zed_notify_email "${subject}" "${pathname}"; rv=$?
[ "${rv}" -eq 0 ] && num_success=$((num_success + 1))
[ "${rv}" -eq 1 ] && num_failure=$((num_failure + 1))
zed_notify_pushbullet "${subject}" "${pathname}"; rv=$?
[ "${rv}" -eq 0 ] && num_success=$((num_success + 1))
[ "${rv}" -eq 1 ] && num_failure=$((num_failure + 1))
zed_notify_slack_webhook "${subject}" "${pathname}"; rv=$?
[ "${rv}" -eq 0 ] && num_success=$((num_success + 1))
[ "${rv}" -eq 1 ] && num_failure=$((num_failure + 1))
zed_notify_pushover "${subject}" "${pathname}"; rv=$?
[ "${rv}" -eq 0 ] && num_success=$((num_success + 1))
[ "${rv}" -eq 1 ] && num_failure=$((num_failure + 1))
[ "${num_success}" -gt 0 ] && return 0
[ "${num_failure}" -gt 0 ] && return 1
return 2
}
# zed_notify_email (subject, pathname)
#
# Send a notification via email to the address specified by ZED_EMAIL_ADDR.
#
# Requires the mail executable to be installed in the standard PATH, or
# ZED_EMAIL_PROG to be defined with the pathname of an executable capable of
# reading a message body from stdin.
#
# Command-line options to the mail executable can be specified in
# ZED_EMAIL_OPTS. This undergoes the following keyword substitutions:
# - @ADDRESS@ is replaced with the space-delimited recipient email address(es)
# - @SUBJECT@ is replaced with the notification subject
#
# Arguments
# subject: notification subject
# pathname: pathname containing the notification message (OPTIONAL)
#
# Globals
# ZED_EMAIL_PROG
# ZED_EMAIL_OPTS
# ZED_EMAIL_ADDR
#
# Return
# 0: notification sent
# 1: notification failed
# 2: not configured
#
zed_notify_email()
{
local subject="$1"
local pathname="${2:-"/dev/null"}"
: "${ZED_EMAIL_PROG:="mail"}"
: "${ZED_EMAIL_OPTS:="-s '@SUBJECT@' @ADDRESS@"}"
# For backward compatibility with ZED_EMAIL.
if [ -n "${ZED_EMAIL}" ] && [ -z "${ZED_EMAIL_ADDR}" ]; then
ZED_EMAIL_ADDR="${ZED_EMAIL}"
fi
[ -n "${ZED_EMAIL_ADDR}" ] || return 2
zed_check_cmd "${ZED_EMAIL_PROG}" || return 1
[ -n "${subject}" ] || return 1
if [ ! -r "${pathname}" ]; then
zed_log_err \
"$(basename "${ZED_EMAIL_PROG}") cannot read \"${pathname}\""
return 1
fi
ZED_EMAIL_OPTS="$(echo "${ZED_EMAIL_OPTS}" \
| sed -e "s/@ADDRESS@/${ZED_EMAIL_ADDR}/g" \
-e "s/@SUBJECT@/${subject}/g")"
# shellcheck disable=SC2086
- ${ZED_EMAIL_PROG} ${ZED_EMAIL_OPTS} < "${pathname}" >/dev/null 2>&1
+ eval ${ZED_EMAIL_PROG} ${ZED_EMAIL_OPTS} < "${pathname}" >/dev/null 2>&1
rv=$?
if [ "${rv}" -ne 0 ]; then
zed_log_err "$(basename "${ZED_EMAIL_PROG}") exit=${rv}"
return 1
fi
return 0
}
# zed_notify_pushbullet (subject, pathname)
#
# Send a notification via Pushbullet .
# The access token (ZED_PUSHBULLET_ACCESS_TOKEN) identifies this client to the
# Pushbullet server. The optional channel tag (ZED_PUSHBULLET_CHANNEL_TAG) is
# for pushing to notification feeds that can be subscribed to; if a channel is
# not defined, push notifications will instead be sent to all devices
# associated with the account specified by the access token.
#
# Requires awk, curl, and sed executables to be installed in the standard PATH.
#
# References
# https://docs.pushbullet.com/
# https://www.pushbullet.com/security
#
# Arguments
# subject: notification subject
# pathname: pathname containing the notification message (OPTIONAL)
#
# Globals
# ZED_PUSHBULLET_ACCESS_TOKEN
# ZED_PUSHBULLET_CHANNEL_TAG
#
# Return
# 0: notification sent
# 1: notification failed
# 2: not configured
#
zed_notify_pushbullet()
{
local subject="$1"
local pathname="${2:-"/dev/null"}"
local msg_body
local msg_tag
local msg_json
local msg_out
local msg_err
local url="https://api.pushbullet.com/v2/pushes"
[ -n "${ZED_PUSHBULLET_ACCESS_TOKEN}" ] || return 2
[ -n "${subject}" ] || return 1
if [ ! -r "${pathname}" ]; then
zed_log_err "pushbullet cannot read \"${pathname}\""
return 1
fi
zed_check_cmd "awk" "curl" "sed" || return 1
# Escape the following characters in the message body for JSON:
# newline, backslash, double quote, horizontal tab, vertical tab,
# and carriage return.
#
msg_body="$(awk '{ ORS="\\n" } { gsub(/\\/, "\\\\"); gsub(/"/, "\\\"");
gsub(/\t/, "\\t"); gsub(/\f/, "\\f"); gsub(/\r/, "\\r"); print }' \
"${pathname}")"
# Push to a channel if one is configured.
#
[ -n "${ZED_PUSHBULLET_CHANNEL_TAG}" ] && msg_tag="$(printf \
'"channel_tag": "%s", ' "${ZED_PUSHBULLET_CHANNEL_TAG}")"
# Construct the JSON message for pushing a note.
#
msg_json="$(printf '{%s"type": "note", "title": "%s", "body": "%s"}' \
"${msg_tag}" "${subject}" "${msg_body}")"
# Send the POST request and check for errors.
#
msg_out="$(curl -u "${ZED_PUSHBULLET_ACCESS_TOKEN}:" -X POST "${url}" \
--header "Content-Type: application/json" --data-binary "${msg_json}" \
2>/dev/null)"; rv=$?
if [ "${rv}" -ne 0 ]; then
zed_log_err "curl exit=${rv}"
return 1
fi
msg_err="$(echo "${msg_out}" \
| sed -n -e 's/.*"error" *:.*"message" *: *"\([^"]*\)".*/\1/p')"
if [ -n "${msg_err}" ]; then
zed_log_err "pushbullet \"${msg_err}"\"
return 1
fi
return 0
}
# zed_notify_slack_webhook (subject, pathname)
#
# Notification via Slack Webhook .
# The Webhook URL (ZED_SLACK_WEBHOOK_URL) identifies this client to the
# Slack channel.
#
# Requires awk, curl, and sed executables to be installed in the standard PATH.
#
# References
# https://api.slack.com/incoming-webhooks
#
# Arguments
# subject: notification subject
# pathname: pathname containing the notification message (OPTIONAL)
#
# Globals
# ZED_SLACK_WEBHOOK_URL
#
# Return
# 0: notification sent
# 1: notification failed
# 2: not configured
#
zed_notify_slack_webhook()
{
[ -n "${ZED_SLACK_WEBHOOK_URL}" ] || return 2
local subject="$1"
local pathname="${2:-"/dev/null"}"
local msg_body
local msg_tag
local msg_json
local msg_out
local msg_err
local url="${ZED_SLACK_WEBHOOK_URL}"
[ -n "${subject}" ] || return 1
if [ ! -r "${pathname}" ]; then
zed_log_err "slack webhook cannot read \"${pathname}\""
return 1
fi
zed_check_cmd "awk" "curl" "sed" || return 1
# Escape the following characters in the message body for JSON:
# newline, backslash, double quote, horizontal tab, vertical tab,
# and carriage return.
#
msg_body="$(awk '{ ORS="\\n" } { gsub(/\\/, "\\\\"); gsub(/"/, "\\\"");
gsub(/\t/, "\\t"); gsub(/\f/, "\\f"); gsub(/\r/, "\\r"); print }' \
"${pathname}")"
# Construct the JSON message for posting.
#
msg_json="$(printf '{"text": "*%s*\n%s"}' "${subject}" "${msg_body}" )"
# Send the POST request and check for errors.
#
msg_out="$(curl -X POST "${url}" \
--header "Content-Type: application/json" --data-binary "${msg_json}" \
2>/dev/null)"; rv=$?
if [ "${rv}" -ne 0 ]; then
zed_log_err "curl exit=${rv}"
return 1
fi
msg_err="$(echo "${msg_out}" \
| sed -n -e 's/.*"error" *:.*"message" *: *"\([^"]*\)".*/\1/p')"
if [ -n "${msg_err}" ]; then
zed_log_err "slack webhook \"${msg_err}"\"
return 1
fi
return 0
}
# zed_notify_pushover (subject, pathname)
#
# Send a notification via Pushover .
# The access token (ZED_PUSHOVER_TOKEN) identifies this client to the
# Pushover server. The user token (ZED_PUSHOVER_USER) defines the user or
# group to which the notification will be sent.
#
# Requires curl and sed executables to be installed in the standard PATH.
#
# References
# https://pushover.net/api
#
# Arguments
# subject: notification subject
# pathname: pathname containing the notification message (OPTIONAL)
#
# Globals
# ZED_PUSHOVER_TOKEN
# ZED_PUSHOVER_USER
#
# Return
# 0: notification sent
# 1: notification failed
# 2: not configured
#
zed_notify_pushover()
{
local subject="$1"
local pathname="${2:-"/dev/null"}"
local msg_body
local msg_out
local msg_err
local url="https://api.pushover.net/1/messages.json"
[ -n "${ZED_PUSHOVER_TOKEN}" ] && [ -n "${ZED_PUSHOVER_USER}" ] || return 2
if [ ! -r "${pathname}" ]; then
zed_log_err "pushover cannot read \"${pathname}\""
return 1
fi
zed_check_cmd "curl" "sed" || return 1
# Read the message body in.
#
msg_body="$(cat "${pathname}")"
if [ -z "${msg_body}" ]
then
msg_body=$subject
subject=""
fi
# Send the POST request and check for errors.
#
msg_out="$( \
curl \
--form-string "token=${ZED_PUSHOVER_TOKEN}" \
--form-string "user=${ZED_PUSHOVER_USER}" \
--form-string "message=${msg_body}" \
--form-string "title=${subject}" \
"${url}" \
2>/dev/null \
)"; rv=$?
if [ "${rv}" -ne 0 ]; then
zed_log_err "curl exit=${rv}"
return 1
fi
msg_err="$(echo "${msg_out}" \
| sed -n -e 's/.*"errors" *:.*\[\(.*\)\].*/\1/p')"
if [ -n "${msg_err}" ]; then
zed_log_err "pushover \"${msg_err}"\"
return 1
fi
return 0
}
# zed_rate_limit (tag, [interval])
#
# Check whether an event of a given type [tag] has already occurred within the
# last [interval] seconds.
#
# This function obtains a lock on the statefile using file descriptor 9.
#
# Arguments
# tag: arbitrary string for grouping related events to rate-limit
# interval: time interval in seconds (OPTIONAL)
#
# Globals
# ZED_NOTIFY_INTERVAL_SECS
# ZED_RUNDIR
#
# Return
# 0 if the event should be processed
# 1 if the event should be dropped
#
# State File Format
# time;tag
#
zed_rate_limit()
{
local tag="$1"
local interval="${2:-${ZED_NOTIFY_INTERVAL_SECS}}"
local lockfile="zed.zedlet.state.lock"
local lockfile_fd=9
local statefile="${ZED_RUNDIR}/zed.zedlet.state"
local time_now
local time_prev
local umask_bak
local rv=0
[ -n "${tag}" ] || return 0
zed_lock "${lockfile}" "${lockfile_fd}"
time_now="$(date +%s)"
time_prev="$(grep -E "^[0-9]+;${tag}\$" "${statefile}" 2>/dev/null \
| tail -1 | cut -d\; -f1)"
if [ -n "${time_prev}" ] \
&& [ "$((time_now - time_prev))" -lt "${interval}" ]; then
rv=1
else
umask_bak="$(umask)"
umask 077
grep -E -v "^[0-9]+;${tag}\$" "${statefile}" 2>/dev/null \
> "${statefile}.$$"
echo "${time_now};${tag}" >> "${statefile}.$$"
mv -f "${statefile}.$$" "${statefile}"
umask "${umask_bak}"
fi
zed_unlock "${lockfile}" "${lockfile_fd}"
return "${rv}"
}
# zed_guid_to_pool (guid)
#
# Convert a pool GUID into its pool name (like "tank")
# Arguments
# guid: pool GUID (decimal or hex)
#
# Return
# Pool name
#
zed_guid_to_pool()
{
if [ -z "$1" ] ; then
return
fi
guid="$(printf "%u" "$1")"
$ZPOOL get -H -ovalue,name guid | awk '$1 == '"$guid"' {print $2; exit}'
}
# zed_exit_if_ignoring_this_event
#
# Exit the script if we should ignore this event, as determined by
# $ZED_SYSLOG_SUBCLASS_INCLUDE and $ZED_SYSLOG_SUBCLASS_EXCLUDE in zed.rc.
# This function assumes you've imported the normal zed variables.
zed_exit_if_ignoring_this_event()
{
if [ -n "${ZED_SYSLOG_SUBCLASS_INCLUDE}" ]; then
eval "case ${ZEVENT_SUBCLASS} in
${ZED_SYSLOG_SUBCLASS_INCLUDE});;
*) exit 0;;
esac"
elif [ -n "${ZED_SYSLOG_SUBCLASS_EXCLUDE}" ]; then
eval "case ${ZEVENT_SUBCLASS} in
${ZED_SYSLOG_SUBCLASS_EXCLUDE}) exit 0;;
*);;
esac"
fi
}
diff --git a/sys/contrib/openzfs/contrib/dracut/90zfs/zfs-lib.sh.in b/sys/contrib/openzfs/contrib/dracut/90zfs/zfs-lib.sh.in
index 10b0b701a233..defc0bfc8e76 100755
--- a/sys/contrib/openzfs/contrib/dracut/90zfs/zfs-lib.sh.in
+++ b/sys/contrib/openzfs/contrib/dracut/90zfs/zfs-lib.sh.in
@@ -1,207 +1,207 @@
#!/bin/sh
command -v getarg >/dev/null || . /lib/dracut-lib.sh
command -v getargbool >/dev/null || {
# Compatibility with older Dracut versions.
# With apologies to the Dracut developers.
getargbool() {
_default="$1"; shift
! _b=$(getarg "$@") && [ -z "$_b" ] && _b="$_default"
if [ -n "$_b" ]; then
[ "$_b" = "0" ] && return 1
[ "$_b" = "no" ] && return 1
[ "$_b" = "off" ] && return 1
fi
return 0
}
}
OLDIFS="${IFS}"
NEWLINE="
"
TAB=" "
ZPOOL_IMPORT_OPTS=""
if getargbool 0 zfs_force -y zfs.force -y zfsforce ; then
warn "ZFS: Will force-import pools if necessary."
ZPOOL_IMPORT_OPTS="${ZPOOL_IMPORT_OPTS} -f"
fi
# find_bootfs
# returns the first dataset with the bootfs attribute.
find_bootfs() {
IFS="${NEWLINE}"
for dataset in $(zpool list -H -o bootfs); do
case "${dataset}" in
"" | "-")
continue
;;
"no pools available")
IFS="${OLDIFS}"
return 1
;;
*)
IFS="${OLDIFS}"
echo "${dataset}"
return 0
;;
esac
done
IFS="${OLDIFS}"
return 1
}
# import_pool POOL
# imports the given zfs pool if it isn't imported already.
import_pool() {
pool="${1}"
if ! zpool list -H "${pool}" > /dev/null 2>&1; then
info "ZFS: Importing pool ${pool}..."
# shellcheck disable=SC2086
if ! zpool import -N ${ZPOOL_IMPORT_OPTS} "${pool}" ; then
warn "ZFS: Unable to import pool ${pool}"
return 1
fi
fi
return 0
}
_mount_dataset_cb() {
mount -o zfsutil -t zfs "${1}" "${NEWROOT}${2}"
}
# mount_dataset DATASET
# mounts the given zfs dataset.
mount_dataset() {
dataset="${1}"
mountpoint="$(zfs get -H -o value mountpoint "${dataset}")"
ret=0
# We need zfsutil for non-legacy mounts and not for legacy mounts.
if [ "${mountpoint}" = "legacy" ] ; then
mount -t zfs "${dataset}" "${NEWROOT}" || ret=$?
else
mount -o zfsutil -t zfs "${dataset}" "${NEWROOT}" || ret=$?
if [ "$ret" = "0" ]; then
for_relevant_root_children "${dataset}" _mount_dataset_cb || ret=$?
fi
fi
return ${ret}
}
# for_relevant_root_children DATASET EXEC
# Runs "EXEC dataset mountpoint" for all children of DATASET that are needed for system bringup
# Used by zfs-generator.sh and friends, too!
for_relevant_root_children() {
dataset="${1}"
exec="${2}"
zfs list -t filesystem -Ho name,mountpoint,canmount -r "${dataset}" |
(
_ret=0
while IFS="${TAB}" read -r dataset mountpoint canmount; do
[ "$canmount" != "on" ] && continue
case "$mountpoint" in
/etc|/bin|/lib|/lib??|/libx32|/usr)
# If these aren't mounted we may not be able to get to the real init at all, or pollute the dataset holding the rootfs
"${exec}" "${dataset}" "${mountpoint}" || _ret=$?
;;
*)
# Up to the real init to remount everything else it might need
;;
esac
done
exit ${_ret}
)
}
# export_all OPTS
# exports all imported zfs pools.
export_all() {
ret=0
IFS="${NEWLINE}"
for pool in $(zpool list -H -o name) ; do
if zpool list -H "${pool}" > /dev/null 2>&1; then
zpool export "${pool}" "$@" || ret=$?
fi
done
IFS="${OLDIFS}"
return ${ret}
}
# ask_for_password
#
# Wraps around plymouth ask-for-password and adds fallback to tty password ask
# if plymouth is not present.
#
# --cmd command
# Command to execute. Required.
# --prompt prompt
# Password prompt. Note that function already adds ':' at the end.
# Recommended.
# --tries n
# How many times repeat command on its failure. Default is 3.
# --ply-[cmd|prompt|tries]
# Command/prompt/tries specific for plymouth password ask only.
# --tty-[cmd|prompt|tries]
# Command/prompt/tries specific for tty password ask only.
# --tty-echo-off
# Turn off input echo before tty command is executed and turn on after.
# It's useful when password is read from stdin.
ask_for_password() {
ply_tries=3
tty_tries=3
while [ "$#" -gt 0 ]; do
case "$1" in
--cmd) ply_cmd="$2"; tty_cmd="$2"; shift;;
--ply-cmd) ply_cmd="$2"; shift;;
--tty-cmd) tty_cmd="$2"; shift;;
--prompt) ply_prompt="$2"; tty_prompt="$2"; shift;;
--ply-prompt) ply_prompt="$2"; shift;;
--tty-prompt) tty_prompt="$2"; shift;;
--tries) ply_tries="$2"; tty_tries="$2"; shift;;
--ply-tries) ply_tries="$2"; shift;;
--tty-tries) tty_tries="$2"; shift;;
--tty-echo-off) tty_echo_off=yes;;
esac
shift
done
{ flock -s 9;
# Prompt for password with plymouth, if installed and running.
if plymouth --ping 2>/dev/null; then
plymouth ask-for-password \
- --prompt "$ply_prompt" --number-of-tries="$ply_tries" \
- --command="$ply_cmd"
+ --prompt "$ply_prompt" --number-of-tries="$ply_tries" | \
+ eval "$ply_cmd"
ret=$?
else
if [ "$tty_echo_off" = yes ]; then
stty_orig="$(stty -g)"
stty -echo
fi
i=1
while [ "$i" -le "$tty_tries" ]; do
[ -n "$tty_prompt" ] && \
printf "%s [%i/%i]:" "$tty_prompt" "$i" "$tty_tries" >&2
eval "$tty_cmd" && ret=0 && break
ret=$?
i=$((i+1))
[ -n "$tty_prompt" ] && printf '\n' >&2
done
unset i
[ "$tty_echo_off" = yes ] && stty "$stty_orig"
fi
} 9>/.console_lock
[ $ret -ne 0 ] && echo "Wrong password" >&2
return $ret
}
diff --git a/sys/contrib/openzfs/module/zfs/arc.c b/sys/contrib/openzfs/module/zfs/arc.c
index 04d275dd80f4..3484fff3b4d4 100644
--- a/sys/contrib/openzfs/module/zfs/arc.c
+++ b/sys/contrib/openzfs/module/zfs/arc.c
@@ -1,11047 +1,11048 @@
/*
* 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) 2018, Joyent, Inc.
* Copyright (c) 2011, 2020, Delphix. All rights reserved.
* Copyright (c) 2014, Saso Kiselkov. All rights reserved.
* Copyright (c) 2017, Nexenta Systems, Inc. All rights reserved.
* Copyright (c) 2019, loli10K . All rights reserved.
* Copyright (c) 2020, George Amanakis. All rights reserved.
* Copyright (c) 2019, Klara Inc.
* Copyright (c) 2019, Allan Jude
* Copyright (c) 2020, The FreeBSD Foundation [1]
*
* [1] Portions of this software were developed by Allan Jude
* under sponsorship from the FreeBSD Foundation.
*/
/*
* DVA-based Adjustable Replacement Cache
*
* While much of the theory of operation used here is
* based on the self-tuning, low overhead replacement cache
* presented by Megiddo and Modha at FAST 2003, there are some
* significant differences:
*
* 1. The Megiddo and Modha model assumes any page is evictable.
* Pages in its cache cannot be "locked" into memory. This makes
* the eviction algorithm simple: evict the last page in the list.
* This also make the performance characteristics easy to reason
* about. Our cache is not so simple. At any given moment, some
* subset of the blocks in the cache are un-evictable because we
* have handed out a reference to them. Blocks are only evictable
* when there are no external references active. This makes
* eviction far more problematic: we choose to evict the evictable
* blocks that are the "lowest" in the list.
*
* There are times when it is not possible to evict the requested
* space. In these circumstances we are unable to adjust the cache
* size. To prevent the cache growing unbounded at these times we
* implement a "cache throttle" that slows the flow of new data
* into the cache until we can make space available.
*
* 2. The Megiddo and Modha model assumes a fixed cache size.
* Pages are evicted when the cache is full and there is a cache
* miss. Our model has a variable sized cache. It grows with
* high use, but also tries to react to memory pressure from the
* operating system: decreasing its size when system memory is
* tight.
*
* 3. The Megiddo and Modha model assumes a fixed page size. All
* elements of the cache are therefore exactly the same size. So
* when adjusting the cache size following a cache miss, its simply
* a matter of choosing a single page to evict. In our model, we
* have variable sized cache blocks (ranging from 512 bytes to
* 128K bytes). We therefore choose a set of blocks to evict to make
* space for a cache miss that approximates as closely as possible
* the space used by the new block.
*
* See also: "ARC: A Self-Tuning, Low Overhead Replacement Cache"
* by N. Megiddo & D. Modha, FAST 2003
*/
/*
* The locking model:
*
* A new reference to a cache buffer can be obtained in two
* ways: 1) via a hash table lookup using the DVA as a key,
* or 2) via one of the ARC lists. The arc_read() interface
* uses method 1, while the internal ARC algorithms for
* adjusting the cache use method 2. We therefore provide two
* types of locks: 1) the hash table lock array, and 2) the
* ARC list locks.
*
* Buffers do not have their own mutexes, rather they rely on the
* hash table mutexes for the bulk of their protection (i.e. most
* fields in the arc_buf_hdr_t are protected by these mutexes).
*
* buf_hash_find() returns the appropriate mutex (held) when it
* locates the requested buffer in the hash table. It returns
* NULL for the mutex if the buffer was not in the table.
*
* buf_hash_remove() expects the appropriate hash mutex to be
* already held before it is invoked.
*
* Each ARC state also has a mutex which is used to protect the
* buffer list associated with the state. When attempting to
* obtain a hash table lock while holding an ARC list lock you
* must use: mutex_tryenter() to avoid deadlock. Also note that
* the active state mutex must be held before the ghost state mutex.
*
* It as also possible to register a callback which is run when the
* arc_meta_limit is reached and no buffers can be safely evicted. In
* this case the arc user should drop a reference on some arc buffers so
* they can be reclaimed and the arc_meta_limit honored. For example,
* when using the ZPL each dentry holds a references on a znode. These
* dentries must be pruned before the arc buffer holding the znode can
* be safely evicted.
*
* Note that the majority of the performance stats are manipulated
* with atomic operations.
*
* The L2ARC uses the l2ad_mtx on each vdev for the following:
*
* - L2ARC buflist creation
* - L2ARC buflist eviction
* - L2ARC write completion, which walks L2ARC buflists
* - ARC header destruction, as it removes from L2ARC buflists
* - ARC header release, as it removes from L2ARC buflists
*/
/*
* ARC operation:
*
* Every block that is in the ARC is tracked by an arc_buf_hdr_t structure.
* This structure can point either to a block that is still in the cache or to
* one that is only accessible in an L2 ARC device, or it can provide
* information about a block that was recently evicted. If a block is
* only accessible in the L2ARC, then the arc_buf_hdr_t only has enough
* information to retrieve it from the L2ARC device. This information is
* stored in the l2arc_buf_hdr_t sub-structure of the arc_buf_hdr_t. A block
* that is in this state cannot access the data directly.
*
* Blocks that are actively being referenced or have not been evicted
* are cached in the L1ARC. The L1ARC (l1arc_buf_hdr_t) is a structure within
* the arc_buf_hdr_t that will point to the data block in memory. A block can
* only be read by a consumer if it has an l1arc_buf_hdr_t. The L1ARC
* caches data in two ways -- in a list of ARC buffers (arc_buf_t) and
* also in the arc_buf_hdr_t's private physical data block pointer (b_pabd).
*
* The L1ARC's data pointer may or may not be uncompressed. The ARC has the
* ability to store the physical data (b_pabd) associated with the DVA of the
* arc_buf_hdr_t. Since the b_pabd is a copy of the on-disk physical block,
* it will match its on-disk compression characteristics. This behavior can be
* disabled by setting 'zfs_compressed_arc_enabled' to B_FALSE. When the
* compressed ARC functionality is disabled, the b_pabd will point to an
* uncompressed version of the on-disk data.
*
* Data in the L1ARC is not accessed by consumers of the ARC directly. Each
* arc_buf_hdr_t can have multiple ARC buffers (arc_buf_t) which reference it.
* Each ARC buffer (arc_buf_t) is being actively accessed by a specific ARC
* consumer. The ARC will provide references to this data and will keep it
* cached until it is no longer in use. The ARC caches only the L1ARC's physical
* data block and will evict any arc_buf_t that is no longer referenced. The
* amount of memory consumed by the arc_buf_ts' data buffers can be seen via the
* "overhead_size" kstat.
*
* Depending on the consumer, an arc_buf_t can be requested in uncompressed or
* compressed form. The typical case is that consumers will want uncompressed
* data, and when that happens a new data buffer is allocated where the data is
* decompressed for them to use. Currently the only consumer who wants
* compressed arc_buf_t's is "zfs send", when it streams data exactly as it
* exists on disk. When this happens, the arc_buf_t's data buffer is shared
* with the arc_buf_hdr_t.
*
* Here is a diagram showing an arc_buf_hdr_t referenced by two arc_buf_t's. The
* first one is owned by a compressed send consumer (and therefore references
* the same compressed data buffer as the arc_buf_hdr_t) and the second could be
* used by any other consumer (and has its own uncompressed copy of the data
* buffer).
*
* arc_buf_hdr_t
* +-----------+
* | fields |
* | common to |
* | L1- and |
* | L2ARC |
* +-----------+
* | l2arc_buf_hdr_t
* | |
* +-----------+
* | l1arc_buf_hdr_t
* | | arc_buf_t
* | b_buf +------------>+-----------+ arc_buf_t
* | b_pabd +-+ |b_next +---->+-----------+
* +-----------+ | |-----------| |b_next +-->NULL
* | |b_comp = T | +-----------+
* | |b_data +-+ |b_comp = F |
* | +-----------+ | |b_data +-+
* +->+------+ | +-----------+ |
* compressed | | | |
* data | |<--------------+ | uncompressed
* +------+ compressed, | data
* shared +-->+------+
* data | |
* | |
* +------+
*
* When a consumer reads a block, the ARC must first look to see if the
* arc_buf_hdr_t is cached. If the hdr is cached then the ARC allocates a new
* arc_buf_t and either copies uncompressed data into a new data buffer from an
* existing uncompressed arc_buf_t, decompresses the hdr's b_pabd buffer into a
* new data buffer, or shares the hdr's b_pabd buffer, depending on whether the
* hdr is compressed and the desired compression characteristics of the
* arc_buf_t consumer. If the arc_buf_t ends up sharing data with the
* arc_buf_hdr_t and both of them are uncompressed then the arc_buf_t must be
* the last buffer in the hdr's b_buf list, however a shared compressed buf can
* be anywhere in the hdr's list.
*
* The diagram below shows an example of an uncompressed ARC hdr that is
* sharing its data with an arc_buf_t (note that the shared uncompressed buf is
* the last element in the buf list):
*
* arc_buf_hdr_t
* +-----------+
* | |
* | |
* | |
* +-----------+
* l2arc_buf_hdr_t| |
* | |
* +-----------+
* l1arc_buf_hdr_t| |
* | | arc_buf_t (shared)
* | b_buf +------------>+---------+ arc_buf_t
* | | |b_next +---->+---------+
* | b_pabd +-+ |---------| |b_next +-->NULL
* +-----------+ | | | +---------+
* | |b_data +-+ | |
* | +---------+ | |b_data +-+
* +->+------+ | +---------+ |
* | | | |
* uncompressed | | | |
* data +------+ | |
* ^ +->+------+ |
* | uncompressed | | |
* | data | | |
* | +------+ |
* +---------------------------------+
*
* Writing to the ARC requires that the ARC first discard the hdr's b_pabd
* since the physical block is about to be rewritten. The new data contents
* will be contained in the arc_buf_t. As the I/O pipeline performs the write,
* it may compress the data before writing it to disk. The ARC will be called
* with the transformed data and will bcopy the transformed on-disk block into
* a newly allocated b_pabd. Writes are always done into buffers which have
* either been loaned (and hence are new and don't have other readers) or
* buffers which have been released (and hence have their own hdr, if there
* were originally other readers of the buf's original hdr). This ensures that
* the ARC only needs to update a single buf and its hdr after a write occurs.
*
* When the L2ARC is in use, it will also take advantage of the b_pabd. The
* L2ARC will always write the contents of b_pabd to the L2ARC. This means
* that when compressed ARC is enabled that the L2ARC blocks are identical
* to the on-disk block in the main data pool. This provides a significant
* advantage since the ARC can leverage the bp's checksum when reading from the
* L2ARC to determine if the contents are valid. However, if the compressed
* ARC is disabled, then the L2ARC's block must be transformed to look
* like the physical block in the main data pool before comparing the
* checksum and determining its validity.
*
* The L1ARC has a slightly different system for storing encrypted data.
* Raw (encrypted + possibly compressed) data has a few subtle differences from
* data that is just compressed. The biggest difference is that it is not
* possible to decrypt encrypted data (or vice-versa) if the keys aren't loaded.
* The other difference is that encryption cannot be treated as a suggestion.
* If a caller would prefer compressed data, but they actually wind up with
* uncompressed data the worst thing that could happen is there might be a
* performance hit. If the caller requests encrypted data, however, we must be
* sure they actually get it or else secret information could be leaked. Raw
* data is stored in hdr->b_crypt_hdr.b_rabd. An encrypted header, therefore,
* may have both an encrypted version and a decrypted version of its data at
* once. When a caller needs a raw arc_buf_t, it is allocated and the data is
* copied out of this header. To avoid complications with b_pabd, raw buffers
* cannot be shared.
*/
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#ifndef _KERNEL
/* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */
boolean_t arc_watch = B_FALSE;
#endif
/*
* This thread's job is to keep enough free memory in the system, by
* calling arc_kmem_reap_soon() plus arc_reduce_target_size(), which improves
* arc_available_memory().
*/
static zthr_t *arc_reap_zthr;
/*
* This thread's job is to keep arc_size under arc_c, by calling
* arc_evict(), which improves arc_is_overflowing().
*/
static zthr_t *arc_evict_zthr;
static kmutex_t arc_evict_lock;
static boolean_t arc_evict_needed = B_FALSE;
/*
* Count of bytes evicted since boot.
*/
static uint64_t arc_evict_count;
/*
* List of arc_evict_waiter_t's, representing threads waiting for the
* arc_evict_count to reach specific values.
*/
static list_t arc_evict_waiters;
/*
* When arc_is_overflowing(), arc_get_data_impl() waits for this percent of
* the requested amount of data to be evicted. For example, by default for
* every 2KB that's evicted, 1KB of it may be "reused" by a new allocation.
* Since this is above 100%, it ensures that progress is made towards getting
* arc_size under arc_c. Since this is finite, it ensures that allocations
* can still happen, even during the potentially long time that arc_size is
* more than arc_c.
*/
int zfs_arc_eviction_pct = 200;
/*
* The number of headers to evict in arc_evict_state_impl() before
* dropping the sublist lock and evicting from another sublist. A lower
* value means we're more likely to evict the "correct" header (i.e. the
* oldest header in the arc state), but comes with higher overhead
* (i.e. more invocations of arc_evict_state_impl()).
*/
int zfs_arc_evict_batch_limit = 10;
/* number of seconds before growing cache again */
int arc_grow_retry = 5;
/*
* Minimum time between calls to arc_kmem_reap_soon().
*/
int arc_kmem_cache_reap_retry_ms = 1000;
/* shift of arc_c for calculating overflow limit in arc_get_data_impl */
int zfs_arc_overflow_shift = 8;
/* shift of arc_c for calculating both min and max arc_p */
int arc_p_min_shift = 4;
/* log2(fraction of arc to reclaim) */
int arc_shrink_shift = 7;
/* percent of pagecache to reclaim arc to */
#ifdef _KERNEL
uint_t zfs_arc_pc_percent = 0;
#endif
/*
* log2(fraction of ARC which must be free to allow growing).
* I.e. If there is less than arc_c >> arc_no_grow_shift free memory,
* when reading a new block into the ARC, we will evict an equal-sized block
* from the ARC.
*
* This must be less than arc_shrink_shift, so that when we shrink the ARC,
* we will still not allow it to grow.
*/
int arc_no_grow_shift = 5;
/*
* minimum lifespan of a prefetch block in clock ticks
* (initialized in arc_init())
*/
static int arc_min_prefetch_ms;
static int arc_min_prescient_prefetch_ms;
/*
* If this percent of memory is free, don't throttle.
*/
int arc_lotsfree_percent = 10;
/*
* The arc has filled available memory and has now warmed up.
*/
boolean_t arc_warm;
/*
* These tunables are for performance analysis.
*/
unsigned long zfs_arc_max = 0;
unsigned long zfs_arc_min = 0;
unsigned long zfs_arc_meta_limit = 0;
unsigned long zfs_arc_meta_min = 0;
unsigned long zfs_arc_dnode_limit = 0;
unsigned long zfs_arc_dnode_reduce_percent = 10;
int zfs_arc_grow_retry = 0;
int zfs_arc_shrink_shift = 0;
int zfs_arc_p_min_shift = 0;
int zfs_arc_average_blocksize = 8 * 1024; /* 8KB */
/*
* ARC dirty data constraints for arc_tempreserve_space() throttle.
*/
unsigned long zfs_arc_dirty_limit_percent = 50; /* total dirty data limit */
unsigned long zfs_arc_anon_limit_percent = 25; /* anon block dirty limit */
unsigned long zfs_arc_pool_dirty_percent = 20; /* each pool's anon allowance */
/*
* Enable or disable compressed arc buffers.
*/
int zfs_compressed_arc_enabled = B_TRUE;
/*
* ARC will evict meta buffers that exceed arc_meta_limit. This
* tunable make arc_meta_limit adjustable for different workloads.
*/
unsigned long zfs_arc_meta_limit_percent = 75;
/*
* Percentage that can be consumed by dnodes of ARC meta buffers.
*/
unsigned long zfs_arc_dnode_limit_percent = 10;
/*
* These tunables are Linux specific
*/
unsigned long zfs_arc_sys_free = 0;
int zfs_arc_min_prefetch_ms = 0;
int zfs_arc_min_prescient_prefetch_ms = 0;
int zfs_arc_p_dampener_disable = 1;
int zfs_arc_meta_prune = 10000;
int zfs_arc_meta_strategy = ARC_STRATEGY_META_BALANCED;
int zfs_arc_meta_adjust_restarts = 4096;
int zfs_arc_lotsfree_percent = 10;
/* The 6 states: */
arc_state_t ARC_anon;
arc_state_t ARC_mru;
arc_state_t ARC_mru_ghost;
arc_state_t ARC_mfu;
arc_state_t ARC_mfu_ghost;
arc_state_t ARC_l2c_only;
arc_stats_t arc_stats = {
{ "hits", KSTAT_DATA_UINT64 },
{ "misses", KSTAT_DATA_UINT64 },
{ "demand_data_hits", KSTAT_DATA_UINT64 },
{ "demand_data_misses", KSTAT_DATA_UINT64 },
{ "demand_metadata_hits", KSTAT_DATA_UINT64 },
{ "demand_metadata_misses", KSTAT_DATA_UINT64 },
{ "prefetch_data_hits", KSTAT_DATA_UINT64 },
{ "prefetch_data_misses", KSTAT_DATA_UINT64 },
{ "prefetch_metadata_hits", KSTAT_DATA_UINT64 },
{ "prefetch_metadata_misses", KSTAT_DATA_UINT64 },
{ "mru_hits", KSTAT_DATA_UINT64 },
{ "mru_ghost_hits", KSTAT_DATA_UINT64 },
{ "mfu_hits", KSTAT_DATA_UINT64 },
{ "mfu_ghost_hits", KSTAT_DATA_UINT64 },
{ "deleted", KSTAT_DATA_UINT64 },
{ "mutex_miss", KSTAT_DATA_UINT64 },
{ "access_skip", KSTAT_DATA_UINT64 },
{ "evict_skip", KSTAT_DATA_UINT64 },
{ "evict_not_enough", KSTAT_DATA_UINT64 },
{ "evict_l2_cached", KSTAT_DATA_UINT64 },
{ "evict_l2_eligible", KSTAT_DATA_UINT64 },
{ "evict_l2_eligible_mfu", KSTAT_DATA_UINT64 },
{ "evict_l2_eligible_mru", KSTAT_DATA_UINT64 },
{ "evict_l2_ineligible", KSTAT_DATA_UINT64 },
{ "evict_l2_skip", KSTAT_DATA_UINT64 },
{ "hash_elements", KSTAT_DATA_UINT64 },
{ "hash_elements_max", KSTAT_DATA_UINT64 },
{ "hash_collisions", KSTAT_DATA_UINT64 },
{ "hash_chains", KSTAT_DATA_UINT64 },
{ "hash_chain_max", KSTAT_DATA_UINT64 },
{ "p", KSTAT_DATA_UINT64 },
{ "c", KSTAT_DATA_UINT64 },
{ "c_min", KSTAT_DATA_UINT64 },
{ "c_max", KSTAT_DATA_UINT64 },
{ "size", KSTAT_DATA_UINT64 },
{ "compressed_size", KSTAT_DATA_UINT64 },
{ "uncompressed_size", KSTAT_DATA_UINT64 },
{ "overhead_size", KSTAT_DATA_UINT64 },
{ "hdr_size", KSTAT_DATA_UINT64 },
{ "data_size", KSTAT_DATA_UINT64 },
{ "metadata_size", KSTAT_DATA_UINT64 },
{ "dbuf_size", KSTAT_DATA_UINT64 },
{ "dnode_size", KSTAT_DATA_UINT64 },
{ "bonus_size", KSTAT_DATA_UINT64 },
#if defined(COMPAT_FREEBSD11)
{ "other_size", KSTAT_DATA_UINT64 },
#endif
{ "anon_size", KSTAT_DATA_UINT64 },
{ "anon_evictable_data", KSTAT_DATA_UINT64 },
{ "anon_evictable_metadata", KSTAT_DATA_UINT64 },
{ "mru_size", KSTAT_DATA_UINT64 },
{ "mru_evictable_data", KSTAT_DATA_UINT64 },
{ "mru_evictable_metadata", KSTAT_DATA_UINT64 },
{ "mru_ghost_size", KSTAT_DATA_UINT64 },
{ "mru_ghost_evictable_data", KSTAT_DATA_UINT64 },
{ "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64 },
{ "mfu_size", KSTAT_DATA_UINT64 },
{ "mfu_evictable_data", KSTAT_DATA_UINT64 },
{ "mfu_evictable_metadata", KSTAT_DATA_UINT64 },
{ "mfu_ghost_size", KSTAT_DATA_UINT64 },
{ "mfu_ghost_evictable_data", KSTAT_DATA_UINT64 },
{ "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64 },
{ "l2_hits", KSTAT_DATA_UINT64 },
{ "l2_misses", KSTAT_DATA_UINT64 },
{ "l2_prefetch_asize", KSTAT_DATA_UINT64 },
{ "l2_mru_asize", KSTAT_DATA_UINT64 },
{ "l2_mfu_asize", KSTAT_DATA_UINT64 },
{ "l2_bufc_data_asize", KSTAT_DATA_UINT64 },
{ "l2_bufc_metadata_asize", KSTAT_DATA_UINT64 },
{ "l2_feeds", KSTAT_DATA_UINT64 },
{ "l2_rw_clash", KSTAT_DATA_UINT64 },
{ "l2_read_bytes", KSTAT_DATA_UINT64 },
{ "l2_write_bytes", KSTAT_DATA_UINT64 },
{ "l2_writes_sent", KSTAT_DATA_UINT64 },
{ "l2_writes_done", KSTAT_DATA_UINT64 },
{ "l2_writes_error", KSTAT_DATA_UINT64 },
{ "l2_writes_lock_retry", KSTAT_DATA_UINT64 },
{ "l2_evict_lock_retry", KSTAT_DATA_UINT64 },
{ "l2_evict_reading", KSTAT_DATA_UINT64 },
{ "l2_evict_l1cached", KSTAT_DATA_UINT64 },
{ "l2_free_on_write", KSTAT_DATA_UINT64 },
{ "l2_abort_lowmem", KSTAT_DATA_UINT64 },
{ "l2_cksum_bad", KSTAT_DATA_UINT64 },
{ "l2_io_error", KSTAT_DATA_UINT64 },
{ "l2_size", KSTAT_DATA_UINT64 },
{ "l2_asize", KSTAT_DATA_UINT64 },
{ "l2_hdr_size", KSTAT_DATA_UINT64 },
{ "l2_log_blk_writes", KSTAT_DATA_UINT64 },
{ "l2_log_blk_avg_asize", KSTAT_DATA_UINT64 },
{ "l2_log_blk_asize", KSTAT_DATA_UINT64 },
{ "l2_log_blk_count", KSTAT_DATA_UINT64 },
{ "l2_data_to_meta_ratio", KSTAT_DATA_UINT64 },
{ "l2_rebuild_success", KSTAT_DATA_UINT64 },
{ "l2_rebuild_unsupported", KSTAT_DATA_UINT64 },
{ "l2_rebuild_io_errors", KSTAT_DATA_UINT64 },
{ "l2_rebuild_dh_errors", KSTAT_DATA_UINT64 },
{ "l2_rebuild_cksum_lb_errors", KSTAT_DATA_UINT64 },
{ "l2_rebuild_lowmem", KSTAT_DATA_UINT64 },
{ "l2_rebuild_size", KSTAT_DATA_UINT64 },
{ "l2_rebuild_asize", KSTAT_DATA_UINT64 },
{ "l2_rebuild_bufs", KSTAT_DATA_UINT64 },
{ "l2_rebuild_bufs_precached", KSTAT_DATA_UINT64 },
{ "l2_rebuild_log_blks", KSTAT_DATA_UINT64 },
{ "memory_throttle_count", KSTAT_DATA_UINT64 },
{ "memory_direct_count", KSTAT_DATA_UINT64 },
{ "memory_indirect_count", KSTAT_DATA_UINT64 },
{ "memory_all_bytes", KSTAT_DATA_UINT64 },
{ "memory_free_bytes", KSTAT_DATA_UINT64 },
{ "memory_available_bytes", KSTAT_DATA_INT64 },
{ "arc_no_grow", KSTAT_DATA_UINT64 },
{ "arc_tempreserve", KSTAT_DATA_UINT64 },
{ "arc_loaned_bytes", KSTAT_DATA_UINT64 },
{ "arc_prune", KSTAT_DATA_UINT64 },
{ "arc_meta_used", KSTAT_DATA_UINT64 },
{ "arc_meta_limit", KSTAT_DATA_UINT64 },
{ "arc_dnode_limit", KSTAT_DATA_UINT64 },
{ "arc_meta_max", KSTAT_DATA_UINT64 },
{ "arc_meta_min", KSTAT_DATA_UINT64 },
{ "async_upgrade_sync", KSTAT_DATA_UINT64 },
{ "demand_hit_predictive_prefetch", KSTAT_DATA_UINT64 },
{ "demand_hit_prescient_prefetch", KSTAT_DATA_UINT64 },
{ "arc_need_free", KSTAT_DATA_UINT64 },
{ "arc_sys_free", KSTAT_DATA_UINT64 },
{ "arc_raw_size", KSTAT_DATA_UINT64 },
{ "cached_only_in_progress", KSTAT_DATA_UINT64 },
{ "abd_chunk_waste_size", KSTAT_DATA_UINT64 },
};
arc_sums_t arc_sums;
#define ARCSTAT_MAX(stat, val) { \
uint64_t m; \
while ((val) > (m = arc_stats.stat.value.ui64) && \
(m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \
continue; \
}
/*
* We define a macro to allow ARC hits/misses to be easily broken down by
* two separate conditions, giving a total of four different subtypes for
* each of hits and misses (so eight statistics total).
*/
#define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \
if (cond1) { \
if (cond2) { \
ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \
} else { \
ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \
} \
} else { \
if (cond2) { \
ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \
} else { \
ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\
} \
}
/*
* This macro allows us to use kstats as floating averages. Each time we
* update this kstat, we first factor it and the update value by
* ARCSTAT_AVG_FACTOR to shrink the new value's contribution to the overall
* average. This macro assumes that integer loads and stores are atomic, but
* is not safe for multiple writers updating the kstat in parallel (only the
* last writer's update will remain).
*/
#define ARCSTAT_F_AVG_FACTOR 3
#define ARCSTAT_F_AVG(stat, value) \
do { \
uint64_t x = ARCSTAT(stat); \
x = x - x / ARCSTAT_F_AVG_FACTOR + \
(value) / ARCSTAT_F_AVG_FACTOR; \
ARCSTAT(stat) = x; \
_NOTE(CONSTCOND) \
} while (0)
kstat_t *arc_ksp;
static arc_state_t *arc_anon;
static arc_state_t *arc_mru_ghost;
static arc_state_t *arc_mfu_ghost;
static arc_state_t *arc_l2c_only;
arc_state_t *arc_mru;
arc_state_t *arc_mfu;
/*
* There are several ARC variables that are critical to export as kstats --
* but we don't want to have to grovel around in the kstat whenever we wish to
* manipulate them. For these variables, we therefore define them to be in
* terms of the statistic variable. This assures that we are not introducing
* the possibility of inconsistency by having shadow copies of the variables,
* while still allowing the code to be readable.
*/
#define arc_tempreserve ARCSTAT(arcstat_tempreserve)
#define arc_loaned_bytes ARCSTAT(arcstat_loaned_bytes)
#define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */
/* max size for dnodes */
#define arc_dnode_size_limit ARCSTAT(arcstat_dnode_limit)
#define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */
#define arc_need_free ARCSTAT(arcstat_need_free) /* waiting to be evicted */
hrtime_t arc_growtime;
list_t arc_prune_list;
kmutex_t arc_prune_mtx;
taskq_t *arc_prune_taskq;
#define GHOST_STATE(state) \
((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \
(state) == arc_l2c_only)
#define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE)
#define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS)
#define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR)
#define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH)
#define HDR_PRESCIENT_PREFETCH(hdr) \
((hdr)->b_flags & ARC_FLAG_PRESCIENT_PREFETCH)
#define HDR_COMPRESSION_ENABLED(hdr) \
((hdr)->b_flags & ARC_FLAG_COMPRESSED_ARC)
#define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE)
#define HDR_L2_READING(hdr) \
(((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \
((hdr)->b_flags & ARC_FLAG_HAS_L2HDR))
#define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING)
#define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED)
#define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD)
#define HDR_PROTECTED(hdr) ((hdr)->b_flags & ARC_FLAG_PROTECTED)
#define HDR_NOAUTH(hdr) ((hdr)->b_flags & ARC_FLAG_NOAUTH)
#define HDR_SHARED_DATA(hdr) ((hdr)->b_flags & ARC_FLAG_SHARED_DATA)
#define HDR_ISTYPE_METADATA(hdr) \
((hdr)->b_flags & ARC_FLAG_BUFC_METADATA)
#define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr))
#define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR)
#define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)
#define HDR_HAS_RABD(hdr) \
(HDR_HAS_L1HDR(hdr) && HDR_PROTECTED(hdr) && \
(hdr)->b_crypt_hdr.b_rabd != NULL)
#define HDR_ENCRYPTED(hdr) \
(HDR_PROTECTED(hdr) && DMU_OT_IS_ENCRYPTED((hdr)->b_crypt_hdr.b_ot))
#define HDR_AUTHENTICATED(hdr) \
(HDR_PROTECTED(hdr) && !DMU_OT_IS_ENCRYPTED((hdr)->b_crypt_hdr.b_ot))
/* For storing compression mode in b_flags */
#define HDR_COMPRESS_OFFSET (highbit64(ARC_FLAG_COMPRESS_0) - 1)
#define HDR_GET_COMPRESS(hdr) ((enum zio_compress)BF32_GET((hdr)->b_flags, \
HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS))
#define HDR_SET_COMPRESS(hdr, cmp) BF32_SET((hdr)->b_flags, \
HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS, (cmp));
#define ARC_BUF_LAST(buf) ((buf)->b_next == NULL)
#define ARC_BUF_SHARED(buf) ((buf)->b_flags & ARC_BUF_FLAG_SHARED)
#define ARC_BUF_COMPRESSED(buf) ((buf)->b_flags & ARC_BUF_FLAG_COMPRESSED)
#define ARC_BUF_ENCRYPTED(buf) ((buf)->b_flags & ARC_BUF_FLAG_ENCRYPTED)
/*
* Other sizes
*/
#define HDR_FULL_CRYPT_SIZE ((int64_t)sizeof (arc_buf_hdr_t))
#define HDR_FULL_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_crypt_hdr))
#define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr))
/*
* Hash table routines
*/
#define HT_LOCK_ALIGN 64
#define HT_LOCK_PAD (P2NPHASE(sizeof (kmutex_t), (HT_LOCK_ALIGN)))
struct ht_lock {
kmutex_t ht_lock;
#ifdef _KERNEL
unsigned char pad[HT_LOCK_PAD];
#endif
};
#define BUF_LOCKS 8192
typedef struct buf_hash_table {
uint64_t ht_mask;
arc_buf_hdr_t **ht_table;
struct ht_lock ht_locks[BUF_LOCKS];
} buf_hash_table_t;
static buf_hash_table_t buf_hash_table;
#define BUF_HASH_INDEX(spa, dva, birth) \
(buf_hash(spa, dva, birth) & buf_hash_table.ht_mask)
#define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)])
#define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock))
#define HDR_LOCK(hdr) \
(BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth)))
uint64_t zfs_crc64_table[256];
/*
* Level 2 ARC
*/
#define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */
#define L2ARC_HEADROOM 2 /* num of writes */
/*
* If we discover during ARC scan any buffers to be compressed, we boost
* our headroom for the next scanning cycle by this percentage multiple.
*/
#define L2ARC_HEADROOM_BOOST 200
#define L2ARC_FEED_SECS 1 /* caching interval secs */
#define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */
/*
* We can feed L2ARC from two states of ARC buffers, mru and mfu,
* and each of the state has two types: data and metadata.
*/
#define L2ARC_FEED_TYPES 4
/* L2ARC Performance Tunables */
unsigned long l2arc_write_max = L2ARC_WRITE_SIZE; /* def max write size */
unsigned long l2arc_write_boost = L2ARC_WRITE_SIZE; /* extra warmup write */
unsigned long l2arc_headroom = L2ARC_HEADROOM; /* # of dev writes */
unsigned long l2arc_headroom_boost = L2ARC_HEADROOM_BOOST;
unsigned long l2arc_feed_secs = L2ARC_FEED_SECS; /* interval seconds */
unsigned long l2arc_feed_min_ms = L2ARC_FEED_MIN_MS; /* min interval msecs */
int l2arc_noprefetch = B_TRUE; /* don't cache prefetch bufs */
int l2arc_feed_again = B_TRUE; /* turbo warmup */
int l2arc_norw = B_FALSE; /* no reads during writes */
int l2arc_meta_percent = 33; /* limit on headers size */
/*
* L2ARC Internals
*/
static list_t L2ARC_dev_list; /* device list */
static list_t *l2arc_dev_list; /* device list pointer */
static kmutex_t l2arc_dev_mtx; /* device list mutex */
static l2arc_dev_t *l2arc_dev_last; /* last device used */
static list_t L2ARC_free_on_write; /* free after write buf list */
static list_t *l2arc_free_on_write; /* free after write list ptr */
static kmutex_t l2arc_free_on_write_mtx; /* mutex for list */
static uint64_t l2arc_ndev; /* number of devices */
typedef struct l2arc_read_callback {
arc_buf_hdr_t *l2rcb_hdr; /* read header */
blkptr_t l2rcb_bp; /* original blkptr */
zbookmark_phys_t l2rcb_zb; /* original bookmark */
int l2rcb_flags; /* original flags */
abd_t *l2rcb_abd; /* temporary buffer */
} l2arc_read_callback_t;
typedef struct l2arc_data_free {
/* protected by l2arc_free_on_write_mtx */
abd_t *l2df_abd;
size_t l2df_size;
arc_buf_contents_t l2df_type;
list_node_t l2df_list_node;
} l2arc_data_free_t;
typedef enum arc_fill_flags {
ARC_FILL_LOCKED = 1 << 0, /* hdr lock is held */
ARC_FILL_COMPRESSED = 1 << 1, /* fill with compressed data */
ARC_FILL_ENCRYPTED = 1 << 2, /* fill with encrypted data */
ARC_FILL_NOAUTH = 1 << 3, /* don't attempt to authenticate */
ARC_FILL_IN_PLACE = 1 << 4 /* fill in place (special case) */
} arc_fill_flags_t;
static kmutex_t l2arc_feed_thr_lock;
static kcondvar_t l2arc_feed_thr_cv;
static uint8_t l2arc_thread_exit;
static kmutex_t l2arc_rebuild_thr_lock;
static kcondvar_t l2arc_rebuild_thr_cv;
enum arc_hdr_alloc_flags {
ARC_HDR_ALLOC_RDATA = 0x1,
ARC_HDR_DO_ADAPT = 0x2,
};
static abd_t *arc_get_data_abd(arc_buf_hdr_t *, uint64_t, void *, boolean_t);
static void *arc_get_data_buf(arc_buf_hdr_t *, uint64_t, void *);
static void arc_get_data_impl(arc_buf_hdr_t *, uint64_t, void *, boolean_t);
static void arc_free_data_abd(arc_buf_hdr_t *, abd_t *, uint64_t, void *);
static void arc_free_data_buf(arc_buf_hdr_t *, void *, uint64_t, void *);
static void arc_free_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag);
static void arc_hdr_free_abd(arc_buf_hdr_t *, boolean_t);
static void arc_hdr_alloc_abd(arc_buf_hdr_t *, int);
static void arc_access(arc_buf_hdr_t *, kmutex_t *);
static void arc_buf_watch(arc_buf_t *);
static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *);
static uint32_t arc_bufc_to_flags(arc_buf_contents_t);
static inline void arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags);
static inline void arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags);
static boolean_t l2arc_write_eligible(uint64_t, arc_buf_hdr_t *);
static void l2arc_read_done(zio_t *);
static void l2arc_do_free_on_write(void);
static void l2arc_hdr_arcstats_update(arc_buf_hdr_t *hdr, boolean_t incr,
boolean_t state_only);
#define l2arc_hdr_arcstats_increment(hdr) \
l2arc_hdr_arcstats_update((hdr), B_TRUE, B_FALSE)
#define l2arc_hdr_arcstats_decrement(hdr) \
l2arc_hdr_arcstats_update((hdr), B_FALSE, B_FALSE)
#define l2arc_hdr_arcstats_increment_state(hdr) \
l2arc_hdr_arcstats_update((hdr), B_TRUE, B_TRUE)
#define l2arc_hdr_arcstats_decrement_state(hdr) \
l2arc_hdr_arcstats_update((hdr), B_FALSE, B_TRUE)
/*
* l2arc_mfuonly : A ZFS module parameter that controls whether only MFU
* metadata and data are cached from ARC into L2ARC.
*/
int l2arc_mfuonly = 0;
/*
* L2ARC TRIM
* l2arc_trim_ahead : A ZFS module parameter that controls how much ahead of
* the current write size (l2arc_write_max) we should TRIM if we
* have filled the device. It is defined as a percentage of the
* write size. If set to 100 we trim twice the space required to
* accommodate upcoming writes. A minimum of 64MB will be trimmed.
* It also enables TRIM of the whole L2ARC device upon creation or
* addition to an existing pool or if the header of the device is
* invalid upon importing a pool or onlining a cache device. The
* default is 0, which disables TRIM on L2ARC altogether as it can
* put significant stress on the underlying storage devices. This
* will vary depending of how well the specific device handles
* these commands.
*/
unsigned long l2arc_trim_ahead = 0;
/*
* Performance tuning of L2ARC persistence:
*
* l2arc_rebuild_enabled : A ZFS module parameter that controls whether adding
* an L2ARC device (either at pool import or later) will attempt
* to rebuild L2ARC buffer contents.
* l2arc_rebuild_blocks_min_l2size : A ZFS module parameter that controls
* whether log blocks are written to the L2ARC device. If the L2ARC
* device is less than 1GB, the amount of data l2arc_evict()
* evicts is significant compared to the amount of restored L2ARC
* data. In this case do not write log blocks in L2ARC in order
* not to waste space.
*/
int l2arc_rebuild_enabled = B_TRUE;
unsigned long l2arc_rebuild_blocks_min_l2size = 1024 * 1024 * 1024;
/* L2ARC persistence rebuild control routines. */
void l2arc_rebuild_vdev(vdev_t *vd, boolean_t reopen);
static void l2arc_dev_rebuild_thread(void *arg);
static int l2arc_rebuild(l2arc_dev_t *dev);
/* L2ARC persistence read I/O routines. */
static int l2arc_dev_hdr_read(l2arc_dev_t *dev);
static int l2arc_log_blk_read(l2arc_dev_t *dev,
const l2arc_log_blkptr_t *this_lp, const l2arc_log_blkptr_t *next_lp,
l2arc_log_blk_phys_t *this_lb, l2arc_log_blk_phys_t *next_lb,
zio_t *this_io, zio_t **next_io);
static zio_t *l2arc_log_blk_fetch(vdev_t *vd,
const l2arc_log_blkptr_t *lp, l2arc_log_blk_phys_t *lb);
static void l2arc_log_blk_fetch_abort(zio_t *zio);
/* L2ARC persistence block restoration routines. */
static void l2arc_log_blk_restore(l2arc_dev_t *dev,
const l2arc_log_blk_phys_t *lb, uint64_t lb_asize);
static void l2arc_hdr_restore(const l2arc_log_ent_phys_t *le,
l2arc_dev_t *dev);
/* L2ARC persistence write I/O routines. */
static void l2arc_log_blk_commit(l2arc_dev_t *dev, zio_t *pio,
l2arc_write_callback_t *cb);
/* L2ARC persistence auxiliary routines. */
boolean_t l2arc_log_blkptr_valid(l2arc_dev_t *dev,
const l2arc_log_blkptr_t *lbp);
static boolean_t l2arc_log_blk_insert(l2arc_dev_t *dev,
const arc_buf_hdr_t *ab);
boolean_t l2arc_range_check_overlap(uint64_t bottom,
uint64_t top, uint64_t check);
static void l2arc_blk_fetch_done(zio_t *zio);
static inline uint64_t
l2arc_log_blk_overhead(uint64_t write_sz, l2arc_dev_t *dev);
/*
* We use Cityhash for this. It's fast, and has good hash properties without
* requiring any large static buffers.
*/
static uint64_t
buf_hash(uint64_t spa, const dva_t *dva, uint64_t birth)
{
return (cityhash4(spa, dva->dva_word[0], dva->dva_word[1], birth));
}
#define HDR_EMPTY(hdr) \
((hdr)->b_dva.dva_word[0] == 0 && \
(hdr)->b_dva.dva_word[1] == 0)
#define HDR_EMPTY_OR_LOCKED(hdr) \
(HDR_EMPTY(hdr) || MUTEX_HELD(HDR_LOCK(hdr)))
#define HDR_EQUAL(spa, dva, birth, hdr) \
((hdr)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \
((hdr)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \
((hdr)->b_birth == birth) && ((hdr)->b_spa == spa)
static void
buf_discard_identity(arc_buf_hdr_t *hdr)
{
hdr->b_dva.dva_word[0] = 0;
hdr->b_dva.dva_word[1] = 0;
hdr->b_birth = 0;
}
static arc_buf_hdr_t *
buf_hash_find(uint64_t spa, const blkptr_t *bp, kmutex_t **lockp)
{
const dva_t *dva = BP_IDENTITY(bp);
uint64_t birth = BP_PHYSICAL_BIRTH(bp);
uint64_t idx = BUF_HASH_INDEX(spa, dva, birth);
kmutex_t *hash_lock = BUF_HASH_LOCK(idx);
arc_buf_hdr_t *hdr;
mutex_enter(hash_lock);
for (hdr = buf_hash_table.ht_table[idx]; hdr != NULL;
hdr = hdr->b_hash_next) {
if (HDR_EQUAL(spa, dva, birth, hdr)) {
*lockp = hash_lock;
return (hdr);
}
}
mutex_exit(hash_lock);
*lockp = NULL;
return (NULL);
}
/*
* Insert an entry into the hash table. If there is already an element
* equal to elem in the hash table, then the already existing element
* will be returned and the new element will not be inserted.
* Otherwise returns NULL.
* If lockp == NULL, the caller is assumed to already hold the hash lock.
*/
static arc_buf_hdr_t *
buf_hash_insert(arc_buf_hdr_t *hdr, kmutex_t **lockp)
{
uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth);
kmutex_t *hash_lock = BUF_HASH_LOCK(idx);
arc_buf_hdr_t *fhdr;
uint32_t i;
ASSERT(!DVA_IS_EMPTY(&hdr->b_dva));
ASSERT(hdr->b_birth != 0);
ASSERT(!HDR_IN_HASH_TABLE(hdr));
if (lockp != NULL) {
*lockp = hash_lock;
mutex_enter(hash_lock);
} else {
ASSERT(MUTEX_HELD(hash_lock));
}
for (fhdr = buf_hash_table.ht_table[idx], i = 0; fhdr != NULL;
fhdr = fhdr->b_hash_next, i++) {
if (HDR_EQUAL(hdr->b_spa, &hdr->b_dva, hdr->b_birth, fhdr))
return (fhdr);
}
hdr->b_hash_next = buf_hash_table.ht_table[idx];
buf_hash_table.ht_table[idx] = hdr;
arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE);
/* collect some hash table performance data */
if (i > 0) {
ARCSTAT_BUMP(arcstat_hash_collisions);
if (i == 1)
ARCSTAT_BUMP(arcstat_hash_chains);
ARCSTAT_MAX(arcstat_hash_chain_max, i);
}
uint64_t he = atomic_inc_64_nv(
&arc_stats.arcstat_hash_elements.value.ui64);
ARCSTAT_MAX(arcstat_hash_elements_max, he);
return (NULL);
}
static void
buf_hash_remove(arc_buf_hdr_t *hdr)
{
arc_buf_hdr_t *fhdr, **hdrp;
uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth);
ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx)));
ASSERT(HDR_IN_HASH_TABLE(hdr));
hdrp = &buf_hash_table.ht_table[idx];
while ((fhdr = *hdrp) != hdr) {
ASSERT3P(fhdr, !=, NULL);
hdrp = &fhdr->b_hash_next;
}
*hdrp = hdr->b_hash_next;
hdr->b_hash_next = NULL;
arc_hdr_clear_flags(hdr, ARC_FLAG_IN_HASH_TABLE);
/* collect some hash table performance data */
atomic_dec_64(&arc_stats.arcstat_hash_elements.value.ui64);
if (buf_hash_table.ht_table[idx] &&
buf_hash_table.ht_table[idx]->b_hash_next == NULL)
ARCSTAT_BUMPDOWN(arcstat_hash_chains);
}
/*
* Global data structures and functions for the buf kmem cache.
*/
static kmem_cache_t *hdr_full_cache;
static kmem_cache_t *hdr_full_crypt_cache;
static kmem_cache_t *hdr_l2only_cache;
static kmem_cache_t *buf_cache;
static void
buf_fini(void)
{
int i;
#if defined(_KERNEL)
/*
* Large allocations which do not require contiguous pages
* should be using vmem_free() in the linux kernel\
*/
vmem_free(buf_hash_table.ht_table,
(buf_hash_table.ht_mask + 1) * sizeof (void *));
#else
kmem_free(buf_hash_table.ht_table,
(buf_hash_table.ht_mask + 1) * sizeof (void *));
#endif
for (i = 0; i < BUF_LOCKS; i++)
mutex_destroy(&buf_hash_table.ht_locks[i].ht_lock);
kmem_cache_destroy(hdr_full_cache);
kmem_cache_destroy(hdr_full_crypt_cache);
kmem_cache_destroy(hdr_l2only_cache);
kmem_cache_destroy(buf_cache);
}
/*
* Constructor callback - called when the cache is empty
* and a new buf is requested.
*/
/* ARGSUSED */
static int
hdr_full_cons(void *vbuf, void *unused, int kmflag)
{
arc_buf_hdr_t *hdr = vbuf;
bzero(hdr, HDR_FULL_SIZE);
hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS;
cv_init(&hdr->b_l1hdr.b_cv, NULL, CV_DEFAULT, NULL);
zfs_refcount_create(&hdr->b_l1hdr.b_refcnt);
mutex_init(&hdr->b_l1hdr.b_freeze_lock, NULL, MUTEX_DEFAULT, NULL);
list_link_init(&hdr->b_l1hdr.b_arc_node);
list_link_init(&hdr->b_l2hdr.b_l2node);
multilist_link_init(&hdr->b_l1hdr.b_arc_node);
arc_space_consume(HDR_FULL_SIZE, ARC_SPACE_HDRS);
return (0);
}
/* ARGSUSED */
static int
hdr_full_crypt_cons(void *vbuf, void *unused, int kmflag)
{
arc_buf_hdr_t *hdr = vbuf;
hdr_full_cons(vbuf, unused, kmflag);
bzero(&hdr->b_crypt_hdr, sizeof (hdr->b_crypt_hdr));
arc_space_consume(sizeof (hdr->b_crypt_hdr), ARC_SPACE_HDRS);
return (0);
}
/* ARGSUSED */
static int
hdr_l2only_cons(void *vbuf, void *unused, int kmflag)
{
arc_buf_hdr_t *hdr = vbuf;
bzero(hdr, HDR_L2ONLY_SIZE);
arc_space_consume(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS);
return (0);
}
/* ARGSUSED */
static int
buf_cons(void *vbuf, void *unused, int kmflag)
{
arc_buf_t *buf = vbuf;
bzero(buf, sizeof (arc_buf_t));
mutex_init(&buf->b_evict_lock, NULL, MUTEX_DEFAULT, NULL);
arc_space_consume(sizeof (arc_buf_t), ARC_SPACE_HDRS);
return (0);
}
/*
* Destructor callback - called when a cached buf is
* no longer required.
*/
/* ARGSUSED */
static void
hdr_full_dest(void *vbuf, void *unused)
{
arc_buf_hdr_t *hdr = vbuf;
ASSERT(HDR_EMPTY(hdr));
cv_destroy(&hdr->b_l1hdr.b_cv);
zfs_refcount_destroy(&hdr->b_l1hdr.b_refcnt);
mutex_destroy(&hdr->b_l1hdr.b_freeze_lock);
ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node));
arc_space_return(HDR_FULL_SIZE, ARC_SPACE_HDRS);
}
/* ARGSUSED */
static void
hdr_full_crypt_dest(void *vbuf, void *unused)
{
arc_buf_hdr_t *hdr = vbuf;
hdr_full_dest(vbuf, unused);
arc_space_return(sizeof (hdr->b_crypt_hdr), ARC_SPACE_HDRS);
}
/* ARGSUSED */
static void
hdr_l2only_dest(void *vbuf, void *unused)
{
arc_buf_hdr_t *hdr __maybe_unused = vbuf;
ASSERT(HDR_EMPTY(hdr));
arc_space_return(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS);
}
/* ARGSUSED */
static void
buf_dest(void *vbuf, void *unused)
{
arc_buf_t *buf = vbuf;
mutex_destroy(&buf->b_evict_lock);
arc_space_return(sizeof (arc_buf_t), ARC_SPACE_HDRS);
}
static void
buf_init(void)
{
uint64_t *ct = NULL;
uint64_t hsize = 1ULL << 12;
int i, j;
/*
* The hash table is big enough to fill all of physical memory
* with an average block size of zfs_arc_average_blocksize (default 8K).
* By default, the table will take up
* totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers).
*/
while (hsize * zfs_arc_average_blocksize < arc_all_memory())
hsize <<= 1;
retry:
buf_hash_table.ht_mask = hsize - 1;
#if defined(_KERNEL)
/*
* Large allocations which do not require contiguous pages
* should be using vmem_alloc() in the linux kernel
*/
buf_hash_table.ht_table =
vmem_zalloc(hsize * sizeof (void*), KM_SLEEP);
#else
buf_hash_table.ht_table =
kmem_zalloc(hsize * sizeof (void*), KM_NOSLEEP);
#endif
if (buf_hash_table.ht_table == NULL) {
ASSERT(hsize > (1ULL << 8));
hsize >>= 1;
goto retry;
}
hdr_full_cache = kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE,
0, hdr_full_cons, hdr_full_dest, NULL, NULL, NULL, 0);
hdr_full_crypt_cache = kmem_cache_create("arc_buf_hdr_t_full_crypt",
HDR_FULL_CRYPT_SIZE, 0, hdr_full_crypt_cons, hdr_full_crypt_dest,
NULL, NULL, NULL, 0);
hdr_l2only_cache = kmem_cache_create("arc_buf_hdr_t_l2only",
HDR_L2ONLY_SIZE, 0, hdr_l2only_cons, hdr_l2only_dest, NULL,
NULL, NULL, 0);
buf_cache = kmem_cache_create("arc_buf_t", sizeof (arc_buf_t),
0, buf_cons, buf_dest, NULL, NULL, NULL, 0);
for (i = 0; i < 256; i++)
for (ct = zfs_crc64_table + i, *ct = i, j = 8; j > 0; j--)
*ct = (*ct >> 1) ^ (-(*ct & 1) & ZFS_CRC64_POLY);
for (i = 0; i < BUF_LOCKS; i++) {
mutex_init(&buf_hash_table.ht_locks[i].ht_lock,
NULL, MUTEX_DEFAULT, NULL);
}
}
#define ARC_MINTIME (hz>>4) /* 62 ms */
/*
* This is the size that the buf occupies in memory. If the buf is compressed,
* it will correspond to the compressed size. You should use this method of
* getting the buf size unless you explicitly need the logical size.
*/
uint64_t
arc_buf_size(arc_buf_t *buf)
{
return (ARC_BUF_COMPRESSED(buf) ?
HDR_GET_PSIZE(buf->b_hdr) : HDR_GET_LSIZE(buf->b_hdr));
}
uint64_t
arc_buf_lsize(arc_buf_t *buf)
{
return (HDR_GET_LSIZE(buf->b_hdr));
}
/*
* This function will return B_TRUE if the buffer is encrypted in memory.
* This buffer can be decrypted by calling arc_untransform().
*/
boolean_t
arc_is_encrypted(arc_buf_t *buf)
{
return (ARC_BUF_ENCRYPTED(buf) != 0);
}
/*
* Returns B_TRUE if the buffer represents data that has not had its MAC
* verified yet.
*/
boolean_t
arc_is_unauthenticated(arc_buf_t *buf)
{
return (HDR_NOAUTH(buf->b_hdr) != 0);
}
void
arc_get_raw_params(arc_buf_t *buf, boolean_t *byteorder, uint8_t *salt,
uint8_t *iv, uint8_t *mac)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
ASSERT(HDR_PROTECTED(hdr));
bcopy(hdr->b_crypt_hdr.b_salt, salt, ZIO_DATA_SALT_LEN);
bcopy(hdr->b_crypt_hdr.b_iv, iv, ZIO_DATA_IV_LEN);
bcopy(hdr->b_crypt_hdr.b_mac, mac, ZIO_DATA_MAC_LEN);
*byteorder = (hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS) ?
ZFS_HOST_BYTEORDER : !ZFS_HOST_BYTEORDER;
}
/*
* Indicates how this buffer is compressed in memory. If it is not compressed
* the value will be ZIO_COMPRESS_OFF. It can be made normally readable with
* arc_untransform() as long as it is also unencrypted.
*/
enum zio_compress
arc_get_compression(arc_buf_t *buf)
{
return (ARC_BUF_COMPRESSED(buf) ?
HDR_GET_COMPRESS(buf->b_hdr) : ZIO_COMPRESS_OFF);
}
/*
* Return the compression algorithm used to store this data in the ARC. If ARC
* compression is enabled or this is an encrypted block, this will be the same
* as what's used to store it on-disk. Otherwise, this will be ZIO_COMPRESS_OFF.
*/
static inline enum zio_compress
arc_hdr_get_compress(arc_buf_hdr_t *hdr)
{
return (HDR_COMPRESSION_ENABLED(hdr) ?
HDR_GET_COMPRESS(hdr) : ZIO_COMPRESS_OFF);
}
uint8_t
arc_get_complevel(arc_buf_t *buf)
{
return (buf->b_hdr->b_complevel);
}
static inline boolean_t
arc_buf_is_shared(arc_buf_t *buf)
{
boolean_t shared = (buf->b_data != NULL &&
buf->b_hdr->b_l1hdr.b_pabd != NULL &&
abd_is_linear(buf->b_hdr->b_l1hdr.b_pabd) &&
buf->b_data == abd_to_buf(buf->b_hdr->b_l1hdr.b_pabd));
IMPLY(shared, HDR_SHARED_DATA(buf->b_hdr));
IMPLY(shared, ARC_BUF_SHARED(buf));
IMPLY(shared, ARC_BUF_COMPRESSED(buf) || ARC_BUF_LAST(buf));
/*
* It would be nice to assert arc_can_share() too, but the "hdr isn't
* already being shared" requirement prevents us from doing that.
*/
return (shared);
}
/*
* Free the checksum associated with this header. If there is no checksum, this
* is a no-op.
*/
static inline void
arc_cksum_free(arc_buf_hdr_t *hdr)
{
ASSERT(HDR_HAS_L1HDR(hdr));
mutex_enter(&hdr->b_l1hdr.b_freeze_lock);
if (hdr->b_l1hdr.b_freeze_cksum != NULL) {
kmem_free(hdr->b_l1hdr.b_freeze_cksum, sizeof (zio_cksum_t));
hdr->b_l1hdr.b_freeze_cksum = NULL;
}
mutex_exit(&hdr->b_l1hdr.b_freeze_lock);
}
/*
* Return true iff at least one of the bufs on hdr is not compressed.
* Encrypted buffers count as compressed.
*/
static boolean_t
arc_hdr_has_uncompressed_buf(arc_buf_hdr_t *hdr)
{
ASSERT(hdr->b_l1hdr.b_state == arc_anon || HDR_EMPTY_OR_LOCKED(hdr));
for (arc_buf_t *b = hdr->b_l1hdr.b_buf; b != NULL; b = b->b_next) {
if (!ARC_BUF_COMPRESSED(b)) {
return (B_TRUE);
}
}
return (B_FALSE);
}
/*
* If we've turned on the ZFS_DEBUG_MODIFY flag, verify that the buf's data
* matches the checksum that is stored in the hdr. If there is no checksum,
* or if the buf is compressed, this is a no-op.
*/
static void
arc_cksum_verify(arc_buf_t *buf)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
zio_cksum_t zc;
if (!(zfs_flags & ZFS_DEBUG_MODIFY))
return;
if (ARC_BUF_COMPRESSED(buf))
return;
ASSERT(HDR_HAS_L1HDR(hdr));
mutex_enter(&hdr->b_l1hdr.b_freeze_lock);
if (hdr->b_l1hdr.b_freeze_cksum == NULL || HDR_IO_ERROR(hdr)) {
mutex_exit(&hdr->b_l1hdr.b_freeze_lock);
return;
}
fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL, &zc);
if (!ZIO_CHECKSUM_EQUAL(*hdr->b_l1hdr.b_freeze_cksum, zc))
panic("buffer modified while frozen!");
mutex_exit(&hdr->b_l1hdr.b_freeze_lock);
}
/*
* This function makes the assumption that data stored in the L2ARC
* will be transformed exactly as it is in the main pool. Because of
* this we can verify the checksum against the reading process's bp.
*/
static boolean_t
arc_cksum_is_equal(arc_buf_hdr_t *hdr, zio_t *zio)
{
ASSERT(!BP_IS_EMBEDDED(zio->io_bp));
VERIFY3U(BP_GET_PSIZE(zio->io_bp), ==, HDR_GET_PSIZE(hdr));
/*
* Block pointers always store the checksum for the logical data.
* If the block pointer has the gang bit set, then the checksum
* it represents is for the reconstituted data and not for an
* individual gang member. The zio pipeline, however, must be able to
* determine the checksum of each of the gang constituents so it
* treats the checksum comparison differently than what we need
* for l2arc blocks. This prevents us from using the
* zio_checksum_error() interface directly. Instead we must call the
* zio_checksum_error_impl() so that we can ensure the checksum is
* generated using the correct checksum algorithm and accounts for the
* logical I/O size and not just a gang fragment.
*/
return (zio_checksum_error_impl(zio->io_spa, zio->io_bp,
BP_GET_CHECKSUM(zio->io_bp), zio->io_abd, zio->io_size,
zio->io_offset, NULL) == 0);
}
/*
* Given a buf full of data, if ZFS_DEBUG_MODIFY is enabled this computes a
* checksum and attaches it to the buf's hdr so that we can ensure that the buf
* isn't modified later on. If buf is compressed or there is already a checksum
* on the hdr, this is a no-op (we only checksum uncompressed bufs).
*/
static void
arc_cksum_compute(arc_buf_t *buf)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
if (!(zfs_flags & ZFS_DEBUG_MODIFY))
return;
ASSERT(HDR_HAS_L1HDR(hdr));
mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock);
if (hdr->b_l1hdr.b_freeze_cksum != NULL || ARC_BUF_COMPRESSED(buf)) {
mutex_exit(&hdr->b_l1hdr.b_freeze_lock);
return;
}
ASSERT(!ARC_BUF_ENCRYPTED(buf));
ASSERT(!ARC_BUF_COMPRESSED(buf));
hdr->b_l1hdr.b_freeze_cksum = kmem_alloc(sizeof (zio_cksum_t),
KM_SLEEP);
fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL,
hdr->b_l1hdr.b_freeze_cksum);
mutex_exit(&hdr->b_l1hdr.b_freeze_lock);
arc_buf_watch(buf);
}
#ifndef _KERNEL
void
arc_buf_sigsegv(int sig, siginfo_t *si, void *unused)
{
panic("Got SIGSEGV at address: 0x%lx\n", (long)si->si_addr);
}
#endif
/* ARGSUSED */
static void
arc_buf_unwatch(arc_buf_t *buf)
{
#ifndef _KERNEL
if (arc_watch) {
ASSERT0(mprotect(buf->b_data, arc_buf_size(buf),
PROT_READ | PROT_WRITE));
}
#endif
}
/* ARGSUSED */
static void
arc_buf_watch(arc_buf_t *buf)
{
#ifndef _KERNEL
if (arc_watch)
ASSERT0(mprotect(buf->b_data, arc_buf_size(buf),
PROT_READ));
#endif
}
static arc_buf_contents_t
arc_buf_type(arc_buf_hdr_t *hdr)
{
arc_buf_contents_t type;
if (HDR_ISTYPE_METADATA(hdr)) {
type = ARC_BUFC_METADATA;
} else {
type = ARC_BUFC_DATA;
}
VERIFY3U(hdr->b_type, ==, type);
return (type);
}
boolean_t
arc_is_metadata(arc_buf_t *buf)
{
return (HDR_ISTYPE_METADATA(buf->b_hdr) != 0);
}
static uint32_t
arc_bufc_to_flags(arc_buf_contents_t type)
{
switch (type) {
case ARC_BUFC_DATA:
/* metadata field is 0 if buffer contains normal data */
return (0);
case ARC_BUFC_METADATA:
return (ARC_FLAG_BUFC_METADATA);
default:
break;
}
panic("undefined ARC buffer type!");
return ((uint32_t)-1);
}
void
arc_buf_thaw(arc_buf_t *buf)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon);
ASSERT(!HDR_IO_IN_PROGRESS(hdr));
arc_cksum_verify(buf);
/*
* Compressed buffers do not manipulate the b_freeze_cksum.
*/
if (ARC_BUF_COMPRESSED(buf))
return;
ASSERT(HDR_HAS_L1HDR(hdr));
arc_cksum_free(hdr);
arc_buf_unwatch(buf);
}
void
arc_buf_freeze(arc_buf_t *buf)
{
if (!(zfs_flags & ZFS_DEBUG_MODIFY))
return;
if (ARC_BUF_COMPRESSED(buf))
return;
ASSERT(HDR_HAS_L1HDR(buf->b_hdr));
arc_cksum_compute(buf);
}
/*
* The arc_buf_hdr_t's b_flags should never be modified directly. Instead,
* the following functions should be used to ensure that the flags are
* updated in a thread-safe way. When manipulating the flags either
* the hash_lock must be held or the hdr must be undiscoverable. This
* ensures that we're not racing with any other threads when updating
* the flags.
*/
static inline void
arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags)
{
ASSERT(HDR_EMPTY_OR_LOCKED(hdr));
hdr->b_flags |= flags;
}
static inline void
arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags)
{
ASSERT(HDR_EMPTY_OR_LOCKED(hdr));
hdr->b_flags &= ~flags;
}
/*
* Setting the compression bits in the arc_buf_hdr_t's b_flags is
* done in a special way since we have to clear and set bits
* at the same time. Consumers that wish to set the compression bits
* must use this function to ensure that the flags are updated in
* thread-safe manner.
*/
static void
arc_hdr_set_compress(arc_buf_hdr_t *hdr, enum zio_compress cmp)
{
ASSERT(HDR_EMPTY_OR_LOCKED(hdr));
/*
* Holes and embedded blocks will always have a psize = 0 so
* we ignore the compression of the blkptr and set the
* want to uncompress them. Mark them as uncompressed.
*/
if (!zfs_compressed_arc_enabled || HDR_GET_PSIZE(hdr) == 0) {
arc_hdr_clear_flags(hdr, ARC_FLAG_COMPRESSED_ARC);
ASSERT(!HDR_COMPRESSION_ENABLED(hdr));
} else {
arc_hdr_set_flags(hdr, ARC_FLAG_COMPRESSED_ARC);
ASSERT(HDR_COMPRESSION_ENABLED(hdr));
}
HDR_SET_COMPRESS(hdr, cmp);
ASSERT3U(HDR_GET_COMPRESS(hdr), ==, cmp);
}
/*
* Looks for another buf on the same hdr which has the data decompressed, copies
* from it, and returns true. If no such buf exists, returns false.
*/
static boolean_t
arc_buf_try_copy_decompressed_data(arc_buf_t *buf)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
boolean_t copied = B_FALSE;
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT3P(buf->b_data, !=, NULL);
ASSERT(!ARC_BUF_COMPRESSED(buf));
for (arc_buf_t *from = hdr->b_l1hdr.b_buf; from != NULL;
from = from->b_next) {
/* can't use our own data buffer */
if (from == buf) {
continue;
}
if (!ARC_BUF_COMPRESSED(from)) {
bcopy(from->b_data, buf->b_data, arc_buf_size(buf));
copied = B_TRUE;
break;
}
}
/*
* There were no decompressed bufs, so there should not be a
* checksum on the hdr either.
*/
if (zfs_flags & ZFS_DEBUG_MODIFY)
EQUIV(!copied, hdr->b_l1hdr.b_freeze_cksum == NULL);
return (copied);
}
/*
* Allocates an ARC buf header that's in an evicted & L2-cached state.
* This is used during l2arc reconstruction to make empty ARC buffers
* which circumvent the regular disk->arc->l2arc path and instead come
* into being in the reverse order, i.e. l2arc->arc.
*/
static arc_buf_hdr_t *
arc_buf_alloc_l2only(size_t size, arc_buf_contents_t type, l2arc_dev_t *dev,
dva_t dva, uint64_t daddr, int32_t psize, uint64_t birth,
enum zio_compress compress, uint8_t complevel, boolean_t protected,
boolean_t prefetch, arc_state_type_t arcs_state)
{
arc_buf_hdr_t *hdr;
ASSERT(size != 0);
hdr = kmem_cache_alloc(hdr_l2only_cache, KM_SLEEP);
hdr->b_birth = birth;
hdr->b_type = type;
hdr->b_flags = 0;
arc_hdr_set_flags(hdr, arc_bufc_to_flags(type) | ARC_FLAG_HAS_L2HDR);
HDR_SET_LSIZE(hdr, size);
HDR_SET_PSIZE(hdr, psize);
arc_hdr_set_compress(hdr, compress);
hdr->b_complevel = complevel;
if (protected)
arc_hdr_set_flags(hdr, ARC_FLAG_PROTECTED);
if (prefetch)
arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH);
hdr->b_spa = spa_load_guid(dev->l2ad_vdev->vdev_spa);
hdr->b_dva = dva;
hdr->b_l2hdr.b_dev = dev;
hdr->b_l2hdr.b_daddr = daddr;
hdr->b_l2hdr.b_arcs_state = arcs_state;
return (hdr);
}
/*
* Return the size of the block, b_pabd, that is stored in the arc_buf_hdr_t.
*/
static uint64_t
arc_hdr_size(arc_buf_hdr_t *hdr)
{
uint64_t size;
if (arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF &&
HDR_GET_PSIZE(hdr) > 0) {
size = HDR_GET_PSIZE(hdr);
} else {
ASSERT3U(HDR_GET_LSIZE(hdr), !=, 0);
size = HDR_GET_LSIZE(hdr);
}
return (size);
}
static int
arc_hdr_authenticate(arc_buf_hdr_t *hdr, spa_t *spa, uint64_t dsobj)
{
int ret;
uint64_t csize;
uint64_t lsize = HDR_GET_LSIZE(hdr);
uint64_t psize = HDR_GET_PSIZE(hdr);
void *tmpbuf = NULL;
abd_t *abd = hdr->b_l1hdr.b_pabd;
ASSERT(HDR_EMPTY_OR_LOCKED(hdr));
ASSERT(HDR_AUTHENTICATED(hdr));
ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
/*
* The MAC is calculated on the compressed data that is stored on disk.
* However, if compressed arc is disabled we will only have the
* decompressed data available to us now. Compress it into a temporary
* abd so we can verify the MAC. The performance overhead of this will
* be relatively low, since most objects in an encrypted objset will
* be encrypted (instead of authenticated) anyway.
*/
if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF &&
!HDR_COMPRESSION_ENABLED(hdr)) {
tmpbuf = zio_buf_alloc(lsize);
abd = abd_get_from_buf(tmpbuf, lsize);
abd_take_ownership_of_buf(abd, B_TRUE);
csize = zio_compress_data(HDR_GET_COMPRESS(hdr),
hdr->b_l1hdr.b_pabd, tmpbuf, lsize, hdr->b_complevel);
ASSERT3U(csize, <=, psize);
abd_zero_off(abd, csize, psize - csize);
}
/*
* Authentication is best effort. We authenticate whenever the key is
* available. If we succeed we clear ARC_FLAG_NOAUTH.
*/
if (hdr->b_crypt_hdr.b_ot == DMU_OT_OBJSET) {
ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF);
ASSERT3U(lsize, ==, psize);
ret = spa_do_crypt_objset_mac_abd(B_FALSE, spa, dsobj, abd,
psize, hdr->b_l1hdr.b_byteswap != DMU_BSWAP_NUMFUNCS);
} else {
ret = spa_do_crypt_mac_abd(B_FALSE, spa, dsobj, abd, psize,
hdr->b_crypt_hdr.b_mac);
}
if (ret == 0)
arc_hdr_clear_flags(hdr, ARC_FLAG_NOAUTH);
else if (ret != ENOENT)
goto error;
if (tmpbuf != NULL)
abd_free(abd);
return (0);
error:
if (tmpbuf != NULL)
abd_free(abd);
return (ret);
}
/*
* This function will take a header that only has raw encrypted data in
* b_crypt_hdr.b_rabd and decrypt it into a new buffer which is stored in
* b_l1hdr.b_pabd. If designated in the header flags, this function will
* also decompress the data.
*/
static int
arc_hdr_decrypt(arc_buf_hdr_t *hdr, spa_t *spa, const zbookmark_phys_t *zb)
{
int ret;
abd_t *cabd = NULL;
void *tmp = NULL;
boolean_t no_crypt = B_FALSE;
boolean_t bswap = (hdr->b_l1hdr.b_byteswap != DMU_BSWAP_NUMFUNCS);
ASSERT(HDR_EMPTY_OR_LOCKED(hdr));
ASSERT(HDR_ENCRYPTED(hdr));
arc_hdr_alloc_abd(hdr, ARC_HDR_DO_ADAPT);
ret = spa_do_crypt_abd(B_FALSE, spa, zb, hdr->b_crypt_hdr.b_ot,
B_FALSE, bswap, hdr->b_crypt_hdr.b_salt, hdr->b_crypt_hdr.b_iv,
hdr->b_crypt_hdr.b_mac, HDR_GET_PSIZE(hdr), hdr->b_l1hdr.b_pabd,
hdr->b_crypt_hdr.b_rabd, &no_crypt);
if (ret != 0)
goto error;
if (no_crypt) {
abd_copy(hdr->b_l1hdr.b_pabd, hdr->b_crypt_hdr.b_rabd,
HDR_GET_PSIZE(hdr));
}
/*
* If this header has disabled arc compression but the b_pabd is
* compressed after decrypting it, we need to decompress the newly
* decrypted data.
*/
if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF &&
!HDR_COMPRESSION_ENABLED(hdr)) {
/*
* We want to make sure that we are correctly honoring the
* zfs_abd_scatter_enabled setting, so we allocate an abd here
* and then loan a buffer from it, rather than allocating a
* linear buffer and wrapping it in an abd later.
*/
cabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr, B_TRUE);
tmp = abd_borrow_buf(cabd, arc_hdr_size(hdr));
ret = zio_decompress_data(HDR_GET_COMPRESS(hdr),
hdr->b_l1hdr.b_pabd, tmp, HDR_GET_PSIZE(hdr),
HDR_GET_LSIZE(hdr), &hdr->b_complevel);
if (ret != 0) {
abd_return_buf(cabd, tmp, arc_hdr_size(hdr));
goto error;
}
abd_return_buf_copy(cabd, tmp, arc_hdr_size(hdr));
arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd,
arc_hdr_size(hdr), hdr);
hdr->b_l1hdr.b_pabd = cabd;
}
return (0);
error:
arc_hdr_free_abd(hdr, B_FALSE);
if (cabd != NULL)
arc_free_data_buf(hdr, cabd, arc_hdr_size(hdr), hdr);
return (ret);
}
/*
* This function is called during arc_buf_fill() to prepare the header's
* abd plaintext pointer for use. This involves authenticated protected
* data and decrypting encrypted data into the plaintext abd.
*/
static int
arc_fill_hdr_crypt(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, spa_t *spa,
const zbookmark_phys_t *zb, boolean_t noauth)
{
int ret;
ASSERT(HDR_PROTECTED(hdr));
if (hash_lock != NULL)
mutex_enter(hash_lock);
if (HDR_NOAUTH(hdr) && !noauth) {
/*
* The caller requested authenticated data but our data has
* not been authenticated yet. Verify the MAC now if we can.
*/
ret = arc_hdr_authenticate(hdr, spa, zb->zb_objset);
if (ret != 0)
goto error;
} else if (HDR_HAS_RABD(hdr) && hdr->b_l1hdr.b_pabd == NULL) {
/*
* If we only have the encrypted version of the data, but the
* unencrypted version was requested we take this opportunity
* to store the decrypted version in the header for future use.
*/
ret = arc_hdr_decrypt(hdr, spa, zb);
if (ret != 0)
goto error;
}
ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
if (hash_lock != NULL)
mutex_exit(hash_lock);
return (0);
error:
if (hash_lock != NULL)
mutex_exit(hash_lock);
return (ret);
}
/*
* This function is used by the dbuf code to decrypt bonus buffers in place.
* The dbuf code itself doesn't have any locking for decrypting a shared dnode
* block, so we use the hash lock here to protect against concurrent calls to
* arc_buf_fill().
*/
static void
arc_buf_untransform_in_place(arc_buf_t *buf, kmutex_t *hash_lock)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
ASSERT(HDR_ENCRYPTED(hdr));
ASSERT3U(hdr->b_crypt_hdr.b_ot, ==, DMU_OT_DNODE);
ASSERT(HDR_EMPTY_OR_LOCKED(hdr));
ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
zio_crypt_copy_dnode_bonus(hdr->b_l1hdr.b_pabd, buf->b_data,
arc_buf_size(buf));
buf->b_flags &= ~ARC_BUF_FLAG_ENCRYPTED;
buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED;
hdr->b_crypt_hdr.b_ebufcnt -= 1;
}
/*
* Given a buf that has a data buffer attached to it, this function will
* efficiently fill the buf with data of the specified compression setting from
* the hdr and update the hdr's b_freeze_cksum if necessary. If the buf and hdr
* are already sharing a data buf, no copy is performed.
*
* If the buf is marked as compressed but uncompressed data was requested, this
* will allocate a new data buffer for the buf, remove that flag, and fill the
* buf with uncompressed data. You can't request a compressed buf on a hdr with
* uncompressed data, and (since we haven't added support for it yet) if you
* want compressed data your buf must already be marked as compressed and have
* the correct-sized data buffer.
*/
static int
arc_buf_fill(arc_buf_t *buf, spa_t *spa, const zbookmark_phys_t *zb,
arc_fill_flags_t flags)
{
int error = 0;
arc_buf_hdr_t *hdr = buf->b_hdr;
boolean_t hdr_compressed =
(arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF);
boolean_t compressed = (flags & ARC_FILL_COMPRESSED) != 0;
boolean_t encrypted = (flags & ARC_FILL_ENCRYPTED) != 0;
dmu_object_byteswap_t bswap = hdr->b_l1hdr.b_byteswap;
kmutex_t *hash_lock = (flags & ARC_FILL_LOCKED) ? NULL : HDR_LOCK(hdr);
ASSERT3P(buf->b_data, !=, NULL);
IMPLY(compressed, hdr_compressed || ARC_BUF_ENCRYPTED(buf));
IMPLY(compressed, ARC_BUF_COMPRESSED(buf));
IMPLY(encrypted, HDR_ENCRYPTED(hdr));
IMPLY(encrypted, ARC_BUF_ENCRYPTED(buf));
IMPLY(encrypted, ARC_BUF_COMPRESSED(buf));
IMPLY(encrypted, !ARC_BUF_SHARED(buf));
/*
* If the caller wanted encrypted data we just need to copy it from
* b_rabd and potentially byteswap it. We won't be able to do any
* further transforms on it.
*/
if (encrypted) {
ASSERT(HDR_HAS_RABD(hdr));
abd_copy_to_buf(buf->b_data, hdr->b_crypt_hdr.b_rabd,
HDR_GET_PSIZE(hdr));
goto byteswap;
}
/*
* Adjust encrypted and authenticated headers to accommodate
* the request if needed. Dnode blocks (ARC_FILL_IN_PLACE) are
* allowed to fail decryption due to keys not being loaded
* without being marked as an IO error.
*/
if (HDR_PROTECTED(hdr)) {
error = arc_fill_hdr_crypt(hdr, hash_lock, spa,
zb, !!(flags & ARC_FILL_NOAUTH));
if (error == EACCES && (flags & ARC_FILL_IN_PLACE) != 0) {
return (error);
} else if (error != 0) {
if (hash_lock != NULL)
mutex_enter(hash_lock);
arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR);
if (hash_lock != NULL)
mutex_exit(hash_lock);
return (error);
}
}
/*
* There is a special case here for dnode blocks which are
* decrypting their bonus buffers. These blocks may request to
* be decrypted in-place. This is necessary because there may
* be many dnodes pointing into this buffer and there is
* currently no method to synchronize replacing the backing
* b_data buffer and updating all of the pointers. Here we use
* the hash lock to ensure there are no races. If the need
* arises for other types to be decrypted in-place, they must
* add handling here as well.
*/
if ((flags & ARC_FILL_IN_PLACE) != 0) {
ASSERT(!hdr_compressed);
ASSERT(!compressed);
ASSERT(!encrypted);
if (HDR_ENCRYPTED(hdr) && ARC_BUF_ENCRYPTED(buf)) {
ASSERT3U(hdr->b_crypt_hdr.b_ot, ==, DMU_OT_DNODE);
if (hash_lock != NULL)
mutex_enter(hash_lock);
arc_buf_untransform_in_place(buf, hash_lock);
if (hash_lock != NULL)
mutex_exit(hash_lock);
/* Compute the hdr's checksum if necessary */
arc_cksum_compute(buf);
}
return (0);
}
if (hdr_compressed == compressed) {
if (!arc_buf_is_shared(buf)) {
abd_copy_to_buf(buf->b_data, hdr->b_l1hdr.b_pabd,
arc_buf_size(buf));
}
} else {
ASSERT(hdr_compressed);
ASSERT(!compressed);
ASSERT3U(HDR_GET_LSIZE(hdr), !=, HDR_GET_PSIZE(hdr));
/*
* If the buf is sharing its data with the hdr, unlink it and
* allocate a new data buffer for the buf.
*/
if (arc_buf_is_shared(buf)) {
ASSERT(ARC_BUF_COMPRESSED(buf));
/* We need to give the buf its own b_data */
buf->b_flags &= ~ARC_BUF_FLAG_SHARED;
buf->b_data =
arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf);
arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA);
/* Previously overhead was 0; just add new overhead */
ARCSTAT_INCR(arcstat_overhead_size, HDR_GET_LSIZE(hdr));
} else if (ARC_BUF_COMPRESSED(buf)) {
/* We need to reallocate the buf's b_data */
arc_free_data_buf(hdr, buf->b_data, HDR_GET_PSIZE(hdr),
buf);
buf->b_data =
arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf);
/* We increased the size of b_data; update overhead */
ARCSTAT_INCR(arcstat_overhead_size,
HDR_GET_LSIZE(hdr) - HDR_GET_PSIZE(hdr));
}
/*
* Regardless of the buf's previous compression settings, it
* should not be compressed at the end of this function.
*/
buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED;
/*
* Try copying the data from another buf which already has a
* decompressed version. If that's not possible, it's time to
* bite the bullet and decompress the data from the hdr.
*/
if (arc_buf_try_copy_decompressed_data(buf)) {
/* Skip byteswapping and checksumming (already done) */
return (0);
} else {
error = zio_decompress_data(HDR_GET_COMPRESS(hdr),
hdr->b_l1hdr.b_pabd, buf->b_data,
HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr),
&hdr->b_complevel);
/*
* Absent hardware errors or software bugs, this should
* be impossible, but log it anyway so we can debug it.
*/
if (error != 0) {
zfs_dbgmsg(
"hdr %px, compress %d, psize %d, lsize %d",
hdr, arc_hdr_get_compress(hdr),
HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr));
if (hash_lock != NULL)
mutex_enter(hash_lock);
arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR);
if (hash_lock != NULL)
mutex_exit(hash_lock);
return (SET_ERROR(EIO));
}
}
}
byteswap:
/* Byteswap the buf's data if necessary */
if (bswap != DMU_BSWAP_NUMFUNCS) {
ASSERT(!HDR_SHARED_DATA(hdr));
ASSERT3U(bswap, <, DMU_BSWAP_NUMFUNCS);
dmu_ot_byteswap[bswap].ob_func(buf->b_data, HDR_GET_LSIZE(hdr));
}
/* Compute the hdr's checksum if necessary */
arc_cksum_compute(buf);
return (0);
}
/*
* If this function is being called to decrypt an encrypted buffer or verify an
* authenticated one, the key must be loaded and a mapping must be made
* available in the keystore via spa_keystore_create_mapping() or one of its
* callers.
*/
int
arc_untransform(arc_buf_t *buf, spa_t *spa, const zbookmark_phys_t *zb,
boolean_t in_place)
{
int ret;
arc_fill_flags_t flags = 0;
if (in_place)
flags |= ARC_FILL_IN_PLACE;
ret = arc_buf_fill(buf, spa, zb, flags);
if (ret == ECKSUM) {
/*
* Convert authentication and decryption errors to EIO
* (and generate an ereport) before leaving the ARC.
*/
ret = SET_ERROR(EIO);
spa_log_error(spa, zb);
(void) zfs_ereport_post(FM_EREPORT_ZFS_AUTHENTICATION,
spa, NULL, zb, NULL, 0);
}
return (ret);
}
/*
* Increment the amount of evictable space in the arc_state_t's refcount.
* We account for the space used by the hdr and the arc buf individually
* so that we can add and remove them from the refcount individually.
*/
static void
arc_evictable_space_increment(arc_buf_hdr_t *hdr, arc_state_t *state)
{
arc_buf_contents_t type = arc_buf_type(hdr);
ASSERT(HDR_HAS_L1HDR(hdr));
if (GHOST_STATE(state)) {
ASSERT0(hdr->b_l1hdr.b_bufcnt);
ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL);
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
ASSERT(!HDR_HAS_RABD(hdr));
(void) zfs_refcount_add_many(&state->arcs_esize[type],
HDR_GET_LSIZE(hdr), hdr);
return;
}
ASSERT(!GHOST_STATE(state));
if (hdr->b_l1hdr.b_pabd != NULL) {
(void) zfs_refcount_add_many(&state->arcs_esize[type],
arc_hdr_size(hdr), hdr);
}
if (HDR_HAS_RABD(hdr)) {
(void) zfs_refcount_add_many(&state->arcs_esize[type],
HDR_GET_PSIZE(hdr), hdr);
}
for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL;
buf = buf->b_next) {
if (arc_buf_is_shared(buf))
continue;
(void) zfs_refcount_add_many(&state->arcs_esize[type],
arc_buf_size(buf), buf);
}
}
/*
* Decrement the amount of evictable space in the arc_state_t's refcount.
* We account for the space used by the hdr and the arc buf individually
* so that we can add and remove them from the refcount individually.
*/
static void
arc_evictable_space_decrement(arc_buf_hdr_t *hdr, arc_state_t *state)
{
arc_buf_contents_t type = arc_buf_type(hdr);
ASSERT(HDR_HAS_L1HDR(hdr));
if (GHOST_STATE(state)) {
ASSERT0(hdr->b_l1hdr.b_bufcnt);
ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL);
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
ASSERT(!HDR_HAS_RABD(hdr));
(void) zfs_refcount_remove_many(&state->arcs_esize[type],
HDR_GET_LSIZE(hdr), hdr);
return;
}
ASSERT(!GHOST_STATE(state));
if (hdr->b_l1hdr.b_pabd != NULL) {
(void) zfs_refcount_remove_many(&state->arcs_esize[type],
arc_hdr_size(hdr), hdr);
}
if (HDR_HAS_RABD(hdr)) {
(void) zfs_refcount_remove_many(&state->arcs_esize[type],
HDR_GET_PSIZE(hdr), hdr);
}
for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL;
buf = buf->b_next) {
if (arc_buf_is_shared(buf))
continue;
(void) zfs_refcount_remove_many(&state->arcs_esize[type],
arc_buf_size(buf), buf);
}
}
/*
* Add a reference to this hdr indicating that someone is actively
* referencing that memory. When the refcount transitions from 0 to 1,
* we remove it from the respective arc_state_t list to indicate that
* it is not evictable.
*/
static void
add_reference(arc_buf_hdr_t *hdr, void *tag)
{
arc_state_t *state;
ASSERT(HDR_HAS_L1HDR(hdr));
if (!HDR_EMPTY(hdr) && !MUTEX_HELD(HDR_LOCK(hdr))) {
ASSERT(hdr->b_l1hdr.b_state == arc_anon);
ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL);
}
state = hdr->b_l1hdr.b_state;
if ((zfs_refcount_add(&hdr->b_l1hdr.b_refcnt, tag) == 1) &&
(state != arc_anon)) {
/* We don't use the L2-only state list. */
if (state != arc_l2c_only) {
multilist_remove(&state->arcs_list[arc_buf_type(hdr)],
hdr);
arc_evictable_space_decrement(hdr, state);
}
/* remove the prefetch flag if we get a reference */
if (HDR_HAS_L2HDR(hdr))
l2arc_hdr_arcstats_decrement_state(hdr);
arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH);
if (HDR_HAS_L2HDR(hdr))
l2arc_hdr_arcstats_increment_state(hdr);
}
}
/*
* Remove a reference from this hdr. When the reference transitions from
* 1 to 0 and we're not anonymous, then we add this hdr to the arc_state_t's
* list making it eligible for eviction.
*/
static int
remove_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag)
{
int cnt;
arc_state_t *state = hdr->b_l1hdr.b_state;
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT(state == arc_anon || MUTEX_HELD(hash_lock));
ASSERT(!GHOST_STATE(state));
/*
* arc_l2c_only counts as a ghost state so we don't need to explicitly
* check to prevent usage of the arc_l2c_only list.
*/
if (((cnt = zfs_refcount_remove(&hdr->b_l1hdr.b_refcnt, tag)) == 0) &&
(state != arc_anon)) {
multilist_insert(&state->arcs_list[arc_buf_type(hdr)], hdr);
ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0);
arc_evictable_space_increment(hdr, state);
}
return (cnt);
}
/*
* Returns detailed information about a specific arc buffer. When the
* state_index argument is set the function will calculate the arc header
* list position for its arc state. Since this requires a linear traversal
* callers are strongly encourage not to do this. However, it can be helpful
* for targeted analysis so the functionality is provided.
*/
void
arc_buf_info(arc_buf_t *ab, arc_buf_info_t *abi, int state_index)
{
arc_buf_hdr_t *hdr = ab->b_hdr;
l1arc_buf_hdr_t *l1hdr = NULL;
l2arc_buf_hdr_t *l2hdr = NULL;
arc_state_t *state = NULL;
memset(abi, 0, sizeof (arc_buf_info_t));
if (hdr == NULL)
return;
abi->abi_flags = hdr->b_flags;
if (HDR_HAS_L1HDR(hdr)) {
l1hdr = &hdr->b_l1hdr;
state = l1hdr->b_state;
}
if (HDR_HAS_L2HDR(hdr))
l2hdr = &hdr->b_l2hdr;
if (l1hdr) {
abi->abi_bufcnt = l1hdr->b_bufcnt;
abi->abi_access = l1hdr->b_arc_access;
abi->abi_mru_hits = l1hdr->b_mru_hits;
abi->abi_mru_ghost_hits = l1hdr->b_mru_ghost_hits;
abi->abi_mfu_hits = l1hdr->b_mfu_hits;
abi->abi_mfu_ghost_hits = l1hdr->b_mfu_ghost_hits;
abi->abi_holds = zfs_refcount_count(&l1hdr->b_refcnt);
}
if (l2hdr) {
abi->abi_l2arc_dattr = l2hdr->b_daddr;
abi->abi_l2arc_hits = l2hdr->b_hits;
}
abi->abi_state_type = state ? state->arcs_state : ARC_STATE_ANON;
abi->abi_state_contents = arc_buf_type(hdr);
abi->abi_size = arc_hdr_size(hdr);
}
/*
* Move the supplied buffer to the indicated state. The hash lock
* for the buffer must be held by the caller.
*/
static void
arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr,
kmutex_t *hash_lock)
{
arc_state_t *old_state;
int64_t refcnt;
uint32_t bufcnt;
boolean_t update_old, update_new;
arc_buf_contents_t buftype = arc_buf_type(hdr);
/*
* We almost always have an L1 hdr here, since we call arc_hdr_realloc()
* in arc_read() when bringing a buffer out of the L2ARC. However, the
* L1 hdr doesn't always exist when we change state to arc_anon before
* destroying a header, in which case reallocating to add the L1 hdr is
* pointless.
*/
if (HDR_HAS_L1HDR(hdr)) {
old_state = hdr->b_l1hdr.b_state;
refcnt = zfs_refcount_count(&hdr->b_l1hdr.b_refcnt);
bufcnt = hdr->b_l1hdr.b_bufcnt;
update_old = (bufcnt > 0 || hdr->b_l1hdr.b_pabd != NULL ||
HDR_HAS_RABD(hdr));
} else {
old_state = arc_l2c_only;
refcnt = 0;
bufcnt = 0;
update_old = B_FALSE;
}
update_new = update_old;
ASSERT(MUTEX_HELD(hash_lock));
ASSERT3P(new_state, !=, old_state);
ASSERT(!GHOST_STATE(new_state) || bufcnt == 0);
ASSERT(old_state != arc_anon || bufcnt <= 1);
/*
* If this buffer is evictable, transfer it from the
* old state list to the new state list.
*/
if (refcnt == 0) {
if (old_state != arc_anon && old_state != arc_l2c_only) {
ASSERT(HDR_HAS_L1HDR(hdr));
multilist_remove(&old_state->arcs_list[buftype], hdr);
if (GHOST_STATE(old_state)) {
ASSERT0(bufcnt);
ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL);
update_old = B_TRUE;
}
arc_evictable_space_decrement(hdr, old_state);
}
if (new_state != arc_anon && new_state != arc_l2c_only) {
/*
* An L1 header always exists here, since if we're
* moving to some L1-cached state (i.e. not l2c_only or
* anonymous), we realloc the header to add an L1hdr
* beforehand.
*/
ASSERT(HDR_HAS_L1HDR(hdr));
multilist_insert(&new_state->arcs_list[buftype], hdr);
if (GHOST_STATE(new_state)) {
ASSERT0(bufcnt);
ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL);
update_new = B_TRUE;
}
arc_evictable_space_increment(hdr, new_state);
}
}
ASSERT(!HDR_EMPTY(hdr));
if (new_state == arc_anon && HDR_IN_HASH_TABLE(hdr))
buf_hash_remove(hdr);
/* adjust state sizes (ignore arc_l2c_only) */
if (update_new && new_state != arc_l2c_only) {
ASSERT(HDR_HAS_L1HDR(hdr));
if (GHOST_STATE(new_state)) {
ASSERT0(bufcnt);
/*
* When moving a header to a ghost state, we first
* remove all arc buffers. Thus, we'll have a
* bufcnt of zero, and no arc buffer to use for
* the reference. As a result, we use the arc
* header pointer for the reference.
*/
(void) zfs_refcount_add_many(&new_state->arcs_size,
HDR_GET_LSIZE(hdr), hdr);
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
ASSERT(!HDR_HAS_RABD(hdr));
} else {
uint32_t buffers = 0;
/*
* Each individual buffer holds a unique reference,
* thus we must remove each of these references one
* at a time.
*/
for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL;
buf = buf->b_next) {
ASSERT3U(bufcnt, !=, 0);
buffers++;
/*
* When the arc_buf_t is sharing the data
* block with the hdr, the owner of the
* reference belongs to the hdr. Only
* add to the refcount if the arc_buf_t is
* not shared.
*/
if (arc_buf_is_shared(buf))
continue;
(void) zfs_refcount_add_many(
&new_state->arcs_size,
arc_buf_size(buf), buf);
}
ASSERT3U(bufcnt, ==, buffers);
if (hdr->b_l1hdr.b_pabd != NULL) {
(void) zfs_refcount_add_many(
&new_state->arcs_size,
arc_hdr_size(hdr), hdr);
}
if (HDR_HAS_RABD(hdr)) {
(void) zfs_refcount_add_many(
&new_state->arcs_size,
HDR_GET_PSIZE(hdr), hdr);
}
}
}
if (update_old && old_state != arc_l2c_only) {
ASSERT(HDR_HAS_L1HDR(hdr));
if (GHOST_STATE(old_state)) {
ASSERT0(bufcnt);
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
ASSERT(!HDR_HAS_RABD(hdr));
/*
* When moving a header off of a ghost state,
* the header will not contain any arc buffers.
* We use the arc header pointer for the reference
* which is exactly what we did when we put the
* header on the ghost state.
*/
(void) zfs_refcount_remove_many(&old_state->arcs_size,
HDR_GET_LSIZE(hdr), hdr);
} else {
uint32_t buffers = 0;
/*
* Each individual buffer holds a unique reference,
* thus we must remove each of these references one
* at a time.
*/
for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL;
buf = buf->b_next) {
ASSERT3U(bufcnt, !=, 0);
buffers++;
/*
* When the arc_buf_t is sharing the data
* block with the hdr, the owner of the
* reference belongs to the hdr. Only
* add to the refcount if the arc_buf_t is
* not shared.
*/
if (arc_buf_is_shared(buf))
continue;
(void) zfs_refcount_remove_many(
&old_state->arcs_size, arc_buf_size(buf),
buf);
}
ASSERT3U(bufcnt, ==, buffers);
ASSERT(hdr->b_l1hdr.b_pabd != NULL ||
HDR_HAS_RABD(hdr));
if (hdr->b_l1hdr.b_pabd != NULL) {
(void) zfs_refcount_remove_many(
&old_state->arcs_size, arc_hdr_size(hdr),
hdr);
}
if (HDR_HAS_RABD(hdr)) {
(void) zfs_refcount_remove_many(
&old_state->arcs_size, HDR_GET_PSIZE(hdr),
hdr);
}
}
}
if (HDR_HAS_L1HDR(hdr)) {
hdr->b_l1hdr.b_state = new_state;
if (HDR_HAS_L2HDR(hdr) && new_state != arc_l2c_only) {
l2arc_hdr_arcstats_decrement_state(hdr);
hdr->b_l2hdr.b_arcs_state = new_state->arcs_state;
l2arc_hdr_arcstats_increment_state(hdr);
}
}
/*
* L2 headers should never be on the L2 state list since they don't
* have L1 headers allocated.
*/
ASSERT(multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]) &&
multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA]));
}
void
arc_space_consume(uint64_t space, arc_space_type_t type)
{
ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES);
switch (type) {
default:
break;
case ARC_SPACE_DATA:
ARCSTAT_INCR(arcstat_data_size, space);
break;
case ARC_SPACE_META:
ARCSTAT_INCR(arcstat_metadata_size, space);
break;
case ARC_SPACE_BONUS:
ARCSTAT_INCR(arcstat_bonus_size, space);
break;
case ARC_SPACE_DNODE:
aggsum_add(&arc_sums.arcstat_dnode_size, space);
break;
case ARC_SPACE_DBUF:
ARCSTAT_INCR(arcstat_dbuf_size, space);
break;
case ARC_SPACE_HDRS:
ARCSTAT_INCR(arcstat_hdr_size, space);
break;
case ARC_SPACE_L2HDRS:
aggsum_add(&arc_sums.arcstat_l2_hdr_size, space);
break;
case ARC_SPACE_ABD_CHUNK_WASTE:
/*
* Note: this includes space wasted by all scatter ABD's, not
* just those allocated by the ARC. But the vast majority of
* scatter ABD's come from the ARC, because other users are
* very short-lived.
*/
ARCSTAT_INCR(arcstat_abd_chunk_waste_size, space);
break;
}
if (type != ARC_SPACE_DATA && type != ARC_SPACE_ABD_CHUNK_WASTE)
aggsum_add(&arc_sums.arcstat_meta_used, space);
aggsum_add(&arc_sums.arcstat_size, space);
}
void
arc_space_return(uint64_t space, arc_space_type_t type)
{
ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES);
switch (type) {
default:
break;
case ARC_SPACE_DATA:
ARCSTAT_INCR(arcstat_data_size, -space);
break;
case ARC_SPACE_META:
ARCSTAT_INCR(arcstat_metadata_size, -space);
break;
case ARC_SPACE_BONUS:
ARCSTAT_INCR(arcstat_bonus_size, -space);
break;
case ARC_SPACE_DNODE:
aggsum_add(&arc_sums.arcstat_dnode_size, -space);
break;
case ARC_SPACE_DBUF:
ARCSTAT_INCR(arcstat_dbuf_size, -space);
break;
case ARC_SPACE_HDRS:
ARCSTAT_INCR(arcstat_hdr_size, -space);
break;
case ARC_SPACE_L2HDRS:
aggsum_add(&arc_sums.arcstat_l2_hdr_size, -space);
break;
case ARC_SPACE_ABD_CHUNK_WASTE:
ARCSTAT_INCR(arcstat_abd_chunk_waste_size, -space);
break;
}
if (type != ARC_SPACE_DATA && type != ARC_SPACE_ABD_CHUNK_WASTE) {
ASSERT(aggsum_compare(&arc_sums.arcstat_meta_used,
space) >= 0);
ARCSTAT_MAX(arcstat_meta_max,
aggsum_upper_bound(&arc_sums.arcstat_meta_used));
aggsum_add(&arc_sums.arcstat_meta_used, -space);
}
ASSERT(aggsum_compare(&arc_sums.arcstat_size, space) >= 0);
aggsum_add(&arc_sums.arcstat_size, -space);
}
/*
* Given a hdr and a buf, returns whether that buf can share its b_data buffer
* with the hdr's b_pabd.
*/
static boolean_t
arc_can_share(arc_buf_hdr_t *hdr, arc_buf_t *buf)
{
/*
* The criteria for sharing a hdr's data are:
* 1. the buffer is not encrypted
* 2. the hdr's compression matches the buf's compression
* 3. the hdr doesn't need to be byteswapped
* 4. the hdr isn't already being shared
* 5. the buf is either compressed or it is the last buf in the hdr list
*
* Criterion #5 maintains the invariant that shared uncompressed
* bufs must be the final buf in the hdr's b_buf list. Reading this, you
* might ask, "if a compressed buf is allocated first, won't that be the
* last thing in the list?", but in that case it's impossible to create
* a shared uncompressed buf anyway (because the hdr must be compressed
* to have the compressed buf). You might also think that #3 is
* sufficient to make this guarantee, however it's possible
* (specifically in the rare L2ARC write race mentioned in
* arc_buf_alloc_impl()) there will be an existing uncompressed buf that
* is shareable, but wasn't at the time of its allocation. Rather than
* allow a new shared uncompressed buf to be created and then shuffle
* the list around to make it the last element, this simply disallows
* sharing if the new buf isn't the first to be added.
*/
ASSERT3P(buf->b_hdr, ==, hdr);
boolean_t hdr_compressed =
arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF;
boolean_t buf_compressed = ARC_BUF_COMPRESSED(buf) != 0;
return (!ARC_BUF_ENCRYPTED(buf) &&
buf_compressed == hdr_compressed &&
hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS &&
!HDR_SHARED_DATA(hdr) &&
(ARC_BUF_LAST(buf) || ARC_BUF_COMPRESSED(buf)));
}
/*
* Allocate a buf for this hdr. If you care about the data that's in the hdr,
* or if you want a compressed buffer, pass those flags in. Returns 0 if the
* copy was made successfully, or an error code otherwise.
*/
static int
arc_buf_alloc_impl(arc_buf_hdr_t *hdr, spa_t *spa, const zbookmark_phys_t *zb,
void *tag, boolean_t encrypted, boolean_t compressed, boolean_t noauth,
boolean_t fill, arc_buf_t **ret)
{
arc_buf_t *buf;
arc_fill_flags_t flags = ARC_FILL_LOCKED;
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT3U(HDR_GET_LSIZE(hdr), >, 0);
VERIFY(hdr->b_type == ARC_BUFC_DATA ||
hdr->b_type == ARC_BUFC_METADATA);
ASSERT3P(ret, !=, NULL);
ASSERT3P(*ret, ==, NULL);
IMPLY(encrypted, compressed);
hdr->b_l1hdr.b_mru_hits = 0;
hdr->b_l1hdr.b_mru_ghost_hits = 0;
hdr->b_l1hdr.b_mfu_hits = 0;
hdr->b_l1hdr.b_mfu_ghost_hits = 0;
hdr->b_l1hdr.b_l2_hits = 0;
buf = *ret = kmem_cache_alloc(buf_cache, KM_PUSHPAGE);
buf->b_hdr = hdr;
buf->b_data = NULL;
buf->b_next = hdr->b_l1hdr.b_buf;
buf->b_flags = 0;
add_reference(hdr, tag);
/*
* We're about to change the hdr's b_flags. We must either
* hold the hash_lock or be undiscoverable.
*/
ASSERT(HDR_EMPTY_OR_LOCKED(hdr));
/*
* Only honor requests for compressed bufs if the hdr is actually
* compressed. This must be overridden if the buffer is encrypted since
* encrypted buffers cannot be decompressed.
*/
if (encrypted) {
buf->b_flags |= ARC_BUF_FLAG_COMPRESSED;
buf->b_flags |= ARC_BUF_FLAG_ENCRYPTED;
flags |= ARC_FILL_COMPRESSED | ARC_FILL_ENCRYPTED;
} else if (compressed &&
arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF) {
buf->b_flags |= ARC_BUF_FLAG_COMPRESSED;
flags |= ARC_FILL_COMPRESSED;
}
if (noauth) {
ASSERT0(encrypted);
flags |= ARC_FILL_NOAUTH;
}
/*
* If the hdr's data can be shared then we share the data buffer and
* set the appropriate bit in the hdr's b_flags to indicate the hdr is
* sharing it's b_pabd with the arc_buf_t. Otherwise, we allocate a new
* buffer to store the buf's data.
*
* There are two additional restrictions here because we're sharing
* hdr -> buf instead of the usual buf -> hdr. First, the hdr can't be
* actively involved in an L2ARC write, because if this buf is used by
* an arc_write() then the hdr's data buffer will be released when the
* write completes, even though the L2ARC write might still be using it.
* Second, the hdr's ABD must be linear so that the buf's user doesn't
* need to be ABD-aware. It must be allocated via
* zio_[data_]buf_alloc(), not as a page, because we need to be able
* to abd_release_ownership_of_buf(), which isn't allowed on "linear
* page" buffers because the ABD code needs to handle freeing them
* specially.
*/
boolean_t can_share = arc_can_share(hdr, buf) &&
!HDR_L2_WRITING(hdr) &&
hdr->b_l1hdr.b_pabd != NULL &&
abd_is_linear(hdr->b_l1hdr.b_pabd) &&
!abd_is_linear_page(hdr->b_l1hdr.b_pabd);
/* Set up b_data and sharing */
if (can_share) {
buf->b_data = abd_to_buf(hdr->b_l1hdr.b_pabd);
buf->b_flags |= ARC_BUF_FLAG_SHARED;
arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA);
} else {
buf->b_data =
arc_get_data_buf(hdr, arc_buf_size(buf), buf);
ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf));
}
VERIFY3P(buf->b_data, !=, NULL);
hdr->b_l1hdr.b_buf = buf;
hdr->b_l1hdr.b_bufcnt += 1;
if (encrypted)
hdr->b_crypt_hdr.b_ebufcnt += 1;
/*
* If the user wants the data from the hdr, we need to either copy or
* decompress the data.
*/
if (fill) {
ASSERT3P(zb, !=, NULL);
return (arc_buf_fill(buf, spa, zb, flags));
}
return (0);
}
static char *arc_onloan_tag = "onloan";
static inline void
arc_loaned_bytes_update(int64_t delta)
{
atomic_add_64(&arc_loaned_bytes, delta);
/* assert that it did not wrap around */
ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes, 0), >=, 0);
}
/*
* Loan out an anonymous arc buffer. Loaned buffers are not counted as in
* flight data by arc_tempreserve_space() until they are "returned". Loaned
* buffers must be returned to the arc before they can be used by the DMU or
* freed.
*/
arc_buf_t *
arc_loan_buf(spa_t *spa, boolean_t is_metadata, int size)
{
arc_buf_t *buf = arc_alloc_buf(spa, arc_onloan_tag,
is_metadata ? ARC_BUFC_METADATA : ARC_BUFC_DATA, size);
arc_loaned_bytes_update(arc_buf_size(buf));
return (buf);
}
arc_buf_t *
arc_loan_compressed_buf(spa_t *spa, uint64_t psize, uint64_t lsize,
enum zio_compress compression_type, uint8_t complevel)
{
arc_buf_t *buf = arc_alloc_compressed_buf(spa, arc_onloan_tag,
psize, lsize, compression_type, complevel);
arc_loaned_bytes_update(arc_buf_size(buf));
return (buf);
}
arc_buf_t *
arc_loan_raw_buf(spa_t *spa, uint64_t dsobj, boolean_t byteorder,
const uint8_t *salt, const uint8_t *iv, const uint8_t *mac,
dmu_object_type_t ot, uint64_t psize, uint64_t lsize,
enum zio_compress compression_type, uint8_t complevel)
{
arc_buf_t *buf = arc_alloc_raw_buf(spa, arc_onloan_tag, dsobj,
byteorder, salt, iv, mac, ot, psize, lsize, compression_type,
complevel);
atomic_add_64(&arc_loaned_bytes, psize);
return (buf);
}
/*
* Return a loaned arc buffer to the arc.
*/
void
arc_return_buf(arc_buf_t *buf, void *tag)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
ASSERT3P(buf->b_data, !=, NULL);
ASSERT(HDR_HAS_L1HDR(hdr));
(void) zfs_refcount_add(&hdr->b_l1hdr.b_refcnt, tag);
(void) zfs_refcount_remove(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag);
arc_loaned_bytes_update(-arc_buf_size(buf));
}
/* Detach an arc_buf from a dbuf (tag) */
void
arc_loan_inuse_buf(arc_buf_t *buf, void *tag)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
ASSERT3P(buf->b_data, !=, NULL);
ASSERT(HDR_HAS_L1HDR(hdr));
(void) zfs_refcount_add(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag);
(void) zfs_refcount_remove(&hdr->b_l1hdr.b_refcnt, tag);
arc_loaned_bytes_update(arc_buf_size(buf));
}
static void
l2arc_free_abd_on_write(abd_t *abd, size_t size, arc_buf_contents_t type)
{
l2arc_data_free_t *df = kmem_alloc(sizeof (*df), KM_SLEEP);
df->l2df_abd = abd;
df->l2df_size = size;
df->l2df_type = type;
mutex_enter(&l2arc_free_on_write_mtx);
list_insert_head(l2arc_free_on_write, df);
mutex_exit(&l2arc_free_on_write_mtx);
}
static void
arc_hdr_free_on_write(arc_buf_hdr_t *hdr, boolean_t free_rdata)
{
arc_state_t *state = hdr->b_l1hdr.b_state;
arc_buf_contents_t type = arc_buf_type(hdr);
uint64_t size = (free_rdata) ? HDR_GET_PSIZE(hdr) : arc_hdr_size(hdr);
/* protected by hash lock, if in the hash table */
if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) {
ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
ASSERT(state != arc_anon && state != arc_l2c_only);
(void) zfs_refcount_remove_many(&state->arcs_esize[type],
size, hdr);
}
(void) zfs_refcount_remove_many(&state->arcs_size, size, hdr);
if (type == ARC_BUFC_METADATA) {
arc_space_return(size, ARC_SPACE_META);
} else {
ASSERT(type == ARC_BUFC_DATA);
arc_space_return(size, ARC_SPACE_DATA);
}
if (free_rdata) {
l2arc_free_abd_on_write(hdr->b_crypt_hdr.b_rabd, size, type);
} else {
l2arc_free_abd_on_write(hdr->b_l1hdr.b_pabd, size, type);
}
}
/*
* Share the arc_buf_t's data with the hdr. Whenever we are sharing the
* data buffer, we transfer the refcount ownership to the hdr and update
* the appropriate kstats.
*/
static void
arc_share_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf)
{
ASSERT(arc_can_share(hdr, buf));
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
ASSERT(!ARC_BUF_ENCRYPTED(buf));
ASSERT(HDR_EMPTY_OR_LOCKED(hdr));
/*
* Start sharing the data buffer. We transfer the
* refcount ownership to the hdr since it always owns
* the refcount whenever an arc_buf_t is shared.
*/
zfs_refcount_transfer_ownership_many(&hdr->b_l1hdr.b_state->arcs_size,
arc_hdr_size(hdr), buf, hdr);
hdr->b_l1hdr.b_pabd = abd_get_from_buf(buf->b_data, arc_buf_size(buf));
abd_take_ownership_of_buf(hdr->b_l1hdr.b_pabd,
HDR_ISTYPE_METADATA(hdr));
arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA);
buf->b_flags |= ARC_BUF_FLAG_SHARED;
/*
* Since we've transferred ownership to the hdr we need
* to increment its compressed and uncompressed kstats and
* decrement the overhead size.
*/
ARCSTAT_INCR(arcstat_compressed_size, arc_hdr_size(hdr));
ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr));
ARCSTAT_INCR(arcstat_overhead_size, -arc_buf_size(buf));
}
static void
arc_unshare_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf)
{
ASSERT(arc_buf_is_shared(buf));
ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
ASSERT(HDR_EMPTY_OR_LOCKED(hdr));
/*
* We are no longer sharing this buffer so we need
* to transfer its ownership to the rightful owner.
*/
zfs_refcount_transfer_ownership_many(&hdr->b_l1hdr.b_state->arcs_size,
arc_hdr_size(hdr), hdr, buf);
arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA);
abd_release_ownership_of_buf(hdr->b_l1hdr.b_pabd);
abd_free(hdr->b_l1hdr.b_pabd);
hdr->b_l1hdr.b_pabd = NULL;
buf->b_flags &= ~ARC_BUF_FLAG_SHARED;
/*
* Since the buffer is no longer shared between
* the arc buf and the hdr, count it as overhead.
*/
ARCSTAT_INCR(arcstat_compressed_size, -arc_hdr_size(hdr));
ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr));
ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf));
}
/*
* Remove an arc_buf_t from the hdr's buf list and return the last
* arc_buf_t on the list. If no buffers remain on the list then return
* NULL.
*/
static arc_buf_t *
arc_buf_remove(arc_buf_hdr_t *hdr, arc_buf_t *buf)
{
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT(HDR_EMPTY_OR_LOCKED(hdr));
arc_buf_t **bufp = &hdr->b_l1hdr.b_buf;
arc_buf_t *lastbuf = NULL;
/*
* Remove the buf from the hdr list and locate the last
* remaining buffer on the list.
*/
while (*bufp != NULL) {
if (*bufp == buf)
*bufp = buf->b_next;
/*
* If we've removed a buffer in the middle of
* the list then update the lastbuf and update
* bufp.
*/
if (*bufp != NULL) {
lastbuf = *bufp;
bufp = &(*bufp)->b_next;
}
}
buf->b_next = NULL;
ASSERT3P(lastbuf, !=, buf);
IMPLY(hdr->b_l1hdr.b_bufcnt > 0, lastbuf != NULL);
IMPLY(hdr->b_l1hdr.b_bufcnt > 0, hdr->b_l1hdr.b_buf != NULL);
IMPLY(lastbuf != NULL, ARC_BUF_LAST(lastbuf));
return (lastbuf);
}
/*
* Free up buf->b_data and pull the arc_buf_t off of the arc_buf_hdr_t's
* list and free it.
*/
static void
arc_buf_destroy_impl(arc_buf_t *buf)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
/*
* Free up the data associated with the buf but only if we're not
* sharing this with the hdr. If we are sharing it with the hdr, the
* hdr is responsible for doing the free.
*/
if (buf->b_data != NULL) {
/*
* We're about to change the hdr's b_flags. We must either
* hold the hash_lock or be undiscoverable.
*/
ASSERT(HDR_EMPTY_OR_LOCKED(hdr));
arc_cksum_verify(buf);
arc_buf_unwatch(buf);
if (arc_buf_is_shared(buf)) {
arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA);
} else {
uint64_t size = arc_buf_size(buf);
arc_free_data_buf(hdr, buf->b_data, size, buf);
ARCSTAT_INCR(arcstat_overhead_size, -size);
}
buf->b_data = NULL;
ASSERT(hdr->b_l1hdr.b_bufcnt > 0);
hdr->b_l1hdr.b_bufcnt -= 1;
if (ARC_BUF_ENCRYPTED(buf)) {
hdr->b_crypt_hdr.b_ebufcnt -= 1;
/*
* If we have no more encrypted buffers and we've
* already gotten a copy of the decrypted data we can
* free b_rabd to save some space.
*/
if (hdr->b_crypt_hdr.b_ebufcnt == 0 &&
HDR_HAS_RABD(hdr) && hdr->b_l1hdr.b_pabd != NULL &&
!HDR_IO_IN_PROGRESS(hdr)) {
arc_hdr_free_abd(hdr, B_TRUE);
}
}
}
arc_buf_t *lastbuf = arc_buf_remove(hdr, buf);
if (ARC_BUF_SHARED(buf) && !ARC_BUF_COMPRESSED(buf)) {
/*
* If the current arc_buf_t is sharing its data buffer with the
* hdr, then reassign the hdr's b_pabd to share it with the new
* buffer at the end of the list. The shared buffer is always
* the last one on the hdr's buffer list.
*
* There is an equivalent case for compressed bufs, but since
* they aren't guaranteed to be the last buf in the list and
* that is an exceedingly rare case, we just allow that space be
* wasted temporarily. We must also be careful not to share
* encrypted buffers, since they cannot be shared.
*/
if (lastbuf != NULL && !ARC_BUF_ENCRYPTED(lastbuf)) {
/* Only one buf can be shared at once */
VERIFY(!arc_buf_is_shared(lastbuf));
/* hdr is uncompressed so can't have compressed buf */
VERIFY(!ARC_BUF_COMPRESSED(lastbuf));
ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
arc_hdr_free_abd(hdr, B_FALSE);
/*
* We must setup a new shared block between the
* last buffer and the hdr. The data would have
* been allocated by the arc buf so we need to transfer
* ownership to the hdr since it's now being shared.
*/
arc_share_buf(hdr, lastbuf);
}
} else if (HDR_SHARED_DATA(hdr)) {
/*
* Uncompressed shared buffers are always at the end
* of the list. Compressed buffers don't have the
* same requirements. This makes it hard to
* simply assert that the lastbuf is shared so
* we rely on the hdr's compression flags to determine
* if we have a compressed, shared buffer.
*/
ASSERT3P(lastbuf, !=, NULL);
ASSERT(arc_buf_is_shared(lastbuf) ||
arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF);
}
/*
* Free the checksum if we're removing the last uncompressed buf from
* this hdr.
*/
if (!arc_hdr_has_uncompressed_buf(hdr)) {
arc_cksum_free(hdr);
}
/* clean up the buf */
buf->b_hdr = NULL;
kmem_cache_free(buf_cache, buf);
}
static void
arc_hdr_alloc_abd(arc_buf_hdr_t *hdr, int alloc_flags)
{
uint64_t size;
boolean_t alloc_rdata = ((alloc_flags & ARC_HDR_ALLOC_RDATA) != 0);
boolean_t do_adapt = ((alloc_flags & ARC_HDR_DO_ADAPT) != 0);
ASSERT3U(HDR_GET_LSIZE(hdr), >, 0);
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT(!HDR_SHARED_DATA(hdr) || alloc_rdata);
IMPLY(alloc_rdata, HDR_PROTECTED(hdr));
if (alloc_rdata) {
size = HDR_GET_PSIZE(hdr);
ASSERT3P(hdr->b_crypt_hdr.b_rabd, ==, NULL);
hdr->b_crypt_hdr.b_rabd = arc_get_data_abd(hdr, size, hdr,
do_adapt);
ASSERT3P(hdr->b_crypt_hdr.b_rabd, !=, NULL);
ARCSTAT_INCR(arcstat_raw_size, size);
} else {
size = arc_hdr_size(hdr);
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
hdr->b_l1hdr.b_pabd = arc_get_data_abd(hdr, size, hdr,
do_adapt);
ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
}
ARCSTAT_INCR(arcstat_compressed_size, size);
ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr));
}
static void
arc_hdr_free_abd(arc_buf_hdr_t *hdr, boolean_t free_rdata)
{
uint64_t size = (free_rdata) ? HDR_GET_PSIZE(hdr) : arc_hdr_size(hdr);
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr));
IMPLY(free_rdata, HDR_HAS_RABD(hdr));
/*
* If the hdr is currently being written to the l2arc then
* we defer freeing the data by adding it to the l2arc_free_on_write
* list. The l2arc will free the data once it's finished
* writing it to the l2arc device.
*/
if (HDR_L2_WRITING(hdr)) {
arc_hdr_free_on_write(hdr, free_rdata);
ARCSTAT_BUMP(arcstat_l2_free_on_write);
} else if (free_rdata) {
arc_free_data_abd(hdr, hdr->b_crypt_hdr.b_rabd, size, hdr);
} else {
arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd, size, hdr);
}
if (free_rdata) {
hdr->b_crypt_hdr.b_rabd = NULL;
ARCSTAT_INCR(arcstat_raw_size, -size);
} else {
hdr->b_l1hdr.b_pabd = NULL;
}
if (hdr->b_l1hdr.b_pabd == NULL && !HDR_HAS_RABD(hdr))
hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS;
ARCSTAT_INCR(arcstat_compressed_size, -size);
ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr));
}
static arc_buf_hdr_t *
arc_hdr_alloc(uint64_t spa, int32_t psize, int32_t lsize,
boolean_t protected, enum zio_compress compression_type, uint8_t complevel,
arc_buf_contents_t type, boolean_t alloc_rdata)
{
arc_buf_hdr_t *hdr;
int flags = ARC_HDR_DO_ADAPT;
VERIFY(type == ARC_BUFC_DATA || type == ARC_BUFC_METADATA);
if (protected) {
hdr = kmem_cache_alloc(hdr_full_crypt_cache, KM_PUSHPAGE);
} else {
hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE);
}
flags |= alloc_rdata ? ARC_HDR_ALLOC_RDATA : 0;
ASSERT(HDR_EMPTY(hdr));
ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL);
HDR_SET_PSIZE(hdr, psize);
HDR_SET_LSIZE(hdr, lsize);
hdr->b_spa = spa;
hdr->b_type = type;
hdr->b_flags = 0;
arc_hdr_set_flags(hdr, arc_bufc_to_flags(type) | ARC_FLAG_HAS_L1HDR);
arc_hdr_set_compress(hdr, compression_type);
hdr->b_complevel = complevel;
if (protected)
arc_hdr_set_flags(hdr, ARC_FLAG_PROTECTED);
hdr->b_l1hdr.b_state = arc_anon;
hdr->b_l1hdr.b_arc_access = 0;
hdr->b_l1hdr.b_bufcnt = 0;
hdr->b_l1hdr.b_buf = NULL;
/*
* Allocate the hdr's buffer. This will contain either
* the compressed or uncompressed data depending on the block
* it references and compressed arc enablement.
*/
arc_hdr_alloc_abd(hdr, flags);
ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
return (hdr);
}
/*
* Transition between the two allocation states for the arc_buf_hdr struct.
* The arc_buf_hdr struct can be allocated with (hdr_full_cache) or without
* (hdr_l2only_cache) the fields necessary for the L1 cache - the smaller
* version is used when a cache buffer is only in the L2ARC in order to reduce
* memory usage.
*/
static arc_buf_hdr_t *
arc_hdr_realloc(arc_buf_hdr_t *hdr, kmem_cache_t *old, kmem_cache_t *new)
{
ASSERT(HDR_HAS_L2HDR(hdr));
arc_buf_hdr_t *nhdr;
l2arc_dev_t *dev = hdr->b_l2hdr.b_dev;
ASSERT((old == hdr_full_cache && new == hdr_l2only_cache) ||
(old == hdr_l2only_cache && new == hdr_full_cache));
/*
* if the caller wanted a new full header and the header is to be
* encrypted we will actually allocate the header from the full crypt
* cache instead. The same applies to freeing from the old cache.
*/
if (HDR_PROTECTED(hdr) && new == hdr_full_cache)
new = hdr_full_crypt_cache;
if (HDR_PROTECTED(hdr) && old == hdr_full_cache)
old = hdr_full_crypt_cache;
nhdr = kmem_cache_alloc(new, KM_PUSHPAGE);
ASSERT(MUTEX_HELD(HDR_LOCK(hdr)));
buf_hash_remove(hdr);
bcopy(hdr, nhdr, HDR_L2ONLY_SIZE);
if (new == hdr_full_cache || new == hdr_full_crypt_cache) {
arc_hdr_set_flags(nhdr, ARC_FLAG_HAS_L1HDR);
/*
* arc_access and arc_change_state need to be aware that a
* header has just come out of L2ARC, so we set its state to
* l2c_only even though it's about to change.
*/
nhdr->b_l1hdr.b_state = arc_l2c_only;
/* Verify previous threads set to NULL before freeing */
ASSERT3P(nhdr->b_l1hdr.b_pabd, ==, NULL);
ASSERT(!HDR_HAS_RABD(hdr));
} else {
ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL);
ASSERT0(hdr->b_l1hdr.b_bufcnt);
ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL);
/*
* If we've reached here, We must have been called from
* arc_evict_hdr(), as such we should have already been
* removed from any ghost list we were previously on
* (which protects us from racing with arc_evict_state),
* thus no locking is needed during this check.
*/
ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node));
/*
* A buffer must not be moved into the arc_l2c_only
* state if it's not finished being written out to the
* l2arc device. Otherwise, the b_l1hdr.b_pabd field
* might try to be accessed, even though it was removed.
*/
VERIFY(!HDR_L2_WRITING(hdr));
VERIFY3P(hdr->b_l1hdr.b_pabd, ==, NULL);
ASSERT(!HDR_HAS_RABD(hdr));
arc_hdr_clear_flags(nhdr, ARC_FLAG_HAS_L1HDR);
}
/*
* The header has been reallocated so we need to re-insert it into any
* lists it was on.
*/
(void) buf_hash_insert(nhdr, NULL);
ASSERT(list_link_active(&hdr->b_l2hdr.b_l2node));
mutex_enter(&dev->l2ad_mtx);
/*
* We must place the realloc'ed header back into the list at
* the same spot. Otherwise, if it's placed earlier in the list,
* l2arc_write_buffers() could find it during the function's
* write phase, and try to write it out to the l2arc.
*/
list_insert_after(&dev->l2ad_buflist, hdr, nhdr);
list_remove(&dev->l2ad_buflist, hdr);
mutex_exit(&dev->l2ad_mtx);
/*
* Since we're using the pointer address as the tag when
* incrementing and decrementing the l2ad_alloc refcount, we
* must remove the old pointer (that we're about to destroy) and
* add the new pointer to the refcount. Otherwise we'd remove
* the wrong pointer address when calling arc_hdr_destroy() later.
*/
(void) zfs_refcount_remove_many(&dev->l2ad_alloc,
arc_hdr_size(hdr), hdr);
(void) zfs_refcount_add_many(&dev->l2ad_alloc,
arc_hdr_size(nhdr), nhdr);
buf_discard_identity(hdr);
kmem_cache_free(old, hdr);
return (nhdr);
}
/*
* This function allows an L1 header to be reallocated as a crypt
* header and vice versa. If we are going to a crypt header, the
* new fields will be zeroed out.
*/
static arc_buf_hdr_t *
arc_hdr_realloc_crypt(arc_buf_hdr_t *hdr, boolean_t need_crypt)
{
arc_buf_hdr_t *nhdr;
arc_buf_t *buf;
kmem_cache_t *ncache, *ocache;
/*
* This function requires that hdr is in the arc_anon state.
* Therefore it won't have any L2ARC data for us to worry
* about copying.
*/
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT(!HDR_HAS_L2HDR(hdr));
ASSERT3U(!!HDR_PROTECTED(hdr), !=, need_crypt);
ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon);
ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node));
ASSERT(!list_link_active(&hdr->b_l2hdr.b_l2node));
ASSERT3P(hdr->b_hash_next, ==, NULL);
if (need_crypt) {
ncache = hdr_full_crypt_cache;
ocache = hdr_full_cache;
} else {
ncache = hdr_full_cache;
ocache = hdr_full_crypt_cache;
}
nhdr = kmem_cache_alloc(ncache, KM_PUSHPAGE);
/*
* Copy all members that aren't locks or condvars to the new header.
* No lists are pointing to us (as we asserted above), so we don't
* need to worry about the list nodes.
*/
nhdr->b_dva = hdr->b_dva;
nhdr->b_birth = hdr->b_birth;
nhdr->b_type = hdr->b_type;
nhdr->b_flags = hdr->b_flags;
nhdr->b_psize = hdr->b_psize;
nhdr->b_lsize = hdr->b_lsize;
nhdr->b_spa = hdr->b_spa;
nhdr->b_l1hdr.b_freeze_cksum = hdr->b_l1hdr.b_freeze_cksum;
nhdr->b_l1hdr.b_bufcnt = hdr->b_l1hdr.b_bufcnt;
nhdr->b_l1hdr.b_byteswap = hdr->b_l1hdr.b_byteswap;
nhdr->b_l1hdr.b_state = hdr->b_l1hdr.b_state;
nhdr->b_l1hdr.b_arc_access = hdr->b_l1hdr.b_arc_access;
nhdr->b_l1hdr.b_mru_hits = hdr->b_l1hdr.b_mru_hits;
nhdr->b_l1hdr.b_mru_ghost_hits = hdr->b_l1hdr.b_mru_ghost_hits;
nhdr->b_l1hdr.b_mfu_hits = hdr->b_l1hdr.b_mfu_hits;
nhdr->b_l1hdr.b_mfu_ghost_hits = hdr->b_l1hdr.b_mfu_ghost_hits;
nhdr->b_l1hdr.b_l2_hits = hdr->b_l1hdr.b_l2_hits;
nhdr->b_l1hdr.b_acb = hdr->b_l1hdr.b_acb;
nhdr->b_l1hdr.b_pabd = hdr->b_l1hdr.b_pabd;
/*
* This zfs_refcount_add() exists only to ensure that the individual
* arc buffers always point to a header that is referenced, avoiding
* a small race condition that could trigger ASSERTs.
*/
(void) zfs_refcount_add(&nhdr->b_l1hdr.b_refcnt, FTAG);
nhdr->b_l1hdr.b_buf = hdr->b_l1hdr.b_buf;
for (buf = nhdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) {
mutex_enter(&buf->b_evict_lock);
buf->b_hdr = nhdr;
mutex_exit(&buf->b_evict_lock);
}
zfs_refcount_transfer(&nhdr->b_l1hdr.b_refcnt, &hdr->b_l1hdr.b_refcnt);
(void) zfs_refcount_remove(&nhdr->b_l1hdr.b_refcnt, FTAG);
ASSERT0(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt));
if (need_crypt) {
arc_hdr_set_flags(nhdr, ARC_FLAG_PROTECTED);
} else {
arc_hdr_clear_flags(nhdr, ARC_FLAG_PROTECTED);
}
/* unset all members of the original hdr */
bzero(&hdr->b_dva, sizeof (dva_t));
hdr->b_birth = 0;
hdr->b_type = ARC_BUFC_INVALID;
hdr->b_flags = 0;
hdr->b_psize = 0;
hdr->b_lsize = 0;
hdr->b_spa = 0;
hdr->b_l1hdr.b_freeze_cksum = NULL;
hdr->b_l1hdr.b_buf = NULL;
hdr->b_l1hdr.b_bufcnt = 0;
hdr->b_l1hdr.b_byteswap = 0;
hdr->b_l1hdr.b_state = NULL;
hdr->b_l1hdr.b_arc_access = 0;
hdr->b_l1hdr.b_mru_hits = 0;
hdr->b_l1hdr.b_mru_ghost_hits = 0;
hdr->b_l1hdr.b_mfu_hits = 0;
hdr->b_l1hdr.b_mfu_ghost_hits = 0;
hdr->b_l1hdr.b_l2_hits = 0;
hdr->b_l1hdr.b_acb = NULL;
hdr->b_l1hdr.b_pabd = NULL;
if (ocache == hdr_full_crypt_cache) {
ASSERT(!HDR_HAS_RABD(hdr));
hdr->b_crypt_hdr.b_ot = DMU_OT_NONE;
hdr->b_crypt_hdr.b_ebufcnt = 0;
hdr->b_crypt_hdr.b_dsobj = 0;
bzero(hdr->b_crypt_hdr.b_salt, ZIO_DATA_SALT_LEN);
bzero(hdr->b_crypt_hdr.b_iv, ZIO_DATA_IV_LEN);
bzero(hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN);
}
buf_discard_identity(hdr);
kmem_cache_free(ocache, hdr);
return (nhdr);
}
/*
* This function is used by the send / receive code to convert a newly
* allocated arc_buf_t to one that is suitable for a raw encrypted write. It
* is also used to allow the root objset block to be updated without altering
* its embedded MACs. Both block types will always be uncompressed so we do not
* have to worry about compression type or psize.
*/
void
arc_convert_to_raw(arc_buf_t *buf, uint64_t dsobj, boolean_t byteorder,
dmu_object_type_t ot, const uint8_t *salt, const uint8_t *iv,
const uint8_t *mac)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
ASSERT(ot == DMU_OT_DNODE || ot == DMU_OT_OBJSET);
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon);
buf->b_flags |= (ARC_BUF_FLAG_COMPRESSED | ARC_BUF_FLAG_ENCRYPTED);
if (!HDR_PROTECTED(hdr))
hdr = arc_hdr_realloc_crypt(hdr, B_TRUE);
hdr->b_crypt_hdr.b_dsobj = dsobj;
hdr->b_crypt_hdr.b_ot = ot;
hdr->b_l1hdr.b_byteswap = (byteorder == ZFS_HOST_BYTEORDER) ?
DMU_BSWAP_NUMFUNCS : DMU_OT_BYTESWAP(ot);
if (!arc_hdr_has_uncompressed_buf(hdr))
arc_cksum_free(hdr);
if (salt != NULL)
bcopy(salt, hdr->b_crypt_hdr.b_salt, ZIO_DATA_SALT_LEN);
if (iv != NULL)
bcopy(iv, hdr->b_crypt_hdr.b_iv, ZIO_DATA_IV_LEN);
if (mac != NULL)
bcopy(mac, hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN);
}
/*
* Allocate a new arc_buf_hdr_t and arc_buf_t and return the buf to the caller.
* The buf is returned thawed since we expect the consumer to modify it.
*/
arc_buf_t *
arc_alloc_buf(spa_t *spa, void *tag, arc_buf_contents_t type, int32_t size)
{
arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), size, size,
B_FALSE, ZIO_COMPRESS_OFF, 0, type, B_FALSE);
arc_buf_t *buf = NULL;
VERIFY0(arc_buf_alloc_impl(hdr, spa, NULL, tag, B_FALSE, B_FALSE,
B_FALSE, B_FALSE, &buf));
arc_buf_thaw(buf);
return (buf);
}
/*
* Allocate a compressed buf in the same manner as arc_alloc_buf. Don't use this
* for bufs containing metadata.
*/
arc_buf_t *
arc_alloc_compressed_buf(spa_t *spa, void *tag, uint64_t psize, uint64_t lsize,
enum zio_compress compression_type, uint8_t complevel)
{
ASSERT3U(lsize, >, 0);
ASSERT3U(lsize, >=, psize);
ASSERT3U(compression_type, >, ZIO_COMPRESS_OFF);
ASSERT3U(compression_type, <, ZIO_COMPRESS_FUNCTIONS);
arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize,
B_FALSE, compression_type, complevel, ARC_BUFC_DATA, B_FALSE);
arc_buf_t *buf = NULL;
VERIFY0(arc_buf_alloc_impl(hdr, spa, NULL, tag, B_FALSE,
B_TRUE, B_FALSE, B_FALSE, &buf));
arc_buf_thaw(buf);
ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL);
if (!arc_buf_is_shared(buf)) {
/*
* To ensure that the hdr has the correct data in it if we call
* arc_untransform() on this buf before it's been written to
* disk, it's easiest if we just set up sharing between the
* buf and the hdr.
*/
arc_hdr_free_abd(hdr, B_FALSE);
arc_share_buf(hdr, buf);
}
return (buf);
}
arc_buf_t *
arc_alloc_raw_buf(spa_t *spa, void *tag, uint64_t dsobj, boolean_t byteorder,
const uint8_t *salt, const uint8_t *iv, const uint8_t *mac,
dmu_object_type_t ot, uint64_t psize, uint64_t lsize,
enum zio_compress compression_type, uint8_t complevel)
{
arc_buf_hdr_t *hdr;
arc_buf_t *buf;
arc_buf_contents_t type = DMU_OT_IS_METADATA(ot) ?
ARC_BUFC_METADATA : ARC_BUFC_DATA;
ASSERT3U(lsize, >, 0);
ASSERT3U(lsize, >=, psize);
ASSERT3U(compression_type, >=, ZIO_COMPRESS_OFF);
ASSERT3U(compression_type, <, ZIO_COMPRESS_FUNCTIONS);
hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, B_TRUE,
compression_type, complevel, type, B_TRUE);
hdr->b_crypt_hdr.b_dsobj = dsobj;
hdr->b_crypt_hdr.b_ot = ot;
hdr->b_l1hdr.b_byteswap = (byteorder == ZFS_HOST_BYTEORDER) ?
DMU_BSWAP_NUMFUNCS : DMU_OT_BYTESWAP(ot);
bcopy(salt, hdr->b_crypt_hdr.b_salt, ZIO_DATA_SALT_LEN);
bcopy(iv, hdr->b_crypt_hdr.b_iv, ZIO_DATA_IV_LEN);
bcopy(mac, hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN);
/*
* This buffer will be considered encrypted even if the ot is not an
* encrypted type. It will become authenticated instead in
* arc_write_ready().
*/
buf = NULL;
VERIFY0(arc_buf_alloc_impl(hdr, spa, NULL, tag, B_TRUE, B_TRUE,
B_FALSE, B_FALSE, &buf));
arc_buf_thaw(buf);
ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL);
return (buf);
}
static void
l2arc_hdr_arcstats_update(arc_buf_hdr_t *hdr, boolean_t incr,
boolean_t state_only)
{
l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr;
l2arc_dev_t *dev = l2hdr->b_dev;
uint64_t lsize = HDR_GET_LSIZE(hdr);
uint64_t psize = HDR_GET_PSIZE(hdr);
uint64_t asize = vdev_psize_to_asize(dev->l2ad_vdev, psize);
arc_buf_contents_t type = hdr->b_type;
int64_t lsize_s;
int64_t psize_s;
int64_t asize_s;
if (incr) {
lsize_s = lsize;
psize_s = psize;
asize_s = asize;
} else {
lsize_s = -lsize;
psize_s = -psize;
asize_s = -asize;
}
/* If the buffer is a prefetch, count it as such. */
if (HDR_PREFETCH(hdr)) {
ARCSTAT_INCR(arcstat_l2_prefetch_asize, asize_s);
} else {
/*
* We use the value stored in the L2 header upon initial
* caching in L2ARC. This value will be updated in case
* an MRU/MRU_ghost buffer transitions to MFU but the L2ARC
* metadata (log entry) cannot currently be updated. Having
* the ARC state in the L2 header solves the problem of a
* possibly absent L1 header (apparent in buffers restored
* from persistent L2ARC).
*/
switch (hdr->b_l2hdr.b_arcs_state) {
case ARC_STATE_MRU_GHOST:
case ARC_STATE_MRU:
ARCSTAT_INCR(arcstat_l2_mru_asize, asize_s);
break;
case ARC_STATE_MFU_GHOST:
case ARC_STATE_MFU:
ARCSTAT_INCR(arcstat_l2_mfu_asize, asize_s);
break;
default:
break;
}
}
if (state_only)
return;
ARCSTAT_INCR(arcstat_l2_psize, psize_s);
ARCSTAT_INCR(arcstat_l2_lsize, lsize_s);
switch (type) {
case ARC_BUFC_DATA:
ARCSTAT_INCR(arcstat_l2_bufc_data_asize, asize_s);
break;
case ARC_BUFC_METADATA:
ARCSTAT_INCR(arcstat_l2_bufc_metadata_asize, asize_s);
break;
default:
break;
}
}
static void
arc_hdr_l2hdr_destroy(arc_buf_hdr_t *hdr)
{
l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr;
l2arc_dev_t *dev = l2hdr->b_dev;
uint64_t psize = HDR_GET_PSIZE(hdr);
uint64_t asize = vdev_psize_to_asize(dev->l2ad_vdev, psize);
ASSERT(MUTEX_HELD(&dev->l2ad_mtx));
ASSERT(HDR_HAS_L2HDR(hdr));
list_remove(&dev->l2ad_buflist, hdr);
l2arc_hdr_arcstats_decrement(hdr);
vdev_space_update(dev->l2ad_vdev, -asize, 0, 0);
(void) zfs_refcount_remove_many(&dev->l2ad_alloc, arc_hdr_size(hdr),
hdr);
arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR);
}
static void
arc_hdr_destroy(arc_buf_hdr_t *hdr)
{
if (HDR_HAS_L1HDR(hdr)) {
ASSERT(hdr->b_l1hdr.b_buf == NULL ||
hdr->b_l1hdr.b_bufcnt > 0);
ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon);
}
ASSERT(!HDR_IO_IN_PROGRESS(hdr));
ASSERT(!HDR_IN_HASH_TABLE(hdr));
if (HDR_HAS_L2HDR(hdr)) {
l2arc_dev_t *dev = hdr->b_l2hdr.b_dev;
boolean_t buflist_held = MUTEX_HELD(&dev->l2ad_mtx);
if (!buflist_held)
mutex_enter(&dev->l2ad_mtx);
/*
* Even though we checked this conditional above, we
* need to check this again now that we have the
* l2ad_mtx. This is because we could be racing with
* another thread calling l2arc_evict() which might have
* destroyed this header's L2 portion as we were waiting
* to acquire the l2ad_mtx. If that happens, we don't
* want to re-destroy the header's L2 portion.
*/
if (HDR_HAS_L2HDR(hdr))
arc_hdr_l2hdr_destroy(hdr);
if (!buflist_held)
mutex_exit(&dev->l2ad_mtx);
}
/*
* The header's identify can only be safely discarded once it is no
* longer discoverable. This requires removing it from the hash table
* and the l2arc header list. After this point the hash lock can not
* be used to protect the header.
*/
if (!HDR_EMPTY(hdr))
buf_discard_identity(hdr);
if (HDR_HAS_L1HDR(hdr)) {
arc_cksum_free(hdr);
while (hdr->b_l1hdr.b_buf != NULL)
arc_buf_destroy_impl(hdr->b_l1hdr.b_buf);
if (hdr->b_l1hdr.b_pabd != NULL)
arc_hdr_free_abd(hdr, B_FALSE);
if (HDR_HAS_RABD(hdr))
arc_hdr_free_abd(hdr, B_TRUE);
}
ASSERT3P(hdr->b_hash_next, ==, NULL);
if (HDR_HAS_L1HDR(hdr)) {
ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node));
ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL);
if (!HDR_PROTECTED(hdr)) {
kmem_cache_free(hdr_full_cache, hdr);
} else {
kmem_cache_free(hdr_full_crypt_cache, hdr);
}
} else {
kmem_cache_free(hdr_l2only_cache, hdr);
}
}
void
arc_buf_destroy(arc_buf_t *buf, void* tag)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
if (hdr->b_l1hdr.b_state == arc_anon) {
ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1);
ASSERT(!HDR_IO_IN_PROGRESS(hdr));
VERIFY0(remove_reference(hdr, NULL, tag));
arc_hdr_destroy(hdr);
return;
}
kmutex_t *hash_lock = HDR_LOCK(hdr);
mutex_enter(hash_lock);
ASSERT3P(hdr, ==, buf->b_hdr);
ASSERT(hdr->b_l1hdr.b_bufcnt > 0);
ASSERT3P(hash_lock, ==, HDR_LOCK(hdr));
ASSERT3P(hdr->b_l1hdr.b_state, !=, arc_anon);
ASSERT3P(buf->b_data, !=, NULL);
(void) remove_reference(hdr, hash_lock, tag);
arc_buf_destroy_impl(buf);
mutex_exit(hash_lock);
}
/*
* Evict the arc_buf_hdr that is provided as a parameter. The resultant
* state of the header is dependent on its state prior to entering this
* function. The following transitions are possible:
*
* - arc_mru -> arc_mru_ghost
* - arc_mfu -> arc_mfu_ghost
* - arc_mru_ghost -> arc_l2c_only
* - arc_mru_ghost -> deleted
* - arc_mfu_ghost -> arc_l2c_only
* - arc_mfu_ghost -> deleted
*/
static int64_t
arc_evict_hdr(arc_buf_hdr_t *hdr, kmutex_t *hash_lock)
{
arc_state_t *evicted_state, *state;
int64_t bytes_evicted = 0;
int min_lifetime = HDR_PRESCIENT_PREFETCH(hdr) ?
arc_min_prescient_prefetch_ms : arc_min_prefetch_ms;
ASSERT(MUTEX_HELD(hash_lock));
ASSERT(HDR_HAS_L1HDR(hdr));
state = hdr->b_l1hdr.b_state;
if (GHOST_STATE(state)) {
ASSERT(!HDR_IO_IN_PROGRESS(hdr));
ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL);
/*
* l2arc_write_buffers() relies on a header's L1 portion
* (i.e. its b_pabd field) during it's write phase.
* Thus, we cannot push a header onto the arc_l2c_only
* state (removing its L1 piece) until the header is
* done being written to the l2arc.
*/
if (HDR_HAS_L2HDR(hdr) && HDR_L2_WRITING(hdr)) {
ARCSTAT_BUMP(arcstat_evict_l2_skip);
return (bytes_evicted);
}
ARCSTAT_BUMP(arcstat_deleted);
bytes_evicted += HDR_GET_LSIZE(hdr);
DTRACE_PROBE1(arc__delete, arc_buf_hdr_t *, hdr);
if (HDR_HAS_L2HDR(hdr)) {
ASSERT(hdr->b_l1hdr.b_pabd == NULL);
ASSERT(!HDR_HAS_RABD(hdr));
/*
* This buffer is cached on the 2nd Level ARC;
* don't destroy the header.
*/
arc_change_state(arc_l2c_only, hdr, hash_lock);
/*
* dropping from L1+L2 cached to L2-only,
* realloc to remove the L1 header.
*/
hdr = arc_hdr_realloc(hdr, hdr_full_cache,
hdr_l2only_cache);
} else {
arc_change_state(arc_anon, hdr, hash_lock);
arc_hdr_destroy(hdr);
}
return (bytes_evicted);
}
ASSERT(state == arc_mru || state == arc_mfu);
evicted_state = (state == arc_mru) ? arc_mru_ghost : arc_mfu_ghost;
/* prefetch buffers have a minimum lifespan */
if (HDR_IO_IN_PROGRESS(hdr) ||
((hdr->b_flags & (ARC_FLAG_PREFETCH | ARC_FLAG_INDIRECT)) &&
ddi_get_lbolt() - hdr->b_l1hdr.b_arc_access <
MSEC_TO_TICK(min_lifetime))) {
ARCSTAT_BUMP(arcstat_evict_skip);
return (bytes_evicted);
}
ASSERT0(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt));
while (hdr->b_l1hdr.b_buf) {
arc_buf_t *buf = hdr->b_l1hdr.b_buf;
if (!mutex_tryenter(&buf->b_evict_lock)) {
ARCSTAT_BUMP(arcstat_mutex_miss);
break;
}
if (buf->b_data != NULL)
bytes_evicted += HDR_GET_LSIZE(hdr);
mutex_exit(&buf->b_evict_lock);
arc_buf_destroy_impl(buf);
}
if (HDR_HAS_L2HDR(hdr)) {
ARCSTAT_INCR(arcstat_evict_l2_cached, HDR_GET_LSIZE(hdr));
} else {
if (l2arc_write_eligible(hdr->b_spa, hdr)) {
ARCSTAT_INCR(arcstat_evict_l2_eligible,
HDR_GET_LSIZE(hdr));
switch (state->arcs_state) {
case ARC_STATE_MRU:
ARCSTAT_INCR(
arcstat_evict_l2_eligible_mru,
HDR_GET_LSIZE(hdr));
break;
case ARC_STATE_MFU:
ARCSTAT_INCR(
arcstat_evict_l2_eligible_mfu,
HDR_GET_LSIZE(hdr));
break;
default:
break;
}
} else {
ARCSTAT_INCR(arcstat_evict_l2_ineligible,
HDR_GET_LSIZE(hdr));
}
}
if (hdr->b_l1hdr.b_bufcnt == 0) {
arc_cksum_free(hdr);
bytes_evicted += arc_hdr_size(hdr);
/*
* If this hdr is being evicted and has a compressed
* buffer then we discard it here before we change states.
* This ensures that the accounting is updated correctly
* in arc_free_data_impl().
*/
if (hdr->b_l1hdr.b_pabd != NULL)
arc_hdr_free_abd(hdr, B_FALSE);
if (HDR_HAS_RABD(hdr))
arc_hdr_free_abd(hdr, B_TRUE);
arc_change_state(evicted_state, hdr, hash_lock);
ASSERT(HDR_IN_HASH_TABLE(hdr));
arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE);
DTRACE_PROBE1(arc__evict, arc_buf_hdr_t *, hdr);
}
return (bytes_evicted);
}
static void
arc_set_need_free(void)
{
ASSERT(MUTEX_HELD(&arc_evict_lock));
int64_t remaining = arc_free_memory() - arc_sys_free / 2;
arc_evict_waiter_t *aw = list_tail(&arc_evict_waiters);
if (aw == NULL) {
arc_need_free = MAX(-remaining, 0);
} else {
arc_need_free =
MAX(-remaining, (int64_t)(aw->aew_count - arc_evict_count));
}
}
static uint64_t
arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker,
uint64_t spa, int64_t bytes)
{
multilist_sublist_t *mls;
uint64_t bytes_evicted = 0;
arc_buf_hdr_t *hdr;
kmutex_t *hash_lock;
int evict_count = 0;
ASSERT3P(marker, !=, NULL);
IMPLY(bytes < 0, bytes == ARC_EVICT_ALL);
mls = multilist_sublist_lock(ml, idx);
for (hdr = multilist_sublist_prev(mls, marker); hdr != NULL;
hdr = multilist_sublist_prev(mls, marker)) {
if ((bytes != ARC_EVICT_ALL && bytes_evicted >= bytes) ||
(evict_count >= zfs_arc_evict_batch_limit))
break;
/*
* To keep our iteration location, move the marker
* forward. Since we're not holding hdr's hash lock, we
* must be very careful and not remove 'hdr' from the
* sublist. Otherwise, other consumers might mistake the
* 'hdr' as not being on a sublist when they call the
* multilist_link_active() function (they all rely on
* the hash lock protecting concurrent insertions and
* removals). multilist_sublist_move_forward() was
* specifically implemented to ensure this is the case
* (only 'marker' will be removed and re-inserted).
*/
multilist_sublist_move_forward(mls, marker);
/*
* The only case where the b_spa field should ever be
* zero, is the marker headers inserted by
* arc_evict_state(). It's possible for multiple threads
* to be calling arc_evict_state() concurrently (e.g.
* dsl_pool_close() and zio_inject_fault()), so we must
* skip any markers we see from these other threads.
*/
if (hdr->b_spa == 0)
continue;
/* we're only interested in evicting buffers of a certain spa */
if (spa != 0 && hdr->b_spa != spa) {
ARCSTAT_BUMP(arcstat_evict_skip);
continue;
}
hash_lock = HDR_LOCK(hdr);
/*
* We aren't calling this function from any code path
* that would already be holding a hash lock, so we're
* asserting on this assumption to be defensive in case
* this ever changes. Without this check, it would be
* possible to incorrectly increment arcstat_mutex_miss
* below (e.g. if the code changed such that we called
* this function with a hash lock held).
*/
ASSERT(!MUTEX_HELD(hash_lock));
if (mutex_tryenter(hash_lock)) {
uint64_t evicted = arc_evict_hdr(hdr, hash_lock);
mutex_exit(hash_lock);
bytes_evicted += evicted;
/*
* If evicted is zero, arc_evict_hdr() must have
* decided to skip this header, don't increment
* evict_count in this case.
*/
if (evicted != 0)
evict_count++;
} else {
ARCSTAT_BUMP(arcstat_mutex_miss);
}
}
multilist_sublist_unlock(mls);
/*
* Increment the count of evicted bytes, and wake up any threads that
* are waiting for the count to reach this value. Since the list is
* ordered by ascending aew_count, we pop off the beginning of the
* list until we reach the end, or a waiter that's past the current
* "count". Doing this outside the loop reduces the number of times
* we need to acquire the global arc_evict_lock.
*
* Only wake when there's sufficient free memory in the system
* (specifically, arc_sys_free/2, which by default is a bit more than
* 1/64th of RAM). See the comments in arc_wait_for_eviction().
*/
mutex_enter(&arc_evict_lock);
arc_evict_count += bytes_evicted;
if (arc_free_memory() > arc_sys_free / 2) {
arc_evict_waiter_t *aw;
while ((aw = list_head(&arc_evict_waiters)) != NULL &&
aw->aew_count <= arc_evict_count) {
list_remove(&arc_evict_waiters, aw);
cv_broadcast(&aw->aew_cv);
}
}
arc_set_need_free();
mutex_exit(&arc_evict_lock);
/*
* If the ARC size is reduced from arc_c_max to arc_c_min (especially
* if the average cached block is small), eviction can be on-CPU for
* many seconds. To ensure that other threads that may be bound to
* this CPU are able to make progress, make a voluntary preemption
* call here.
*/
cond_resched();
return (bytes_evicted);
}
/*
* Evict buffers from the given arc state, until we've removed the
* specified number of bytes. Move the removed buffers to the
* appropriate evict state.
*
* This function makes a "best effort". It skips over any buffers
* it can't get a hash_lock on, and so, may not catch all candidates.
* It may also return without evicting as much space as requested.
*
* If bytes is specified using the special value ARC_EVICT_ALL, this
* will evict all available (i.e. unlocked and evictable) buffers from
* the given arc state; which is used by arc_flush().
*/
static uint64_t
arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes,
arc_buf_contents_t type)
{
uint64_t total_evicted = 0;
multilist_t *ml = &state->arcs_list[type];
int num_sublists;
arc_buf_hdr_t **markers;
IMPLY(bytes < 0, bytes == ARC_EVICT_ALL);
num_sublists = multilist_get_num_sublists(ml);
/*
* If we've tried to evict from each sublist, made some
* progress, but still have not hit the target number of bytes
* to evict, we want to keep trying. The markers allow us to
* pick up where we left off for each individual sublist, rather
* than starting from the tail each time.
*/
markers = kmem_zalloc(sizeof (*markers) * num_sublists, KM_SLEEP);
for (int i = 0; i < num_sublists; i++) {
multilist_sublist_t *mls;
markers[i] = kmem_cache_alloc(hdr_full_cache, KM_SLEEP);
/*
* A b_spa of 0 is used to indicate that this header is
* a marker. This fact is used in arc_evict_type() and
* arc_evict_state_impl().
*/
markers[i]->b_spa = 0;
mls = multilist_sublist_lock(ml, i);
multilist_sublist_insert_tail(mls, markers[i]);
multilist_sublist_unlock(mls);
}
/*
* While we haven't hit our target number of bytes to evict, or
* we're evicting all available buffers.
*/
while (total_evicted < bytes || bytes == ARC_EVICT_ALL) {
int sublist_idx = multilist_get_random_index(ml);
uint64_t scan_evicted = 0;
/*
* Try to reduce pinned dnodes with a floor of arc_dnode_limit.
* Request that 10% of the LRUs be scanned by the superblock
* shrinker.
*/
if (type == ARC_BUFC_DATA && aggsum_compare(
&arc_sums.arcstat_dnode_size, arc_dnode_size_limit) > 0) {
arc_prune_async((aggsum_upper_bound(
&arc_sums.arcstat_dnode_size) -
arc_dnode_size_limit) / sizeof (dnode_t) /
zfs_arc_dnode_reduce_percent);
}
/*
* Start eviction using a randomly selected sublist,
* this is to try and evenly balance eviction across all
* sublists. Always starting at the same sublist
* (e.g. index 0) would cause evictions to favor certain
* sublists over others.
*/
for (int i = 0; i < num_sublists; i++) {
uint64_t bytes_remaining;
uint64_t bytes_evicted;
if (bytes == ARC_EVICT_ALL)
bytes_remaining = ARC_EVICT_ALL;
else if (total_evicted < bytes)
bytes_remaining = bytes - total_evicted;
else
break;
bytes_evicted = arc_evict_state_impl(ml, sublist_idx,
markers[sublist_idx], spa, bytes_remaining);
scan_evicted += bytes_evicted;
total_evicted += bytes_evicted;
/* we've reached the end, wrap to the beginning */
if (++sublist_idx >= num_sublists)
sublist_idx = 0;
}
/*
* If we didn't evict anything during this scan, we have
* no reason to believe we'll evict more during another
* scan, so break the loop.
*/
if (scan_evicted == 0) {
/* This isn't possible, let's make that obvious */
ASSERT3S(bytes, !=, 0);
/*
* When bytes is ARC_EVICT_ALL, the only way to
* break the loop is when scan_evicted is zero.
* In that case, we actually have evicted enough,
* so we don't want to increment the kstat.
*/
if (bytes != ARC_EVICT_ALL) {
ASSERT3S(total_evicted, <, bytes);
ARCSTAT_BUMP(arcstat_evict_not_enough);
}
break;
}
}
for (int i = 0; i < num_sublists; i++) {
multilist_sublist_t *mls = multilist_sublist_lock(ml, i);
multilist_sublist_remove(mls, markers[i]);
multilist_sublist_unlock(mls);
kmem_cache_free(hdr_full_cache, markers[i]);
}
kmem_free(markers, sizeof (*markers) * num_sublists);
return (total_evicted);
}
/*
* Flush all "evictable" data of the given type from the arc state
* specified. This will not evict any "active" buffers (i.e. referenced).
*
* When 'retry' is set to B_FALSE, the function will make a single pass
* over the state and evict any buffers that it can. Since it doesn't
* continually retry the eviction, it might end up leaving some buffers
* in the ARC due to lock misses.
*
* When 'retry' is set to B_TRUE, the function will continually retry the
* eviction until *all* evictable buffers have been removed from the
* state. As a result, if concurrent insertions into the state are
* allowed (e.g. if the ARC isn't shutting down), this function might
* wind up in an infinite loop, continually trying to evict buffers.
*/
static uint64_t
arc_flush_state(arc_state_t *state, uint64_t spa, arc_buf_contents_t type,
boolean_t retry)
{
uint64_t evicted = 0;
while (zfs_refcount_count(&state->arcs_esize[type]) != 0) {
evicted += arc_evict_state(state, spa, ARC_EVICT_ALL, type);
if (!retry)
break;
}
return (evicted);
}
/*
* Evict the specified number of bytes from the state specified,
* restricting eviction to the spa and type given. This function
* prevents us from trying to evict more from a state's list than
* is "evictable", and to skip evicting altogether when passed a
* negative value for "bytes". In contrast, arc_evict_state() will
* evict everything it can, when passed a negative value for "bytes".
*/
static uint64_t
arc_evict_impl(arc_state_t *state, uint64_t spa, int64_t bytes,
arc_buf_contents_t type)
{
int64_t delta;
if (bytes > 0 && zfs_refcount_count(&state->arcs_esize[type]) > 0) {
delta = MIN(zfs_refcount_count(&state->arcs_esize[type]),
bytes);
return (arc_evict_state(state, spa, delta, type));
}
return (0);
}
/*
* The goal of this function is to evict enough meta data buffers from the
* ARC in order to enforce the arc_meta_limit. Achieving this is slightly
* more complicated than it appears because it is common for data buffers
* to have holds on meta data buffers. In addition, dnode meta data buffers
* will be held by the dnodes in the block preventing them from being freed.
* This means we can't simply traverse the ARC and expect to always find
* enough unheld meta data buffer to release.
*
* Therefore, this function has been updated to make alternating passes
* over the ARC releasing data buffers and then newly unheld meta data
* buffers. This ensures forward progress is maintained and meta_used
* will decrease. Normally this is sufficient, but if required the ARC
* will call the registered prune callbacks causing dentry and inodes to
* be dropped from the VFS cache. This will make dnode meta data buffers
* available for reclaim.
*/
static uint64_t
arc_evict_meta_balanced(uint64_t meta_used)
{
int64_t delta, prune = 0, adjustmnt;
uint64_t total_evicted = 0;
arc_buf_contents_t type = ARC_BUFC_DATA;
int restarts = MAX(zfs_arc_meta_adjust_restarts, 0);
restart:
/*
* This slightly differs than the way we evict from the mru in
* arc_evict because we don't have a "target" value (i.e. no
* "meta" arc_p). As a result, I think we can completely
* cannibalize the metadata in the MRU before we evict the
* metadata from the MFU. I think we probably need to implement a
* "metadata arc_p" value to do this properly.
*/
adjustmnt = meta_used - arc_meta_limit;
if (adjustmnt > 0 &&
zfs_refcount_count(&arc_mru->arcs_esize[type]) > 0) {
delta = MIN(zfs_refcount_count(&arc_mru->arcs_esize[type]),
adjustmnt);
total_evicted += arc_evict_impl(arc_mru, 0, delta, type);
adjustmnt -= delta;
}
/*
* We can't afford to recalculate adjustmnt here. If we do,
* new metadata buffers can sneak into the MRU or ANON lists,
* thus penalize the MFU metadata. Although the fudge factor is
* small, it has been empirically shown to be significant for
* certain workloads (e.g. creating many empty directories). As
* such, we use the original calculation for adjustmnt, and
* simply decrement the amount of data evicted from the MRU.
*/
if (adjustmnt > 0 &&
zfs_refcount_count(&arc_mfu->arcs_esize[type]) > 0) {
delta = MIN(zfs_refcount_count(&arc_mfu->arcs_esize[type]),
adjustmnt);
total_evicted += arc_evict_impl(arc_mfu, 0, delta, type);
}
adjustmnt = meta_used - arc_meta_limit;
if (adjustmnt > 0 &&
zfs_refcount_count(&arc_mru_ghost->arcs_esize[type]) > 0) {
delta = MIN(adjustmnt,
zfs_refcount_count(&arc_mru_ghost->arcs_esize[type]));
total_evicted += arc_evict_impl(arc_mru_ghost, 0, delta, type);
adjustmnt -= delta;
}
if (adjustmnt > 0 &&
zfs_refcount_count(&arc_mfu_ghost->arcs_esize[type]) > 0) {
delta = MIN(adjustmnt,
zfs_refcount_count(&arc_mfu_ghost->arcs_esize[type]));
total_evicted += arc_evict_impl(arc_mfu_ghost, 0, delta, type);
}
/*
* If after attempting to make the requested adjustment to the ARC
* the meta limit is still being exceeded then request that the
* higher layers drop some cached objects which have holds on ARC
* meta buffers. Requests to the upper layers will be made with
* increasingly large scan sizes until the ARC is below the limit.
*/
if (meta_used > arc_meta_limit) {
if (type == ARC_BUFC_DATA) {
type = ARC_BUFC_METADATA;
} else {
type = ARC_BUFC_DATA;
if (zfs_arc_meta_prune) {
prune += zfs_arc_meta_prune;
arc_prune_async(prune);
}
}
if (restarts > 0) {
restarts--;
goto restart;
}
}
return (total_evicted);
}
/*
* Evict metadata buffers from the cache, such that arcstat_meta_used is
* capped by the arc_meta_limit tunable.
*/
static uint64_t
arc_evict_meta_only(uint64_t meta_used)
{
uint64_t total_evicted = 0;
int64_t target;
/*
* If we're over the meta limit, we want to evict enough
* metadata to get back under the meta limit. We don't want to
* evict so much that we drop the MRU below arc_p, though. If
* we're over the meta limit more than we're over arc_p, we
* evict some from the MRU here, and some from the MFU below.
*/
target = MIN((int64_t)(meta_used - arc_meta_limit),
(int64_t)(zfs_refcount_count(&arc_anon->arcs_size) +
zfs_refcount_count(&arc_mru->arcs_size) - arc_p));
total_evicted += arc_evict_impl(arc_mru, 0, target, ARC_BUFC_METADATA);
/*
* Similar to the above, we want to evict enough bytes to get us
* below the meta limit, but not so much as to drop us below the
* space allotted to the MFU (which is defined as arc_c - arc_p).
*/
target = MIN((int64_t)(meta_used - arc_meta_limit),
(int64_t)(zfs_refcount_count(&arc_mfu->arcs_size) -
(arc_c - arc_p)));
total_evicted += arc_evict_impl(arc_mfu, 0, target, ARC_BUFC_METADATA);
return (total_evicted);
}
static uint64_t
arc_evict_meta(uint64_t meta_used)
{
if (zfs_arc_meta_strategy == ARC_STRATEGY_META_ONLY)
return (arc_evict_meta_only(meta_used));
else
return (arc_evict_meta_balanced(meta_used));
}
/*
* Return the type of the oldest buffer in the given arc state
*
* This function will select a random sublist of type ARC_BUFC_DATA and
* a random sublist of type ARC_BUFC_METADATA. The tail of each sublist
* is compared, and the type which contains the "older" buffer will be
* returned.
*/
static arc_buf_contents_t
arc_evict_type(arc_state_t *state)
{
multilist_t *data_ml = &state->arcs_list[ARC_BUFC_DATA];
multilist_t *meta_ml = &state->arcs_list[ARC_BUFC_METADATA];
int data_idx = multilist_get_random_index(data_ml);
int meta_idx = multilist_get_random_index(meta_ml);
multilist_sublist_t *data_mls;
multilist_sublist_t *meta_mls;
arc_buf_contents_t type;
arc_buf_hdr_t *data_hdr;
arc_buf_hdr_t *meta_hdr;
/*
* We keep the sublist lock until we're finished, to prevent
* the headers from being destroyed via arc_evict_state().
*/
data_mls = multilist_sublist_lock(data_ml, data_idx);
meta_mls = multilist_sublist_lock(meta_ml, meta_idx);
/*
* These two loops are to ensure we skip any markers that
* might be at the tail of the lists due to arc_evict_state().
*/
for (data_hdr = multilist_sublist_tail(data_mls); data_hdr != NULL;
data_hdr = multilist_sublist_prev(data_mls, data_hdr)) {
if (data_hdr->b_spa != 0)
break;
}
for (meta_hdr = multilist_sublist_tail(meta_mls); meta_hdr != NULL;
meta_hdr = multilist_sublist_prev(meta_mls, meta_hdr)) {
if (meta_hdr->b_spa != 0)
break;
}
if (data_hdr == NULL && meta_hdr == NULL) {
type = ARC_BUFC_DATA;
} else if (data_hdr == NULL) {
ASSERT3P(meta_hdr, !=, NULL);
type = ARC_BUFC_METADATA;
} else if (meta_hdr == NULL) {
ASSERT3P(data_hdr, !=, NULL);
type = ARC_BUFC_DATA;
} else {
ASSERT3P(data_hdr, !=, NULL);
ASSERT3P(meta_hdr, !=, NULL);
/* The headers can't be on the sublist without an L1 header */
ASSERT(HDR_HAS_L1HDR(data_hdr));
ASSERT(HDR_HAS_L1HDR(meta_hdr));
if (data_hdr->b_l1hdr.b_arc_access <
meta_hdr->b_l1hdr.b_arc_access) {
type = ARC_BUFC_DATA;
} else {
type = ARC_BUFC_METADATA;
}
}
multilist_sublist_unlock(meta_mls);
multilist_sublist_unlock(data_mls);
return (type);
}
/*
* Evict buffers from the cache, such that arcstat_size is capped by arc_c.
*/
static uint64_t
arc_evict(void)
{
uint64_t total_evicted = 0;
uint64_t bytes;
int64_t target;
uint64_t asize = aggsum_value(&arc_sums.arcstat_size);
uint64_t ameta = aggsum_value(&arc_sums.arcstat_meta_used);
/*
* If we're over arc_meta_limit, we want to correct that before
* potentially evicting data buffers below.
*/
total_evicted += arc_evict_meta(ameta);
/*
* Adjust MRU size
*
* If we're over the target cache size, we want to evict enough
* from the list to get back to our target size. We don't want
* to evict too much from the MRU, such that it drops below
* arc_p. So, if we're over our target cache size more than
* the MRU is over arc_p, we'll evict enough to get back to
* arc_p here, and then evict more from the MFU below.
*/
target = MIN((int64_t)(asize - arc_c),
(int64_t)(zfs_refcount_count(&arc_anon->arcs_size) +
zfs_refcount_count(&arc_mru->arcs_size) + ameta - arc_p));
/*
* If we're below arc_meta_min, always prefer to evict data.
* Otherwise, try to satisfy the requested number of bytes to
* evict from the type which contains older buffers; in an
* effort to keep newer buffers in the cache regardless of their
* type. If we cannot satisfy the number of bytes from this
* type, spill over into the next type.
*/
if (arc_evict_type(arc_mru) == ARC_BUFC_METADATA &&
ameta > arc_meta_min) {
bytes = arc_evict_impl(arc_mru, 0, target, ARC_BUFC_METADATA);
total_evicted += bytes;
/*
* If we couldn't evict our target number of bytes from
* metadata, we try to get the rest from data.
*/
target -= bytes;
total_evicted +=
arc_evict_impl(arc_mru, 0, target, ARC_BUFC_DATA);
} else {
bytes = arc_evict_impl(arc_mru, 0, target, ARC_BUFC_DATA);
total_evicted += bytes;
/*
* If we couldn't evict our target number of bytes from
* data, we try to get the rest from metadata.
*/
target -= bytes;
total_evicted +=
arc_evict_impl(arc_mru, 0, target, ARC_BUFC_METADATA);
}
/*
* Re-sum ARC stats after the first round of evictions.
*/
asize = aggsum_value(&arc_sums.arcstat_size);
ameta = aggsum_value(&arc_sums.arcstat_meta_used);
/*
* Adjust MFU size
*
* Now that we've tried to evict enough from the MRU to get its
* size back to arc_p, if we're still above the target cache
* size, we evict the rest from the MFU.
*/
target = asize - arc_c;
if (arc_evict_type(arc_mfu) == ARC_BUFC_METADATA &&
ameta > arc_meta_min) {
bytes = arc_evict_impl(arc_mfu, 0, target, ARC_BUFC_METADATA);
total_evicted += bytes;
/*
* If we couldn't evict our target number of bytes from
* metadata, we try to get the rest from data.
*/
target -= bytes;
total_evicted +=
arc_evict_impl(arc_mfu, 0, target, ARC_BUFC_DATA);
} else {
bytes = arc_evict_impl(arc_mfu, 0, target, ARC_BUFC_DATA);
total_evicted += bytes;
/*
* If we couldn't evict our target number of bytes from
* data, we try to get the rest from data.
*/
target -= bytes;
total_evicted +=
arc_evict_impl(arc_mfu, 0, target, ARC_BUFC_METADATA);
}
/*
* Adjust ghost lists
*
* In addition to the above, the ARC also defines target values
* for the ghost lists. The sum of the mru list and mru ghost
* list should never exceed the target size of the cache, and
* the sum of the mru list, mfu list, mru ghost list, and mfu
* ghost list should never exceed twice the target size of the
* cache. The following logic enforces these limits on the ghost
* caches, and evicts from them as needed.
*/
target = zfs_refcount_count(&arc_mru->arcs_size) +
zfs_refcount_count(&arc_mru_ghost->arcs_size) - arc_c;
bytes = arc_evict_impl(arc_mru_ghost, 0, target, ARC_BUFC_DATA);
total_evicted += bytes;
target -= bytes;
total_evicted +=
arc_evict_impl(arc_mru_ghost, 0, target, ARC_BUFC_METADATA);
/*
* We assume the sum of the mru list and mfu list is less than
* or equal to arc_c (we enforced this above), which means we
* can use the simpler of the two equations below:
*
* mru + mfu + mru ghost + mfu ghost <= 2 * arc_c
* mru ghost + mfu ghost <= arc_c
*/
target = zfs_refcount_count(&arc_mru_ghost->arcs_size) +
zfs_refcount_count(&arc_mfu_ghost->arcs_size) - arc_c;
bytes = arc_evict_impl(arc_mfu_ghost, 0, target, ARC_BUFC_DATA);
total_evicted += bytes;
target -= bytes;
total_evicted +=
arc_evict_impl(arc_mfu_ghost, 0, target, ARC_BUFC_METADATA);
return (total_evicted);
}
void
arc_flush(spa_t *spa, boolean_t retry)
{
uint64_t guid = 0;
/*
* If retry is B_TRUE, a spa must not be specified since we have
* no good way to determine if all of a spa's buffers have been
* evicted from an arc state.
*/
ASSERT(!retry || spa == 0);
if (spa != NULL)
guid = spa_load_guid(spa);
(void) arc_flush_state(arc_mru, guid, ARC_BUFC_DATA, retry);
(void) arc_flush_state(arc_mru, guid, ARC_BUFC_METADATA, retry);
(void) arc_flush_state(arc_mfu, guid, ARC_BUFC_DATA, retry);
(void) arc_flush_state(arc_mfu, guid, ARC_BUFC_METADATA, retry);
(void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_DATA, retry);
(void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_METADATA, retry);
(void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_DATA, retry);
(void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_METADATA, retry);
}
void
arc_reduce_target_size(int64_t to_free)
{
uint64_t asize = aggsum_value(&arc_sums.arcstat_size);
/*
* All callers want the ARC to actually evict (at least) this much
* memory. Therefore we reduce from the lower of the current size and
* the target size. This way, even if arc_c is much higher than
* arc_size (as can be the case after many calls to arc_freed(), we will
* immediately have arc_c < arc_size and therefore the arc_evict_zthr
* will evict.
*/
uint64_t c = MIN(arc_c, asize);
if (c > to_free && c - to_free > arc_c_min) {
arc_c = c - to_free;
atomic_add_64(&arc_p, -(arc_p >> arc_shrink_shift));
if (arc_p > arc_c)
arc_p = (arc_c >> 1);
ASSERT(arc_c >= arc_c_min);
ASSERT((int64_t)arc_p >= 0);
} else {
arc_c = arc_c_min;
}
if (asize > arc_c) {
/* See comment in arc_evict_cb_check() on why lock+flag */
mutex_enter(&arc_evict_lock);
arc_evict_needed = B_TRUE;
mutex_exit(&arc_evict_lock);
zthr_wakeup(arc_evict_zthr);
}
}
/*
* Determine if the system is under memory pressure and is asking
* to reclaim memory. A return value of B_TRUE indicates that the system
* is under memory pressure and that the arc should adjust accordingly.
*/
boolean_t
arc_reclaim_needed(void)
{
return (arc_available_memory() < 0);
}
void
arc_kmem_reap_soon(void)
{
size_t i;
kmem_cache_t *prev_cache = NULL;
kmem_cache_t *prev_data_cache = NULL;
extern kmem_cache_t *zio_buf_cache[];
extern kmem_cache_t *zio_data_buf_cache[];
#ifdef _KERNEL
if ((aggsum_compare(&arc_sums.arcstat_meta_used,
arc_meta_limit) >= 0) && zfs_arc_meta_prune) {
/*
* We are exceeding our meta-data cache limit.
* Prune some entries to release holds on meta-data.
*/
arc_prune_async(zfs_arc_meta_prune);
}
#if defined(_ILP32)
/*
* Reclaim unused memory from all kmem caches.
*/
kmem_reap();
#endif
#endif
for (i = 0; i < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; i++) {
#if defined(_ILP32)
/* reach upper limit of cache size on 32-bit */
if (zio_buf_cache[i] == NULL)
break;
#endif
if (zio_buf_cache[i] != prev_cache) {
prev_cache = zio_buf_cache[i];
kmem_cache_reap_now(zio_buf_cache[i]);
}
if (zio_data_buf_cache[i] != prev_data_cache) {
prev_data_cache = zio_data_buf_cache[i];
kmem_cache_reap_now(zio_data_buf_cache[i]);
}
}
kmem_cache_reap_now(buf_cache);
kmem_cache_reap_now(hdr_full_cache);
kmem_cache_reap_now(hdr_l2only_cache);
kmem_cache_reap_now(zfs_btree_leaf_cache);
abd_cache_reap_now();
}
/* ARGSUSED */
static boolean_t
arc_evict_cb_check(void *arg, zthr_t *zthr)
{
#ifdef ZFS_DEBUG
/*
* This is necessary in order to keep the kstat information
* up to date for tools that display kstat data such as the
* mdb ::arc dcmd and the Linux crash utility. These tools
* typically do not call kstat's update function, but simply
* dump out stats from the most recent update. Without
* this call, these commands may show stale stats for the
* anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even
* with this call, the data might be out of date if the
* evict thread hasn't been woken recently; but that should
* suffice. The arc_state_t structures can be queried
* directly if more accurate information is needed.
*/
if (arc_ksp != NULL)
arc_ksp->ks_update(arc_ksp, KSTAT_READ);
#endif
/*
* We have to rely on arc_wait_for_eviction() to tell us when to
* evict, rather than checking if we are overflowing here, so that we
* are sure to not leave arc_wait_for_eviction() waiting on aew_cv.
* If we have become "not overflowing" since arc_wait_for_eviction()
* checked, we need to wake it up. We could broadcast the CV here,
* but arc_wait_for_eviction() may have not yet gone to sleep. We
* would need to use a mutex to ensure that this function doesn't
* broadcast until arc_wait_for_eviction() has gone to sleep (e.g.
* the arc_evict_lock). However, the lock ordering of such a lock
* would necessarily be incorrect with respect to the zthr_lock,
* which is held before this function is called, and is held by
* arc_wait_for_eviction() when it calls zthr_wakeup().
*/
return (arc_evict_needed);
}
/*
* Keep arc_size under arc_c by running arc_evict which evicts data
* from the ARC.
*/
/* ARGSUSED */
static void
arc_evict_cb(void *arg, zthr_t *zthr)
{
uint64_t evicted = 0;
fstrans_cookie_t cookie = spl_fstrans_mark();
/* Evict from cache */
evicted = arc_evict();
/*
* If evicted is zero, we couldn't evict anything
* via arc_evict(). This could be due to hash lock
* collisions, but more likely due to the majority of
* arc buffers being unevictable. Therefore, even if
* arc_size is above arc_c, another pass is unlikely to
* be helpful and could potentially cause us to enter an
* infinite loop. Additionally, zthr_iscancelled() is
* checked here so that if the arc is shutting down, the
* broadcast will wake any remaining arc evict waiters.
*/
mutex_enter(&arc_evict_lock);
arc_evict_needed = !zthr_iscancelled(arc_evict_zthr) &&
evicted > 0 && aggsum_compare(&arc_sums.arcstat_size, arc_c) > 0;
if (!arc_evict_needed) {
/*
* We're either no longer overflowing, or we
* can't evict anything more, so we should wake
* arc_get_data_impl() sooner.
*/
arc_evict_waiter_t *aw;
while ((aw = list_remove_head(&arc_evict_waiters)) != NULL) {
cv_broadcast(&aw->aew_cv);
}
arc_set_need_free();
}
mutex_exit(&arc_evict_lock);
spl_fstrans_unmark(cookie);
}
/* ARGSUSED */
static boolean_t
arc_reap_cb_check(void *arg, zthr_t *zthr)
{
int64_t free_memory = arc_available_memory();
static int reap_cb_check_counter = 0;
/*
* If a kmem reap is already active, don't schedule more. We must
* check for this because kmem_cache_reap_soon() won't actually
* block on the cache being reaped (this is to prevent callers from
* becoming implicitly blocked by a system-wide kmem reap -- which,
* on a system with many, many full magazines, can take minutes).
*/
if (!kmem_cache_reap_active() && free_memory < 0) {
arc_no_grow = B_TRUE;
arc_warm = B_TRUE;
/*
* Wait at least zfs_grow_retry (default 5) seconds
* before considering growing.
*/
arc_growtime = gethrtime() + SEC2NSEC(arc_grow_retry);
return (B_TRUE);
} else if (free_memory < arc_c >> arc_no_grow_shift) {
arc_no_grow = B_TRUE;
} else if (gethrtime() >= arc_growtime) {
arc_no_grow = B_FALSE;
}
/*
* Called unconditionally every 60 seconds to reclaim unused
* zstd compression and decompression context. This is done
* here to avoid the need for an independent thread.
*/
if (!((reap_cb_check_counter++) % 60))
zfs_zstd_cache_reap_now();
return (B_FALSE);
}
/*
* Keep enough free memory in the system by reaping the ARC's kmem
* caches. To cause more slabs to be reapable, we may reduce the
* target size of the cache (arc_c), causing the arc_evict_cb()
* to free more buffers.
*/
/* ARGSUSED */
static void
arc_reap_cb(void *arg, zthr_t *zthr)
{
int64_t free_memory;
fstrans_cookie_t cookie = spl_fstrans_mark();
/*
* Kick off asynchronous kmem_reap()'s of all our caches.
*/
arc_kmem_reap_soon();
/*
* Wait at least arc_kmem_cache_reap_retry_ms between
* arc_kmem_reap_soon() calls. Without this check it is possible to
* end up in a situation where we spend lots of time reaping
* caches, while we're near arc_c_min. Waiting here also gives the
* subsequent free memory check a chance of finding that the
* asynchronous reap has already freed enough memory, and we don't
* need to call arc_reduce_target_size().
*/
delay((hz * arc_kmem_cache_reap_retry_ms + 999) / 1000);
/*
* Reduce the target size as needed to maintain the amount of free
* memory in the system at a fraction of the arc_size (1/128th by
* default). If oversubscribed (free_memory < 0) then reduce the
* target arc_size by the deficit amount plus the fractional
* amount. If free memory is positive but less than the fractional
* amount, reduce by what is needed to hit the fractional amount.
*/
free_memory = arc_available_memory();
int64_t to_free =
(arc_c >> arc_shrink_shift) - free_memory;
if (to_free > 0) {
arc_reduce_target_size(to_free);
}
spl_fstrans_unmark(cookie);
}
#ifdef _KERNEL
/*
* Determine the amount of memory eligible for eviction contained in the
* ARC. All clean data reported by the ghost lists can always be safely
* evicted. Due to arc_c_min, the same does not hold for all clean data
* contained by the regular mru and mfu lists.
*
* In the case of the regular mru and mfu lists, we need to report as
* much clean data as possible, such that evicting that same reported
* data will not bring arc_size below arc_c_min. Thus, in certain
* circumstances, the total amount of clean data in the mru and mfu
* lists might not actually be evictable.
*
* The following two distinct cases are accounted for:
*
* 1. The sum of the amount of dirty data contained by both the mru and
* mfu lists, plus the ARC's other accounting (e.g. the anon list),
* is greater than or equal to arc_c_min.
* (i.e. amount of dirty data >= arc_c_min)
*
* This is the easy case; all clean data contained by the mru and mfu
* lists is evictable. Evicting all clean data can only drop arc_size
* to the amount of dirty data, which is greater than arc_c_min.
*
* 2. The sum of the amount of dirty data contained by both the mru and
* mfu lists, plus the ARC's other accounting (e.g. the anon list),
* is less than arc_c_min.
* (i.e. arc_c_min > amount of dirty data)
*
* 2.1. arc_size is greater than or equal arc_c_min.
* (i.e. arc_size >= arc_c_min > amount of dirty data)
*
* In this case, not all clean data from the regular mru and mfu
* lists is actually evictable; we must leave enough clean data
* to keep arc_size above arc_c_min. Thus, the maximum amount of
* evictable data from the two lists combined, is exactly the
* difference between arc_size and arc_c_min.
*
* 2.2. arc_size is less than arc_c_min
* (i.e. arc_c_min > arc_size > amount of dirty data)
*
* In this case, none of the data contained in the mru and mfu
* lists is evictable, even if it's clean. Since arc_size is
* already below arc_c_min, evicting any more would only
* increase this negative difference.
*/
#endif /* _KERNEL */
/*
* Adapt arc info given the number of bytes we are trying to add and
* the state that we are coming from. This function is only called
* when we are adding new content to the cache.
*/
static void
arc_adapt(int bytes, arc_state_t *state)
{
int mult;
uint64_t arc_p_min = (arc_c >> arc_p_min_shift);
int64_t mrug_size = zfs_refcount_count(&arc_mru_ghost->arcs_size);
int64_t mfug_size = zfs_refcount_count(&arc_mfu_ghost->arcs_size);
ASSERT(bytes > 0);
/*
* Adapt the target size of the MRU list:
* - if we just hit in the MRU ghost list, then increase
* the target size of the MRU list.
* - if we just hit in the MFU ghost list, then increase
* the target size of the MFU list by decreasing the
* target size of the MRU list.
*/
if (state == arc_mru_ghost) {
mult = (mrug_size >= mfug_size) ? 1 : (mfug_size / mrug_size);
if (!zfs_arc_p_dampener_disable)
mult = MIN(mult, 10); /* avoid wild arc_p adjustment */
arc_p = MIN(arc_c - arc_p_min, arc_p + bytes * mult);
} else if (state == arc_mfu_ghost) {
uint64_t delta;
mult = (mfug_size >= mrug_size) ? 1 : (mrug_size / mfug_size);
if (!zfs_arc_p_dampener_disable)
mult = MIN(mult, 10);
delta = MIN(bytes * mult, arc_p);
arc_p = MAX(arc_p_min, arc_p - delta);
}
ASSERT((int64_t)arc_p >= 0);
/*
* Wake reap thread if we do not have any available memory
*/
if (arc_reclaim_needed()) {
zthr_wakeup(arc_reap_zthr);
return;
}
if (arc_no_grow)
return;
if (arc_c >= arc_c_max)
return;
/*
* If we're within (2 * maxblocksize) bytes of the target
* cache size, increment the target cache size
*/
ASSERT3U(arc_c, >=, 2ULL << SPA_MAXBLOCKSHIFT);
if (aggsum_upper_bound(&arc_sums.arcstat_size) >=
arc_c - (2ULL << SPA_MAXBLOCKSHIFT)) {
atomic_add_64(&arc_c, (int64_t)bytes);
if (arc_c > arc_c_max)
arc_c = arc_c_max;
else if (state == arc_anon)
atomic_add_64(&arc_p, (int64_t)bytes);
if (arc_p > arc_c)
arc_p = arc_c;
}
ASSERT((int64_t)arc_p >= 0);
}
/*
* Check if arc_size has grown past our upper threshold, determined by
* zfs_arc_overflow_shift.
*/
boolean_t
arc_is_overflowing(void)
{
/* Always allow at least one block of overflow */
int64_t overflow = MAX(SPA_MAXBLOCKSIZE,
arc_c >> zfs_arc_overflow_shift);
/*
* We just compare the lower bound here for performance reasons. Our
* primary goals are to make sure that the arc never grows without
* bound, and that it can reach its maximum size. This check
* accomplishes both goals. The maximum amount we could run over by is
* 2 * aggsum_borrow_multiplier * NUM_CPUS * the average size of a block
* in the ARC. In practice, that's in the tens of MB, which is low
* enough to be safe.
*/
return (aggsum_lower_bound(&arc_sums.arcstat_size) >=
(int64_t)arc_c + overflow);
}
static abd_t *
arc_get_data_abd(arc_buf_hdr_t *hdr, uint64_t size, void *tag,
boolean_t do_adapt)
{
arc_buf_contents_t type = arc_buf_type(hdr);
arc_get_data_impl(hdr, size, tag, do_adapt);
if (type == ARC_BUFC_METADATA) {
return (abd_alloc(size, B_TRUE));
} else {
ASSERT(type == ARC_BUFC_DATA);
return (abd_alloc(size, B_FALSE));
}
}
static void *
arc_get_data_buf(arc_buf_hdr_t *hdr, uint64_t size, void *tag)
{
arc_buf_contents_t type = arc_buf_type(hdr);
arc_get_data_impl(hdr, size, tag, B_TRUE);
if (type == ARC_BUFC_METADATA) {
return (zio_buf_alloc(size));
} else {
ASSERT(type == ARC_BUFC_DATA);
return (zio_data_buf_alloc(size));
}
}
/*
* Wait for the specified amount of data (in bytes) to be evicted from the
* ARC, and for there to be sufficient free memory in the system. Waiting for
* eviction ensures that the memory used by the ARC decreases. Waiting for
* free memory ensures that the system won't run out of free pages, regardless
* of ARC behavior and settings. See arc_lowmem_init().
*/
void
arc_wait_for_eviction(uint64_t amount)
{
mutex_enter(&arc_evict_lock);
if (arc_is_overflowing()) {
arc_evict_needed = B_TRUE;
zthr_wakeup(arc_evict_zthr);
if (amount != 0) {
arc_evict_waiter_t aw;
list_link_init(&aw.aew_node);
cv_init(&aw.aew_cv, NULL, CV_DEFAULT, NULL);
uint64_t last_count = 0;
if (!list_is_empty(&arc_evict_waiters)) {
arc_evict_waiter_t *last =
list_tail(&arc_evict_waiters);
last_count = last->aew_count;
}
/*
* Note, the last waiter's count may be less than
* arc_evict_count if we are low on memory in which
* case arc_evict_state_impl() may have deferred
* wakeups (but still incremented arc_evict_count).
*/
aw.aew_count =
MAX(last_count, arc_evict_count) + amount;
list_insert_tail(&arc_evict_waiters, &aw);
arc_set_need_free();
DTRACE_PROBE3(arc__wait__for__eviction,
uint64_t, amount,
uint64_t, arc_evict_count,
uint64_t, aw.aew_count);
/*
* We will be woken up either when arc_evict_count
* reaches aew_count, or when the ARC is no longer
* overflowing and eviction completes.
*/
cv_wait(&aw.aew_cv, &arc_evict_lock);
/*
* In case of "false" wakeup, we will still be on the
* list.
*/
if (list_link_active(&aw.aew_node))
list_remove(&arc_evict_waiters, &aw);
cv_destroy(&aw.aew_cv);
}
}
mutex_exit(&arc_evict_lock);
}
/*
* Allocate a block and return it to the caller. If we are hitting the
* hard limit for the cache size, we must sleep, waiting for the eviction
* thread to catch up. If we're past the target size but below the hard
* limit, we'll only signal the reclaim thread and continue on.
*/
static void
arc_get_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag,
boolean_t do_adapt)
{
arc_state_t *state = hdr->b_l1hdr.b_state;
arc_buf_contents_t type = arc_buf_type(hdr);
if (do_adapt)
arc_adapt(size, state);
/*
* If arc_size is currently overflowing, we must be adding data
* faster than we are evicting. To ensure we don't compound the
* problem by adding more data and forcing arc_size to grow even
* further past it's target size, we wait for the eviction thread to
* make some progress. We also wait for there to be sufficient free
* memory in the system, as measured by arc_free_memory().
*
* Specifically, we wait for zfs_arc_eviction_pct percent of the
* requested size to be evicted. This should be more than 100%, to
* ensure that that progress is also made towards getting arc_size
* under arc_c. See the comment above zfs_arc_eviction_pct.
*
* We do the overflowing check without holding the arc_evict_lock to
* reduce lock contention in this hot path. Note that
* arc_wait_for_eviction() will acquire the lock and check again to
* ensure we are truly overflowing before blocking.
*/
if (arc_is_overflowing()) {
arc_wait_for_eviction(size *
zfs_arc_eviction_pct / 100);
}
VERIFY3U(hdr->b_type, ==, type);
if (type == ARC_BUFC_METADATA) {
arc_space_consume(size, ARC_SPACE_META);
} else {
arc_space_consume(size, ARC_SPACE_DATA);
}
/*
* Update the state size. Note that ghost states have a
* "ghost size" and so don't need to be updated.
*/
if (!GHOST_STATE(state)) {
(void) zfs_refcount_add_many(&state->arcs_size, size, tag);
/*
* If this is reached via arc_read, the link is
* protected by the hash lock. If reached via
* arc_buf_alloc, the header should not be accessed by
* any other thread. And, if reached via arc_read_done,
* the hash lock will protect it if it's found in the
* hash table; otherwise no other thread should be
* trying to [add|remove]_reference it.
*/
if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) {
ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
(void) zfs_refcount_add_many(&state->arcs_esize[type],
size, tag);
}
/*
* If we are growing the cache, and we are adding anonymous
* data, and we have outgrown arc_p, update arc_p
*/
if (aggsum_upper_bound(&arc_sums.arcstat_size) < arc_c &&
hdr->b_l1hdr.b_state == arc_anon &&
(zfs_refcount_count(&arc_anon->arcs_size) +
zfs_refcount_count(&arc_mru->arcs_size) > arc_p))
arc_p = MIN(arc_c, arc_p + size);
}
}
static void
arc_free_data_abd(arc_buf_hdr_t *hdr, abd_t *abd, uint64_t size, void *tag)
{
arc_free_data_impl(hdr, size, tag);
abd_free(abd);
}
static void
arc_free_data_buf(arc_buf_hdr_t *hdr, void *buf, uint64_t size, void *tag)
{
arc_buf_contents_t type = arc_buf_type(hdr);
arc_free_data_impl(hdr, size, tag);
if (type == ARC_BUFC_METADATA) {
zio_buf_free(buf, size);
} else {
ASSERT(type == ARC_BUFC_DATA);
zio_data_buf_free(buf, size);
}
}
/*
* Free the arc data buffer.
*/
static void
arc_free_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag)
{
arc_state_t *state = hdr->b_l1hdr.b_state;
arc_buf_contents_t type = arc_buf_type(hdr);
/* protected by hash lock, if in the hash table */
if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) {
ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
ASSERT(state != arc_anon && state != arc_l2c_only);
(void) zfs_refcount_remove_many(&state->arcs_esize[type],
size, tag);
}
(void) zfs_refcount_remove_many(&state->arcs_size, size, tag);
VERIFY3U(hdr->b_type, ==, type);
if (type == ARC_BUFC_METADATA) {
arc_space_return(size, ARC_SPACE_META);
} else {
ASSERT(type == ARC_BUFC_DATA);
arc_space_return(size, ARC_SPACE_DATA);
}
}
/*
* This routine is called whenever a buffer is accessed.
* NOTE: the hash lock is dropped in this function.
*/
static void
arc_access(arc_buf_hdr_t *hdr, kmutex_t *hash_lock)
{
clock_t now;
ASSERT(MUTEX_HELD(hash_lock));
ASSERT(HDR_HAS_L1HDR(hdr));
if (hdr->b_l1hdr.b_state == arc_anon) {
/*
* This buffer is not in the cache, and does not
* appear in our "ghost" list. Add the new buffer
* to the MRU state.
*/
ASSERT0(hdr->b_l1hdr.b_arc_access);
hdr->b_l1hdr.b_arc_access = ddi_get_lbolt();
DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr);
arc_change_state(arc_mru, hdr, hash_lock);
} else if (hdr->b_l1hdr.b_state == arc_mru) {
now = ddi_get_lbolt();
/*
* If this buffer is here because of a prefetch, then either:
* - clear the flag if this is a "referencing" read
* (any subsequent access will bump this into the MFU state).
* or
* - move the buffer to the head of the list if this is
* another prefetch (to make it less likely to be evicted).
*/
if (HDR_PREFETCH(hdr) || HDR_PRESCIENT_PREFETCH(hdr)) {
if (zfs_refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) {
/* link protected by hash lock */
ASSERT(multilist_link_active(
&hdr->b_l1hdr.b_arc_node));
} else {
if (HDR_HAS_L2HDR(hdr))
l2arc_hdr_arcstats_decrement_state(hdr);
arc_hdr_clear_flags(hdr,
ARC_FLAG_PREFETCH |
ARC_FLAG_PRESCIENT_PREFETCH);
atomic_inc_32(&hdr->b_l1hdr.b_mru_hits);
ARCSTAT_BUMP(arcstat_mru_hits);
if (HDR_HAS_L2HDR(hdr))
l2arc_hdr_arcstats_increment_state(hdr);
}
hdr->b_l1hdr.b_arc_access = now;
return;
}
/*
* This buffer has been "accessed" only once so far,
* but it is still in the cache. Move it to the MFU
* state.
*/
if (ddi_time_after(now, hdr->b_l1hdr.b_arc_access +
ARC_MINTIME)) {
/*
* More than 125ms have passed since we
* instantiated this buffer. Move it to the
* most frequently used state.
*/
hdr->b_l1hdr.b_arc_access = now;
DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr);
arc_change_state(arc_mfu, hdr, hash_lock);
}
atomic_inc_32(&hdr->b_l1hdr.b_mru_hits);
ARCSTAT_BUMP(arcstat_mru_hits);
} else if (hdr->b_l1hdr.b_state == arc_mru_ghost) {
arc_state_t *new_state;
/*
* This buffer has been "accessed" recently, but
* was evicted from the cache. Move it to the
* MFU state.
*/
if (HDR_PREFETCH(hdr) || HDR_PRESCIENT_PREFETCH(hdr)) {
new_state = arc_mru;
if (zfs_refcount_count(&hdr->b_l1hdr.b_refcnt) > 0) {
if (HDR_HAS_L2HDR(hdr))
l2arc_hdr_arcstats_decrement_state(hdr);
arc_hdr_clear_flags(hdr,
ARC_FLAG_PREFETCH |
ARC_FLAG_PRESCIENT_PREFETCH);
if (HDR_HAS_L2HDR(hdr))
l2arc_hdr_arcstats_increment_state(hdr);
}
DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr);
} else {
new_state = arc_mfu;
DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr);
}
hdr->b_l1hdr.b_arc_access = ddi_get_lbolt();
arc_change_state(new_state, hdr, hash_lock);
atomic_inc_32(&hdr->b_l1hdr.b_mru_ghost_hits);
ARCSTAT_BUMP(arcstat_mru_ghost_hits);
} else if (hdr->b_l1hdr.b_state == arc_mfu) {
/*
* This buffer has been accessed more than once and is
* still in the cache. Keep it in the MFU state.
*
* NOTE: an add_reference() that occurred when we did
* the arc_read() will have kicked this off the list.
* If it was a prefetch, we will explicitly move it to
* the head of the list now.
*/
atomic_inc_32(&hdr->b_l1hdr.b_mfu_hits);
ARCSTAT_BUMP(arcstat_mfu_hits);
hdr->b_l1hdr.b_arc_access = ddi_get_lbolt();
} else if (hdr->b_l1hdr.b_state == arc_mfu_ghost) {
arc_state_t *new_state = arc_mfu;
/*
* This buffer has been accessed more than once but has
* been evicted from the cache. Move it back to the
* MFU state.
*/
if (HDR_PREFETCH(hdr) || HDR_PRESCIENT_PREFETCH(hdr)) {
/*
* This is a prefetch access...
* move this block back to the MRU state.
*/
new_state = arc_mru;
}
hdr->b_l1hdr.b_arc_access = ddi_get_lbolt();
DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr);
arc_change_state(new_state, hdr, hash_lock);
atomic_inc_32(&hdr->b_l1hdr.b_mfu_ghost_hits);
ARCSTAT_BUMP(arcstat_mfu_ghost_hits);
} else if (hdr->b_l1hdr.b_state == arc_l2c_only) {
/*
* This buffer is on the 2nd Level ARC.
*/
hdr->b_l1hdr.b_arc_access = ddi_get_lbolt();
DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr);
arc_change_state(arc_mfu, hdr, hash_lock);
} else {
cmn_err(CE_PANIC, "invalid arc state 0x%p",
hdr->b_l1hdr.b_state);
}
}
/*
* This routine is called by dbuf_hold() to update the arc_access() state
* which otherwise would be skipped for entries in the dbuf cache.
*/
void
arc_buf_access(arc_buf_t *buf)
{
mutex_enter(&buf->b_evict_lock);
arc_buf_hdr_t *hdr = buf->b_hdr;
/*
* Avoid taking the hash_lock when possible as an optimization.
* The header must be checked again under the hash_lock in order
* to handle the case where it is concurrently being released.
*/
if (hdr->b_l1hdr.b_state == arc_anon || HDR_EMPTY(hdr)) {
mutex_exit(&buf->b_evict_lock);
return;
}
kmutex_t *hash_lock = HDR_LOCK(hdr);
mutex_enter(hash_lock);
if (hdr->b_l1hdr.b_state == arc_anon || HDR_EMPTY(hdr)) {
mutex_exit(hash_lock);
mutex_exit(&buf->b_evict_lock);
ARCSTAT_BUMP(arcstat_access_skip);
return;
}
mutex_exit(&buf->b_evict_lock);
ASSERT(hdr->b_l1hdr.b_state == arc_mru ||
hdr->b_l1hdr.b_state == arc_mfu);
DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr);
arc_access(hdr, hash_lock);
mutex_exit(hash_lock);
ARCSTAT_BUMP(arcstat_hits);
ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr) && !HDR_PRESCIENT_PREFETCH(hdr),
demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data, metadata, hits);
}
/* a generic arc_read_done_func_t which you can use */
/* ARGSUSED */
void
arc_bcopy_func(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp,
arc_buf_t *buf, void *arg)
{
if (buf == NULL)
return;
bcopy(buf->b_data, arg, arc_buf_size(buf));
arc_buf_destroy(buf, arg);
}
/* a generic arc_read_done_func_t */
/* ARGSUSED */
void
arc_getbuf_func(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp,
arc_buf_t *buf, void *arg)
{
arc_buf_t **bufp = arg;
if (buf == NULL) {
ASSERT(zio == NULL || zio->io_error != 0);
*bufp = NULL;
} else {
ASSERT(zio == NULL || zio->io_error == 0);
*bufp = buf;
ASSERT(buf->b_data != NULL);
}
}
static void
arc_hdr_verify(arc_buf_hdr_t *hdr, blkptr_t *bp)
{
if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) {
ASSERT3U(HDR_GET_PSIZE(hdr), ==, 0);
ASSERT3U(arc_hdr_get_compress(hdr), ==, ZIO_COMPRESS_OFF);
} else {
if (HDR_COMPRESSION_ENABLED(hdr)) {
ASSERT3U(arc_hdr_get_compress(hdr), ==,
BP_GET_COMPRESS(bp));
}
ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(bp));
ASSERT3U(HDR_GET_PSIZE(hdr), ==, BP_GET_PSIZE(bp));
ASSERT3U(!!HDR_PROTECTED(hdr), ==, BP_IS_PROTECTED(bp));
}
}
static void
arc_read_done(zio_t *zio)
{
blkptr_t *bp = zio->io_bp;
arc_buf_hdr_t *hdr = zio->io_private;
kmutex_t *hash_lock = NULL;
arc_callback_t *callback_list;
arc_callback_t *acb;
boolean_t freeable = B_FALSE;
/*
* The hdr was inserted into hash-table and removed from lists
* prior to starting I/O. We should find this header, since
* it's in the hash table, and it should be legit since it's
* not possible to evict it during the I/O. The only possible
* reason for it not to be found is if we were freed during the
* read.
*/
if (HDR_IN_HASH_TABLE(hdr)) {
arc_buf_hdr_t *found;
ASSERT3U(hdr->b_birth, ==, BP_PHYSICAL_BIRTH(zio->io_bp));
ASSERT3U(hdr->b_dva.dva_word[0], ==,
BP_IDENTITY(zio->io_bp)->dva_word[0]);
ASSERT3U(hdr->b_dva.dva_word[1], ==,
BP_IDENTITY(zio->io_bp)->dva_word[1]);
found = buf_hash_find(hdr->b_spa, zio->io_bp, &hash_lock);
ASSERT((found == hdr &&
DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) ||
(found == hdr && HDR_L2_READING(hdr)));
ASSERT3P(hash_lock, !=, NULL);
}
if (BP_IS_PROTECTED(bp)) {
hdr->b_crypt_hdr.b_ot = BP_GET_TYPE(bp);
hdr->b_crypt_hdr.b_dsobj = zio->io_bookmark.zb_objset;
zio_crypt_decode_params_bp(bp, hdr->b_crypt_hdr.b_salt,
hdr->b_crypt_hdr.b_iv);
if (BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) {
void *tmpbuf;
tmpbuf = abd_borrow_buf_copy(zio->io_abd,
sizeof (zil_chain_t));
zio_crypt_decode_mac_zil(tmpbuf,
hdr->b_crypt_hdr.b_mac);
abd_return_buf(zio->io_abd, tmpbuf,
sizeof (zil_chain_t));
} else {
zio_crypt_decode_mac_bp(bp, hdr->b_crypt_hdr.b_mac);
}
}
if (zio->io_error == 0) {
/* byteswap if necessary */
if (BP_SHOULD_BYTESWAP(zio->io_bp)) {
if (BP_GET_LEVEL(zio->io_bp) > 0) {
hdr->b_l1hdr.b_byteswap = DMU_BSWAP_UINT64;
} else {
hdr->b_l1hdr.b_byteswap =
DMU_OT_BYTESWAP(BP_GET_TYPE(zio->io_bp));
}
} else {
hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS;
}
if (!HDR_L2_READING(hdr)) {
hdr->b_complevel = zio->io_prop.zp_complevel;
}
}
arc_hdr_clear_flags(hdr, ARC_FLAG_L2_EVICTED);
if (l2arc_noprefetch && HDR_PREFETCH(hdr))
arc_hdr_clear_flags(hdr, ARC_FLAG_L2CACHE);
callback_list = hdr->b_l1hdr.b_acb;
ASSERT3P(callback_list, !=, NULL);
if (hash_lock && zio->io_error == 0 &&
hdr->b_l1hdr.b_state == arc_anon) {
/*
* Only call arc_access on anonymous buffers. This is because
* if we've issued an I/O for an evicted buffer, we've already
* called arc_access (to prevent any simultaneous readers from
* getting confused).
*/
arc_access(hdr, hash_lock);
}
/*
* If a read request has a callback (i.e. acb_done is not NULL), then we
* make a buf containing the data according to the parameters which were
* passed in. The implementation of arc_buf_alloc_impl() ensures that we
* aren't needlessly decompressing the data multiple times.
*/
int callback_cnt = 0;
for (acb = callback_list; acb != NULL; acb = acb->acb_next) {
if (!acb->acb_done || acb->acb_nobuf)
continue;
callback_cnt++;
if (zio->io_error != 0)
continue;
int error = arc_buf_alloc_impl(hdr, zio->io_spa,
&acb->acb_zb, acb->acb_private, acb->acb_encrypted,
acb->acb_compressed, acb->acb_noauth, B_TRUE,
&acb->acb_buf);
/*
* Assert non-speculative zios didn't fail because an
* encryption key wasn't loaded
*/
ASSERT((zio->io_flags & ZIO_FLAG_SPECULATIVE) ||
error != EACCES);
/*
* If we failed to decrypt, report an error now (as the zio
* layer would have done if it had done the transforms).
*/
if (error == ECKSUM) {
ASSERT(BP_IS_PROTECTED(bp));
error = SET_ERROR(EIO);
if ((zio->io_flags & ZIO_FLAG_SPECULATIVE) == 0) {
spa_log_error(zio->io_spa, &acb->acb_zb);
(void) zfs_ereport_post(
FM_EREPORT_ZFS_AUTHENTICATION,
zio->io_spa, NULL, &acb->acb_zb, zio, 0);
}
}
if (error != 0) {
/*
* Decompression or decryption failed. Set
* io_error so that when we call acb_done
* (below), we will indicate that the read
* failed. Note that in the unusual case
* where one callback is compressed and another
* uncompressed, we will mark all of them
* as failed, even though the uncompressed
* one can't actually fail. In this case,
* the hdr will not be anonymous, because
* if there are multiple callbacks, it's
* because multiple threads found the same
* arc buf in the hash table.
*/
zio->io_error = error;
}
}
/*
* If there are multiple callbacks, we must have the hash lock,
* because the only way for multiple threads to find this hdr is
* in the hash table. This ensures that if there are multiple
* callbacks, the hdr is not anonymous. If it were anonymous,
* we couldn't use arc_buf_destroy() in the error case below.
*/
ASSERT(callback_cnt < 2 || hash_lock != NULL);
hdr->b_l1hdr.b_acb = NULL;
arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS);
if (callback_cnt == 0)
ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr));
ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt) ||
callback_list != NULL);
if (zio->io_error == 0) {
arc_hdr_verify(hdr, zio->io_bp);
} else {
arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR);
if (hdr->b_l1hdr.b_state != arc_anon)
arc_change_state(arc_anon, hdr, hash_lock);
if (HDR_IN_HASH_TABLE(hdr))
buf_hash_remove(hdr);
freeable = zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt);
}
/*
* Broadcast before we drop the hash_lock to avoid the possibility
* that the hdr (and hence the cv) might be freed before we get to
* the cv_broadcast().
*/
cv_broadcast(&hdr->b_l1hdr.b_cv);
if (hash_lock != NULL) {
mutex_exit(hash_lock);
} else {
/*
* This block was freed while we waited for the read to
* complete. It has been removed from the hash table and
* moved to the anonymous state (so that it won't show up
* in the cache).
*/
ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon);
freeable = zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt);
}
/* execute each callback and free its structure */
while ((acb = callback_list) != NULL) {
if (acb->acb_done != NULL) {
if (zio->io_error != 0 && acb->acb_buf != NULL) {
/*
* If arc_buf_alloc_impl() fails during
* decompression, the buf will still be
* allocated, and needs to be freed here.
*/
arc_buf_destroy(acb->acb_buf,
acb->acb_private);
acb->acb_buf = NULL;
}
acb->acb_done(zio, &zio->io_bookmark, zio->io_bp,
acb->acb_buf, acb->acb_private);
}
if (acb->acb_zio_dummy != NULL) {
acb->acb_zio_dummy->io_error = zio->io_error;
zio_nowait(acb->acb_zio_dummy);
}
callback_list = acb->acb_next;
kmem_free(acb, sizeof (arc_callback_t));
}
if (freeable)
arc_hdr_destroy(hdr);
}
/*
* "Read" the block at the specified DVA (in bp) via the
* cache. If the block is found in the cache, invoke the provided
* callback immediately and return. Note that the `zio' parameter
* in the callback will be NULL in this case, since no IO was
* required. If the block is not in the cache pass the read request
* on to the spa with a substitute callback function, so that the
* requested block will be added to the cache.
*
* If a read request arrives for a block that has a read in-progress,
* either wait for the in-progress read to complete (and return the
* results); or, if this is a read with a "done" func, add a record
* to the read to invoke the "done" func when the read completes,
* and return; or just return.
*
* arc_read_done() will invoke all the requested "done" functions
* for readers of this block.
*/
int
arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
arc_read_done_func_t *done, void *private, zio_priority_t priority,
int zio_flags, arc_flags_t *arc_flags, const zbookmark_phys_t *zb)
{
arc_buf_hdr_t *hdr = NULL;
kmutex_t *hash_lock = NULL;
zio_t *rzio;
uint64_t guid = spa_load_guid(spa);
boolean_t compressed_read = (zio_flags & ZIO_FLAG_RAW_COMPRESS) != 0;
boolean_t encrypted_read = BP_IS_ENCRYPTED(bp) &&
(zio_flags & ZIO_FLAG_RAW_ENCRYPT) != 0;
boolean_t noauth_read = BP_IS_AUTHENTICATED(bp) &&
(zio_flags & ZIO_FLAG_RAW_ENCRYPT) != 0;
boolean_t embedded_bp = !!BP_IS_EMBEDDED(bp);
boolean_t no_buf = *arc_flags & ARC_FLAG_NO_BUF;
int rc = 0;
ASSERT(!embedded_bp ||
BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
ASSERT(!BP_IS_HOLE(bp));
ASSERT(!BP_IS_REDACTED(bp));
/*
* Normally SPL_FSTRANS will already be set since kernel threads which
* expect to call the DMU interfaces will set it when created. System
* calls are similarly handled by setting/cleaning the bit in the
* registered callback (module/os/.../zfs/zpl_*).
*
* External consumers such as Lustre which call the exported DMU
* interfaces may not have set SPL_FSTRANS. To avoid a deadlock
* on the hash_lock always set and clear the bit.
*/
fstrans_cookie_t cookie = spl_fstrans_mark();
top:
if (!embedded_bp) {
/*
* Embedded BP's have no DVA and require no I/O to "read".
* Create an anonymous arc buf to back it.
*/
if (!zfs_blkptr_verify(spa, bp, zio_flags &
ZIO_FLAG_CONFIG_WRITER, BLK_VERIFY_LOG)) {
rc = SET_ERROR(ECKSUM);
goto out;
}
hdr = buf_hash_find(guid, bp, &hash_lock);
}
/*
* Determine if we have an L1 cache hit or a cache miss. For simplicity
* we maintain encrypted data separately from compressed / uncompressed
* data. If the user is requesting raw encrypted data and we don't have
* that in the header we will read from disk to guarantee that we can
* get it even if the encryption keys aren't loaded.
*/
if (hdr != NULL && HDR_HAS_L1HDR(hdr) && (HDR_HAS_RABD(hdr) ||
(hdr->b_l1hdr.b_pabd != NULL && !encrypted_read))) {
arc_buf_t *buf = NULL;
*arc_flags |= ARC_FLAG_CACHED;
if (HDR_IO_IN_PROGRESS(hdr)) {
zio_t *head_zio = hdr->b_l1hdr.b_acb->acb_zio_head;
if (*arc_flags & ARC_FLAG_CACHED_ONLY) {
mutex_exit(hash_lock);
ARCSTAT_BUMP(arcstat_cached_only_in_progress);
rc = SET_ERROR(ENOENT);
goto out;
}
ASSERT3P(head_zio, !=, NULL);
if ((hdr->b_flags & ARC_FLAG_PRIO_ASYNC_READ) &&
priority == ZIO_PRIORITY_SYNC_READ) {
/*
* This is a sync read that needs to wait for
* an in-flight async read. Request that the
* zio have its priority upgraded.
*/
zio_change_priority(head_zio, priority);
DTRACE_PROBE1(arc__async__upgrade__sync,
arc_buf_hdr_t *, hdr);
ARCSTAT_BUMP(arcstat_async_upgrade_sync);
}
if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) {
arc_hdr_clear_flags(hdr,
ARC_FLAG_PREDICTIVE_PREFETCH);
}
if (*arc_flags & ARC_FLAG_WAIT) {
cv_wait(&hdr->b_l1hdr.b_cv, hash_lock);
mutex_exit(hash_lock);
goto top;
}
ASSERT(*arc_flags & ARC_FLAG_NOWAIT);
if (done) {
arc_callback_t *acb = NULL;
acb = kmem_zalloc(sizeof (arc_callback_t),
KM_SLEEP);
acb->acb_done = done;
acb->acb_private = private;
acb->acb_compressed = compressed_read;
acb->acb_encrypted = encrypted_read;
acb->acb_noauth = noauth_read;
acb->acb_nobuf = no_buf;
acb->acb_zb = *zb;
if (pio != NULL)
acb->acb_zio_dummy = zio_null(pio,
spa, NULL, NULL, NULL, zio_flags);
ASSERT3P(acb->acb_done, !=, NULL);
acb->acb_zio_head = head_zio;
acb->acb_next = hdr->b_l1hdr.b_acb;
hdr->b_l1hdr.b_acb = acb;
}
mutex_exit(hash_lock);
goto out;
}
ASSERT(hdr->b_l1hdr.b_state == arc_mru ||
hdr->b_l1hdr.b_state == arc_mfu);
if (done && !no_buf) {
if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) {
/*
* This is a demand read which does not have to
* wait for i/o because we did a predictive
* prefetch i/o for it, which has completed.
*/
DTRACE_PROBE1(
arc__demand__hit__predictive__prefetch,
arc_buf_hdr_t *, hdr);
ARCSTAT_BUMP(
arcstat_demand_hit_predictive_prefetch);
arc_hdr_clear_flags(hdr,
ARC_FLAG_PREDICTIVE_PREFETCH);
}
if (hdr->b_flags & ARC_FLAG_PRESCIENT_PREFETCH) {
ARCSTAT_BUMP(
arcstat_demand_hit_prescient_prefetch);
arc_hdr_clear_flags(hdr,
ARC_FLAG_PRESCIENT_PREFETCH);
}
ASSERT(!embedded_bp || !BP_IS_HOLE(bp));
/* Get a buf with the desired data in it. */
rc = arc_buf_alloc_impl(hdr, spa, zb, private,
encrypted_read, compressed_read, noauth_read,
B_TRUE, &buf);
if (rc == ECKSUM) {
/*
* Convert authentication and decryption errors
* to EIO (and generate an ereport if needed)
* before leaving the ARC.
*/
rc = SET_ERROR(EIO);
if ((zio_flags & ZIO_FLAG_SPECULATIVE) == 0) {
spa_log_error(spa, zb);
(void) zfs_ereport_post(
FM_EREPORT_ZFS_AUTHENTICATION,
spa, NULL, zb, NULL, 0);
}
}
if (rc != 0) {
(void) remove_reference(hdr, hash_lock,
private);
arc_buf_destroy_impl(buf);
buf = NULL;
}
/* assert any errors weren't due to unloaded keys */
ASSERT((zio_flags & ZIO_FLAG_SPECULATIVE) ||
rc != EACCES);
} else if (*arc_flags & ARC_FLAG_PREFETCH &&
zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) {
if (HDR_HAS_L2HDR(hdr))
l2arc_hdr_arcstats_decrement_state(hdr);
arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH);
if (HDR_HAS_L2HDR(hdr))
l2arc_hdr_arcstats_increment_state(hdr);
}
DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr);
arc_access(hdr, hash_lock);
if (*arc_flags & ARC_FLAG_PRESCIENT_PREFETCH)
arc_hdr_set_flags(hdr, ARC_FLAG_PRESCIENT_PREFETCH);
if (*arc_flags & ARC_FLAG_L2CACHE)
arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE);
mutex_exit(hash_lock);
ARCSTAT_BUMP(arcstat_hits);
ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr),
demand, prefetch, !HDR_ISTYPE_METADATA(hdr),
data, metadata, hits);
if (done)
done(NULL, zb, bp, buf, private);
} else {
uint64_t lsize = BP_GET_LSIZE(bp);
uint64_t psize = BP_GET_PSIZE(bp);
arc_callback_t *acb;
vdev_t *vd = NULL;
uint64_t addr = 0;
boolean_t devw = B_FALSE;
uint64_t size;
abd_t *hdr_abd;
int alloc_flags = encrypted_read ? ARC_HDR_ALLOC_RDATA : 0;
if (*arc_flags & ARC_FLAG_CACHED_ONLY) {
rc = SET_ERROR(ENOENT);
if (hash_lock != NULL)
mutex_exit(hash_lock);
goto out;
}
if (hdr == NULL) {
/*
* This block is not in the cache or it has
* embedded data.
*/
arc_buf_hdr_t *exists = NULL;
arc_buf_contents_t type = BP_GET_BUFC_TYPE(bp);
hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize,
BP_IS_PROTECTED(bp), BP_GET_COMPRESS(bp), 0, type,
encrypted_read);
if (!embedded_bp) {
hdr->b_dva = *BP_IDENTITY(bp);
hdr->b_birth = BP_PHYSICAL_BIRTH(bp);
exists = buf_hash_insert(hdr, &hash_lock);
}
if (exists != NULL) {
/* somebody beat us to the hash insert */
mutex_exit(hash_lock);
buf_discard_identity(hdr);
arc_hdr_destroy(hdr);
goto top; /* restart the IO request */
}
} else {
/*
* This block is in the ghost cache or encrypted data
* was requested and we didn't have it. If it was
* L2-only (and thus didn't have an L1 hdr),
* we realloc the header to add an L1 hdr.
*/
if (!HDR_HAS_L1HDR(hdr)) {
hdr = arc_hdr_realloc(hdr, hdr_l2only_cache,
hdr_full_cache);
}
if (GHOST_STATE(hdr->b_l1hdr.b_state)) {
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
ASSERT(!HDR_HAS_RABD(hdr));
ASSERT(!HDR_IO_IN_PROGRESS(hdr));
ASSERT0(zfs_refcount_count(
&hdr->b_l1hdr.b_refcnt));
ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL);
ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL);
} else if (HDR_IO_IN_PROGRESS(hdr)) {
/*
* If this header already had an IO in progress
* and we are performing another IO to fetch
* encrypted data we must wait until the first
* IO completes so as not to confuse
* arc_read_done(). This should be very rare
* and so the performance impact shouldn't
* matter.
*/
cv_wait(&hdr->b_l1hdr.b_cv, hash_lock);
mutex_exit(hash_lock);
goto top;
}
/*
* This is a delicate dance that we play here.
* This hdr might be in the ghost list so we access
* it to move it out of the ghost list before we
* initiate the read. If it's a prefetch then
* it won't have a callback so we'll remove the
* reference that arc_buf_alloc_impl() created. We
* do this after we've called arc_access() to
* avoid hitting an assert in remove_reference().
*/
arc_adapt(arc_hdr_size(hdr), hdr->b_l1hdr.b_state);
arc_access(hdr, hash_lock);
arc_hdr_alloc_abd(hdr, alloc_flags);
}
if (encrypted_read) {
ASSERT(HDR_HAS_RABD(hdr));
size = HDR_GET_PSIZE(hdr);
hdr_abd = hdr->b_crypt_hdr.b_rabd;
zio_flags |= ZIO_FLAG_RAW;
} else {
ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
size = arc_hdr_size(hdr);
hdr_abd = hdr->b_l1hdr.b_pabd;
if (arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF) {
zio_flags |= ZIO_FLAG_RAW_COMPRESS;
}
/*
* For authenticated bp's, we do not ask the ZIO layer
* to authenticate them since this will cause the entire
* IO to fail if the key isn't loaded. Instead, we
* defer authentication until arc_buf_fill(), which will
* verify the data when the key is available.
*/
if (BP_IS_AUTHENTICATED(bp))
zio_flags |= ZIO_FLAG_RAW_ENCRYPT;
}
if (*arc_flags & ARC_FLAG_PREFETCH &&
zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) {
if (HDR_HAS_L2HDR(hdr))
l2arc_hdr_arcstats_decrement_state(hdr);
arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH);
if (HDR_HAS_L2HDR(hdr))
l2arc_hdr_arcstats_increment_state(hdr);
}
if (*arc_flags & ARC_FLAG_PRESCIENT_PREFETCH)
arc_hdr_set_flags(hdr, ARC_FLAG_PRESCIENT_PREFETCH);
if (*arc_flags & ARC_FLAG_L2CACHE)
arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE);
if (BP_IS_AUTHENTICATED(bp))
arc_hdr_set_flags(hdr, ARC_FLAG_NOAUTH);
if (BP_GET_LEVEL(bp) > 0)
arc_hdr_set_flags(hdr, ARC_FLAG_INDIRECT);
if (*arc_flags & ARC_FLAG_PREDICTIVE_PREFETCH)
arc_hdr_set_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH);
ASSERT(!GHOST_STATE(hdr->b_l1hdr.b_state));
acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP);
acb->acb_done = done;
acb->acb_private = private;
acb->acb_compressed = compressed_read;
acb->acb_encrypted = encrypted_read;
acb->acb_noauth = noauth_read;
acb->acb_zb = *zb;
ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL);
hdr->b_l1hdr.b_acb = acb;
arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS);
if (HDR_HAS_L2HDR(hdr) &&
(vd = hdr->b_l2hdr.b_dev->l2ad_vdev) != NULL) {
devw = hdr->b_l2hdr.b_dev->l2ad_writing;
addr = hdr->b_l2hdr.b_daddr;
/*
* Lock out L2ARC device removal.
*/
if (vdev_is_dead(vd) ||
!spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER))
vd = NULL;
}
/*
* We count both async reads and scrub IOs as asynchronous so
* that both can be upgraded in the event of a cache hit while
* the read IO is still in-flight.
*/
if (priority == ZIO_PRIORITY_ASYNC_READ ||
priority == ZIO_PRIORITY_SCRUB)
arc_hdr_set_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ);
else
arc_hdr_clear_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ);
/*
* At this point, we have a level 1 cache miss or a blkptr
* with embedded data. Try again in L2ARC if possible.
*/
ASSERT3U(HDR_GET_LSIZE(hdr), ==, lsize);
/*
* Skip ARC stat bump for block pointers with embedded
* data. The data are read from the blkptr itself via
* decode_embedded_bp_compressed().
*/
if (!embedded_bp) {
DTRACE_PROBE4(arc__miss, arc_buf_hdr_t *, hdr,
blkptr_t *, bp, uint64_t, lsize,
zbookmark_phys_t *, zb);
ARCSTAT_BUMP(arcstat_misses);
ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr),
demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data,
metadata, misses);
zfs_racct_read(size, 1);
}
/* Check if the spa even has l2 configured */
const boolean_t spa_has_l2 = l2arc_ndev != 0 &&
spa->spa_l2cache.sav_count > 0;
if (vd != NULL && spa_has_l2 && !(l2arc_norw && devw)) {
/*
* Read from the L2ARC if the following are true:
* 1. The L2ARC vdev was previously cached.
* 2. This buffer still has L2ARC metadata.
* 3. This buffer isn't currently writing to the L2ARC.
* 4. The L2ARC entry wasn't evicted, which may
* also have invalidated the vdev.
* 5. This isn't prefetch or l2arc_noprefetch is 0.
*/
if (HDR_HAS_L2HDR(hdr) &&
!HDR_L2_WRITING(hdr) && !HDR_L2_EVICTED(hdr) &&
!(l2arc_noprefetch && HDR_PREFETCH(hdr))) {
l2arc_read_callback_t *cb;
abd_t *abd;
uint64_t asize;
DTRACE_PROBE1(l2arc__hit, arc_buf_hdr_t *, hdr);
ARCSTAT_BUMP(arcstat_l2_hits);
atomic_inc_32(&hdr->b_l2hdr.b_hits);
cb = kmem_zalloc(sizeof (l2arc_read_callback_t),
KM_SLEEP);
cb->l2rcb_hdr = hdr;
cb->l2rcb_bp = *bp;
cb->l2rcb_zb = *zb;
cb->l2rcb_flags = zio_flags;
/*
* When Compressed ARC is disabled, but the
* L2ARC block is compressed, arc_hdr_size()
* will have returned LSIZE rather than PSIZE.
*/
if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF &&
!HDR_COMPRESSION_ENABLED(hdr) &&
HDR_GET_PSIZE(hdr) != 0) {
size = HDR_GET_PSIZE(hdr);
}
asize = vdev_psize_to_asize(vd, size);
if (asize != size) {
abd = abd_alloc_for_io(asize,
HDR_ISTYPE_METADATA(hdr));
cb->l2rcb_abd = abd;
} else {
abd = hdr_abd;
}
ASSERT(addr >= VDEV_LABEL_START_SIZE &&
addr + asize <= vd->vdev_psize -
VDEV_LABEL_END_SIZE);
/*
* l2arc read. The SCL_L2ARC lock will be
* released by l2arc_read_done().
* Issue a null zio if the underlying buffer
* was squashed to zero size by compression.
*/
ASSERT3U(arc_hdr_get_compress(hdr), !=,
ZIO_COMPRESS_EMPTY);
rzio = zio_read_phys(pio, vd, addr,
asize, abd,
ZIO_CHECKSUM_OFF,
l2arc_read_done, cb, priority,
zio_flags | ZIO_FLAG_DONT_CACHE |
ZIO_FLAG_CANFAIL |
ZIO_FLAG_DONT_PROPAGATE |
ZIO_FLAG_DONT_RETRY, B_FALSE);
acb->acb_zio_head = rzio;
if (hash_lock != NULL)
mutex_exit(hash_lock);
DTRACE_PROBE2(l2arc__read, vdev_t *, vd,
zio_t *, rzio);
ARCSTAT_INCR(arcstat_l2_read_bytes,
HDR_GET_PSIZE(hdr));
if (*arc_flags & ARC_FLAG_NOWAIT) {
zio_nowait(rzio);
goto out;
}
ASSERT(*arc_flags & ARC_FLAG_WAIT);
if (zio_wait(rzio) == 0)
goto out;
/* l2arc read error; goto zio_read() */
if (hash_lock != NULL)
mutex_enter(hash_lock);
} else {
DTRACE_PROBE1(l2arc__miss,
arc_buf_hdr_t *, hdr);
ARCSTAT_BUMP(arcstat_l2_misses);
if (HDR_L2_WRITING(hdr))
ARCSTAT_BUMP(arcstat_l2_rw_clash);
spa_config_exit(spa, SCL_L2ARC, vd);
}
} else {
if (vd != NULL)
spa_config_exit(spa, SCL_L2ARC, vd);
/*
* Only a spa with l2 should contribute to l2
* miss stats. (Including the case of having a
* faulted cache device - that's also a miss.)
*/
if (spa_has_l2) {
/*
* Skip ARC stat bump for block pointers with
* embedded data. The data are read from the
* blkptr itself via
* decode_embedded_bp_compressed().
*/
if (!embedded_bp) {
DTRACE_PROBE1(l2arc__miss,
arc_buf_hdr_t *, hdr);
ARCSTAT_BUMP(arcstat_l2_misses);
}
}
}
rzio = zio_read(pio, spa, bp, hdr_abd, size,
arc_read_done, hdr, priority, zio_flags, zb);
acb->acb_zio_head = rzio;
if (hash_lock != NULL)
mutex_exit(hash_lock);
if (*arc_flags & ARC_FLAG_WAIT) {
rc = zio_wait(rzio);
goto out;
}
ASSERT(*arc_flags & ARC_FLAG_NOWAIT);
zio_nowait(rzio);
}
out:
/* embedded bps don't actually go to disk */
if (!embedded_bp)
spa_read_history_add(spa, zb, *arc_flags);
spl_fstrans_unmark(cookie);
return (rc);
}
arc_prune_t *
arc_add_prune_callback(arc_prune_func_t *func, void *private)
{
arc_prune_t *p;
p = kmem_alloc(sizeof (*p), KM_SLEEP);
p->p_pfunc = func;
p->p_private = private;
list_link_init(&p->p_node);
zfs_refcount_create(&p->p_refcnt);
mutex_enter(&arc_prune_mtx);
zfs_refcount_add(&p->p_refcnt, &arc_prune_list);
list_insert_head(&arc_prune_list, p);
mutex_exit(&arc_prune_mtx);
return (p);
}
void
arc_remove_prune_callback(arc_prune_t *p)
{
boolean_t wait = B_FALSE;
mutex_enter(&arc_prune_mtx);
list_remove(&arc_prune_list, p);
if (zfs_refcount_remove(&p->p_refcnt, &arc_prune_list) > 0)
wait = B_TRUE;
mutex_exit(&arc_prune_mtx);
/* wait for arc_prune_task to finish */
if (wait)
taskq_wait_outstanding(arc_prune_taskq, 0);
ASSERT0(zfs_refcount_count(&p->p_refcnt));
zfs_refcount_destroy(&p->p_refcnt);
kmem_free(p, sizeof (*p));
}
/*
* Notify the arc that a block was freed, and thus will never be used again.
*/
void
arc_freed(spa_t *spa, const blkptr_t *bp)
{
arc_buf_hdr_t *hdr;
kmutex_t *hash_lock;
uint64_t guid = spa_load_guid(spa);
ASSERT(!BP_IS_EMBEDDED(bp));
hdr = buf_hash_find(guid, bp, &hash_lock);
if (hdr == NULL)
return;
/*
* We might be trying to free a block that is still doing I/O
* (i.e. prefetch) or has a reference (i.e. a dedup-ed,
* dmu_sync-ed block). If this block is being prefetched, then it
* would still have the ARC_FLAG_IO_IN_PROGRESS flag set on the hdr
* until the I/O completes. A block may also have a reference if it is
* part of a dedup-ed, dmu_synced write. The dmu_sync() function would
* have written the new block to its final resting place on disk but
* without the dedup flag set. This would have left the hdr in the MRU
* state and discoverable. When the txg finally syncs it detects that
* the block was overridden in open context and issues an override I/O.
* Since this is a dedup block, the override I/O will determine if the
* block is already in the DDT. If so, then it will replace the io_bp
* with the bp from the DDT and allow the I/O to finish. When the I/O
* reaches the done callback, dbuf_write_override_done, it will
* check to see if the io_bp and io_bp_override are identical.
* If they are not, then it indicates that the bp was replaced with
* the bp in the DDT and the override bp is freed. This allows
* us to arrive here with a reference on a block that is being
* freed. So if we have an I/O in progress, or a reference to
* this hdr, then we don't destroy the hdr.
*/
if (!HDR_HAS_L1HDR(hdr) || (!HDR_IO_IN_PROGRESS(hdr) &&
zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt))) {
arc_change_state(arc_anon, hdr, hash_lock);
arc_hdr_destroy(hdr);
mutex_exit(hash_lock);
} else {
mutex_exit(hash_lock);
}
}
/*
* Release this buffer from the cache, making it an anonymous buffer. This
* must be done after a read and prior to modifying the buffer contents.
* If the buffer has more than one reference, we must make
* a new hdr for the buffer.
*/
void
arc_release(arc_buf_t *buf, void *tag)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
/*
* It would be nice to assert that if its DMU metadata (level >
* 0 || it's the dnode file), then it must be syncing context.
* But we don't know that information at this level.
*/
mutex_enter(&buf->b_evict_lock);
ASSERT(HDR_HAS_L1HDR(hdr));
/*
* We don't grab the hash lock prior to this check, because if
* the buffer's header is in the arc_anon state, it won't be
* linked into the hash table.
*/
if (hdr->b_l1hdr.b_state == arc_anon) {
mutex_exit(&buf->b_evict_lock);
ASSERT(!HDR_IO_IN_PROGRESS(hdr));
ASSERT(!HDR_IN_HASH_TABLE(hdr));
ASSERT(!HDR_HAS_L2HDR(hdr));
ASSERT(HDR_EMPTY(hdr));
ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1);
ASSERT3S(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt), ==, 1);
ASSERT(!list_link_active(&hdr->b_l1hdr.b_arc_node));
hdr->b_l1hdr.b_arc_access = 0;
/*
* If the buf is being overridden then it may already
* have a hdr that is not empty.
*/
buf_discard_identity(hdr);
arc_buf_thaw(buf);
return;
}
kmutex_t *hash_lock = HDR_LOCK(hdr);
mutex_enter(hash_lock);
/*
* This assignment is only valid as long as the hash_lock is
* held, we must be careful not to reference state or the
* b_state field after dropping the lock.
*/
arc_state_t *state = hdr->b_l1hdr.b_state;
ASSERT3P(hash_lock, ==, HDR_LOCK(hdr));
ASSERT3P(state, !=, arc_anon);
/* this buffer is not on any list */
ASSERT3S(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt), >, 0);
if (HDR_HAS_L2HDR(hdr)) {
mutex_enter(&hdr->b_l2hdr.b_dev->l2ad_mtx);
/*
* We have to recheck this conditional again now that
* we're holding the l2ad_mtx to prevent a race with
* another thread which might be concurrently calling
* l2arc_evict(). In that case, l2arc_evict() might have
* destroyed the header's L2 portion as we were waiting
* to acquire the l2ad_mtx.
*/
if (HDR_HAS_L2HDR(hdr))
arc_hdr_l2hdr_destroy(hdr);
mutex_exit(&hdr->b_l2hdr.b_dev->l2ad_mtx);
}
/*
* Do we have more than one buf?
*/
if (hdr->b_l1hdr.b_bufcnt > 1) {
arc_buf_hdr_t *nhdr;
uint64_t spa = hdr->b_spa;
uint64_t psize = HDR_GET_PSIZE(hdr);
uint64_t lsize = HDR_GET_LSIZE(hdr);
boolean_t protected = HDR_PROTECTED(hdr);
enum zio_compress compress = arc_hdr_get_compress(hdr);
arc_buf_contents_t type = arc_buf_type(hdr);
VERIFY3U(hdr->b_type, ==, type);
ASSERT(hdr->b_l1hdr.b_buf != buf || buf->b_next != NULL);
(void) remove_reference(hdr, hash_lock, tag);
if (arc_buf_is_shared(buf) && !ARC_BUF_COMPRESSED(buf)) {
ASSERT3P(hdr->b_l1hdr.b_buf, !=, buf);
ASSERT(ARC_BUF_LAST(buf));
}
/*
* Pull the data off of this hdr and attach it to
* a new anonymous hdr. Also find the last buffer
* in the hdr's buffer list.
*/
arc_buf_t *lastbuf = arc_buf_remove(hdr, buf);
ASSERT3P(lastbuf, !=, NULL);
/*
* If the current arc_buf_t and the hdr are sharing their data
* buffer, then we must stop sharing that block.
*/
if (arc_buf_is_shared(buf)) {
ASSERT3P(hdr->b_l1hdr.b_buf, !=, buf);
VERIFY(!arc_buf_is_shared(lastbuf));
/*
* First, sever the block sharing relationship between
* buf and the arc_buf_hdr_t.
*/
arc_unshare_buf(hdr, buf);
/*
* Now we need to recreate the hdr's b_pabd. Since we
* have lastbuf handy, we try to share with it, but if
* we can't then we allocate a new b_pabd and copy the
* data from buf into it.
*/
if (arc_can_share(hdr, lastbuf)) {
arc_share_buf(hdr, lastbuf);
} else {
arc_hdr_alloc_abd(hdr, ARC_HDR_DO_ADAPT);
abd_copy_from_buf(hdr->b_l1hdr.b_pabd,
buf->b_data, psize);
}
VERIFY3P(lastbuf->b_data, !=, NULL);
} else if (HDR_SHARED_DATA(hdr)) {
/*
* Uncompressed shared buffers are always at the end
* of the list. Compressed buffers don't have the
* same requirements. This makes it hard to
* simply assert that the lastbuf is shared so
* we rely on the hdr's compression flags to determine
* if we have a compressed, shared buffer.
*/
ASSERT(arc_buf_is_shared(lastbuf) ||
arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF);
ASSERT(!ARC_BUF_SHARED(buf));
}
ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr));
ASSERT3P(state, !=, arc_l2c_only);
(void) zfs_refcount_remove_many(&state->arcs_size,
arc_buf_size(buf), buf);
if (zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) {
ASSERT3P(state, !=, arc_l2c_only);
(void) zfs_refcount_remove_many(
&state->arcs_esize[type],
arc_buf_size(buf), buf);
}
hdr->b_l1hdr.b_bufcnt -= 1;
if (ARC_BUF_ENCRYPTED(buf))
hdr->b_crypt_hdr.b_ebufcnt -= 1;
arc_cksum_verify(buf);
arc_buf_unwatch(buf);
/* if this is the last uncompressed buf free the checksum */
if (!arc_hdr_has_uncompressed_buf(hdr))
arc_cksum_free(hdr);
mutex_exit(hash_lock);
/*
* Allocate a new hdr. The new hdr will contain a b_pabd
* buffer which will be freed in arc_write().
*/
nhdr = arc_hdr_alloc(spa, psize, lsize, protected,
compress, hdr->b_complevel, type, HDR_HAS_RABD(hdr));
ASSERT3P(nhdr->b_l1hdr.b_buf, ==, NULL);
ASSERT0(nhdr->b_l1hdr.b_bufcnt);
ASSERT0(zfs_refcount_count(&nhdr->b_l1hdr.b_refcnt));
VERIFY3U(nhdr->b_type, ==, type);
ASSERT(!HDR_SHARED_DATA(nhdr));
nhdr->b_l1hdr.b_buf = buf;
nhdr->b_l1hdr.b_bufcnt = 1;
if (ARC_BUF_ENCRYPTED(buf))
nhdr->b_crypt_hdr.b_ebufcnt = 1;
nhdr->b_l1hdr.b_mru_hits = 0;
nhdr->b_l1hdr.b_mru_ghost_hits = 0;
nhdr->b_l1hdr.b_mfu_hits = 0;
nhdr->b_l1hdr.b_mfu_ghost_hits = 0;
nhdr->b_l1hdr.b_l2_hits = 0;
(void) zfs_refcount_add(&nhdr->b_l1hdr.b_refcnt, tag);
buf->b_hdr = nhdr;
mutex_exit(&buf->b_evict_lock);
(void) zfs_refcount_add_many(&arc_anon->arcs_size,
arc_buf_size(buf), buf);
} else {
mutex_exit(&buf->b_evict_lock);
ASSERT(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt) == 1);
/* protected by hash lock, or hdr is on arc_anon */
ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node));
ASSERT(!HDR_IO_IN_PROGRESS(hdr));
hdr->b_l1hdr.b_mru_hits = 0;
hdr->b_l1hdr.b_mru_ghost_hits = 0;
hdr->b_l1hdr.b_mfu_hits = 0;
hdr->b_l1hdr.b_mfu_ghost_hits = 0;
hdr->b_l1hdr.b_l2_hits = 0;
arc_change_state(arc_anon, hdr, hash_lock);
hdr->b_l1hdr.b_arc_access = 0;
mutex_exit(hash_lock);
buf_discard_identity(hdr);
arc_buf_thaw(buf);
}
}
int
arc_released(arc_buf_t *buf)
{
int released;
mutex_enter(&buf->b_evict_lock);
released = (buf->b_data != NULL &&
buf->b_hdr->b_l1hdr.b_state == arc_anon);
mutex_exit(&buf->b_evict_lock);
return (released);
}
#ifdef ZFS_DEBUG
int
arc_referenced(arc_buf_t *buf)
{
int referenced;
mutex_enter(&buf->b_evict_lock);
referenced = (zfs_refcount_count(&buf->b_hdr->b_l1hdr.b_refcnt));
mutex_exit(&buf->b_evict_lock);
return (referenced);
}
#endif
static void
arc_write_ready(zio_t *zio)
{
arc_write_callback_t *callback = zio->io_private;
arc_buf_t *buf = callback->awcb_buf;
arc_buf_hdr_t *hdr = buf->b_hdr;
blkptr_t *bp = zio->io_bp;
uint64_t psize = BP_IS_HOLE(bp) ? 0 : BP_GET_PSIZE(bp);
fstrans_cookie_t cookie = spl_fstrans_mark();
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT(!zfs_refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt));
ASSERT(hdr->b_l1hdr.b_bufcnt > 0);
/*
* If we're reexecuting this zio because the pool suspended, then
* cleanup any state that was previously set the first time the
* callback was invoked.
*/
if (zio->io_flags & ZIO_FLAG_REEXECUTED) {
arc_cksum_free(hdr);
arc_buf_unwatch(buf);
if (hdr->b_l1hdr.b_pabd != NULL) {
if (arc_buf_is_shared(buf)) {
arc_unshare_buf(hdr, buf);
} else {
arc_hdr_free_abd(hdr, B_FALSE);
}
}
if (HDR_HAS_RABD(hdr))
arc_hdr_free_abd(hdr, B_TRUE);
}
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
ASSERT(!HDR_HAS_RABD(hdr));
ASSERT(!HDR_SHARED_DATA(hdr));
ASSERT(!arc_buf_is_shared(buf));
callback->awcb_ready(zio, buf, callback->awcb_private);
if (HDR_IO_IN_PROGRESS(hdr))
ASSERT(zio->io_flags & ZIO_FLAG_REEXECUTED);
arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS);
if (BP_IS_PROTECTED(bp) != !!HDR_PROTECTED(hdr))
hdr = arc_hdr_realloc_crypt(hdr, BP_IS_PROTECTED(bp));
if (BP_IS_PROTECTED(bp)) {
/* ZIL blocks are written through zio_rewrite */
ASSERT3U(BP_GET_TYPE(bp), !=, DMU_OT_INTENT_LOG);
ASSERT(HDR_PROTECTED(hdr));
if (BP_SHOULD_BYTESWAP(bp)) {
if (BP_GET_LEVEL(bp) > 0) {
hdr->b_l1hdr.b_byteswap = DMU_BSWAP_UINT64;
} else {
hdr->b_l1hdr.b_byteswap =
DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
}
} else {
hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS;
}
hdr->b_crypt_hdr.b_ot = BP_GET_TYPE(bp);
hdr->b_crypt_hdr.b_dsobj = zio->io_bookmark.zb_objset;
zio_crypt_decode_params_bp(bp, hdr->b_crypt_hdr.b_salt,
hdr->b_crypt_hdr.b_iv);
zio_crypt_decode_mac_bp(bp, hdr->b_crypt_hdr.b_mac);
}
/*
* If this block was written for raw encryption but the zio layer
* ended up only authenticating it, adjust the buffer flags now.
*/
if (BP_IS_AUTHENTICATED(bp) && ARC_BUF_ENCRYPTED(buf)) {
arc_hdr_set_flags(hdr, ARC_FLAG_NOAUTH);
buf->b_flags &= ~ARC_BUF_FLAG_ENCRYPTED;
if (BP_GET_COMPRESS(bp) == ZIO_COMPRESS_OFF)
buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED;
} else if (BP_IS_HOLE(bp) && ARC_BUF_ENCRYPTED(buf)) {
buf->b_flags &= ~ARC_BUF_FLAG_ENCRYPTED;
buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED;
}
/* this must be done after the buffer flags are adjusted */
arc_cksum_compute(buf);
enum zio_compress compress;
if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) {
compress = ZIO_COMPRESS_OFF;
} else {
ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(bp));
compress = BP_GET_COMPRESS(bp);
}
HDR_SET_PSIZE(hdr, psize);
arc_hdr_set_compress(hdr, compress);
hdr->b_complevel = zio->io_prop.zp_complevel;
if (zio->io_error != 0 || psize == 0)
goto out;
/*
* Fill the hdr with data. If the buffer is encrypted we have no choice
* but to copy the data into b_radb. If the hdr is compressed, the data
* we want is available from the zio, otherwise we can take it from
* the buf.
*
* We might be able to share the buf's data with the hdr here. However,
* doing so would cause the ARC to be full of linear ABDs if we write a
* lot of shareable data. As a compromise, we check whether scattered
* ABDs are allowed, and assume that if they are then the user wants
* the ARC to be primarily filled with them regardless of the data being
* written. Therefore, if they're allowed then we allocate one and copy
* the data into it; otherwise, we share the data directly if we can.
*/
if (ARC_BUF_ENCRYPTED(buf)) {
ASSERT3U(psize, >, 0);
ASSERT(ARC_BUF_COMPRESSED(buf));
arc_hdr_alloc_abd(hdr, ARC_HDR_DO_ADAPT|ARC_HDR_ALLOC_RDATA);
abd_copy(hdr->b_crypt_hdr.b_rabd, zio->io_abd, psize);
} else if (zfs_abd_scatter_enabled || !arc_can_share(hdr, buf)) {
/*
* Ideally, we would always copy the io_abd into b_pabd, but the
* user may have disabled compressed ARC, thus we must check the
* hdr's compression setting rather than the io_bp's.
*/
if (BP_IS_ENCRYPTED(bp)) {
ASSERT3U(psize, >, 0);
arc_hdr_alloc_abd(hdr,
ARC_HDR_DO_ADAPT|ARC_HDR_ALLOC_RDATA);
abd_copy(hdr->b_crypt_hdr.b_rabd, zio->io_abd, psize);
} else if (arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF &&
!ARC_BUF_COMPRESSED(buf)) {
ASSERT3U(psize, >, 0);
arc_hdr_alloc_abd(hdr, ARC_HDR_DO_ADAPT);
abd_copy(hdr->b_l1hdr.b_pabd, zio->io_abd, psize);
} else {
ASSERT3U(zio->io_orig_size, ==, arc_hdr_size(hdr));
arc_hdr_alloc_abd(hdr, ARC_HDR_DO_ADAPT);
abd_copy_from_buf(hdr->b_l1hdr.b_pabd, buf->b_data,
arc_buf_size(buf));
}
} else {
ASSERT3P(buf->b_data, ==, abd_to_buf(zio->io_orig_abd));
ASSERT3U(zio->io_orig_size, ==, arc_buf_size(buf));
ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1);
arc_share_buf(hdr, buf);
}
out:
arc_hdr_verify(hdr, bp);
spl_fstrans_unmark(cookie);
}
static void
arc_write_children_ready(zio_t *zio)
{
arc_write_callback_t *callback = zio->io_private;
arc_buf_t *buf = callback->awcb_buf;
callback->awcb_children_ready(zio, buf, callback->awcb_private);
}
/*
* The SPA calls this callback for each physical write that happens on behalf
* of a logical write. See the comment in dbuf_write_physdone() for details.
*/
static void
arc_write_physdone(zio_t *zio)
{
arc_write_callback_t *cb = zio->io_private;
if (cb->awcb_physdone != NULL)
cb->awcb_physdone(zio, cb->awcb_buf, cb->awcb_private);
}
static void
arc_write_done(zio_t *zio)
{
arc_write_callback_t *callback = zio->io_private;
arc_buf_t *buf = callback->awcb_buf;
arc_buf_hdr_t *hdr = buf->b_hdr;
ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL);
if (zio->io_error == 0) {
arc_hdr_verify(hdr, zio->io_bp);
if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) {
buf_discard_identity(hdr);
} else {
hdr->b_dva = *BP_IDENTITY(zio->io_bp);
hdr->b_birth = BP_PHYSICAL_BIRTH(zio->io_bp);
}
} else {
ASSERT(HDR_EMPTY(hdr));
}
/*
* If the block to be written was all-zero or compressed enough to be
* embedded in the BP, no write was performed so there will be no
* dva/birth/checksum. The buffer must therefore remain anonymous
* (and uncached).
*/
if (!HDR_EMPTY(hdr)) {
arc_buf_hdr_t *exists;
kmutex_t *hash_lock;
ASSERT3U(zio->io_error, ==, 0);
arc_cksum_verify(buf);
exists = buf_hash_insert(hdr, &hash_lock);
if (exists != NULL) {
/*
* This can only happen if we overwrite for
* sync-to-convergence, because we remove
* buffers from the hash table when we arc_free().
*/
if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp))
panic("bad overwrite, hdr=%p exists=%p",
(void *)hdr, (void *)exists);
ASSERT(zfs_refcount_is_zero(
&exists->b_l1hdr.b_refcnt));
arc_change_state(arc_anon, exists, hash_lock);
arc_hdr_destroy(exists);
mutex_exit(hash_lock);
exists = buf_hash_insert(hdr, &hash_lock);
ASSERT3P(exists, ==, NULL);
} else if (zio->io_flags & ZIO_FLAG_NOPWRITE) {
/* nopwrite */
ASSERT(zio->io_prop.zp_nopwrite);
if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp))
panic("bad nopwrite, hdr=%p exists=%p",
(void *)hdr, (void *)exists);
} else {
/* Dedup */
ASSERT(hdr->b_l1hdr.b_bufcnt == 1);
ASSERT(hdr->b_l1hdr.b_state == arc_anon);
ASSERT(BP_GET_DEDUP(zio->io_bp));
ASSERT(BP_GET_LEVEL(zio->io_bp) == 0);
}
}
arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS);
/* if it's not anon, we are doing a scrub */
if (exists == NULL && hdr->b_l1hdr.b_state == arc_anon)
arc_access(hdr, hash_lock);
mutex_exit(hash_lock);
} else {
arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS);
}
ASSERT(!zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt));
callback->awcb_done(zio, buf, callback->awcb_private);
abd_free(zio->io_abd);
kmem_free(callback, sizeof (arc_write_callback_t));
}
zio_t *
arc_write(zio_t *pio, spa_t *spa, uint64_t txg,
blkptr_t *bp, arc_buf_t *buf, boolean_t l2arc,
const zio_prop_t *zp, arc_write_done_func_t *ready,
arc_write_done_func_t *children_ready, arc_write_done_func_t *physdone,
arc_write_done_func_t *done, void *private, zio_priority_t priority,
int zio_flags, const zbookmark_phys_t *zb)
{
arc_buf_hdr_t *hdr = buf->b_hdr;
arc_write_callback_t *callback;
zio_t *zio;
zio_prop_t localprop = *zp;
ASSERT3P(ready, !=, NULL);
ASSERT3P(done, !=, NULL);
ASSERT(!HDR_IO_ERROR(hdr));
ASSERT(!HDR_IO_IN_PROGRESS(hdr));
ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL);
ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0);
if (l2arc)
arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE);
if (ARC_BUF_ENCRYPTED(buf)) {
ASSERT(ARC_BUF_COMPRESSED(buf));
localprop.zp_encrypt = B_TRUE;
localprop.zp_compress = HDR_GET_COMPRESS(hdr);
localprop.zp_complevel = hdr->b_complevel;
localprop.zp_byteorder =
(hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS) ?
ZFS_HOST_BYTEORDER : !ZFS_HOST_BYTEORDER;
bcopy(hdr->b_crypt_hdr.b_salt, localprop.zp_salt,
ZIO_DATA_SALT_LEN);
bcopy(hdr->b_crypt_hdr.b_iv, localprop.zp_iv,
ZIO_DATA_IV_LEN);
bcopy(hdr->b_crypt_hdr.b_mac, localprop.zp_mac,
ZIO_DATA_MAC_LEN);
if (DMU_OT_IS_ENCRYPTED(localprop.zp_type)) {
localprop.zp_nopwrite = B_FALSE;
localprop.zp_copies =
MIN(localprop.zp_copies, SPA_DVAS_PER_BP - 1);
}
zio_flags |= ZIO_FLAG_RAW;
} else if (ARC_BUF_COMPRESSED(buf)) {
ASSERT3U(HDR_GET_LSIZE(hdr), !=, arc_buf_size(buf));
localprop.zp_compress = HDR_GET_COMPRESS(hdr);
localprop.zp_complevel = hdr->b_complevel;
zio_flags |= ZIO_FLAG_RAW_COMPRESS;
}
callback = kmem_zalloc(sizeof (arc_write_callback_t), KM_SLEEP);
callback->awcb_ready = ready;
callback->awcb_children_ready = children_ready;
callback->awcb_physdone = physdone;
callback->awcb_done = done;
callback->awcb_private = private;
callback->awcb_buf = buf;
/*
* The hdr's b_pabd is now stale, free it now. A new data block
* will be allocated when the zio pipeline calls arc_write_ready().
*/
if (hdr->b_l1hdr.b_pabd != NULL) {
/*
* If the buf is currently sharing the data block with
* the hdr then we need to break that relationship here.
* The hdr will remain with a NULL data pointer and the
* buf will take sole ownership of the block.
*/
if (arc_buf_is_shared(buf)) {
arc_unshare_buf(hdr, buf);
} else {
arc_hdr_free_abd(hdr, B_FALSE);
}
VERIFY3P(buf->b_data, !=, NULL);
}
if (HDR_HAS_RABD(hdr))
arc_hdr_free_abd(hdr, B_TRUE);
if (!(zio_flags & ZIO_FLAG_RAW))
arc_hdr_set_compress(hdr, ZIO_COMPRESS_OFF);
ASSERT(!arc_buf_is_shared(buf));
ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
zio = zio_write(pio, spa, txg, bp,
abd_get_from_buf(buf->b_data, HDR_GET_LSIZE(hdr)),
HDR_GET_LSIZE(hdr), arc_buf_size(buf), &localprop, arc_write_ready,
(children_ready != NULL) ? arc_write_children_ready : NULL,
arc_write_physdone, arc_write_done, callback,
priority, zio_flags, zb);
return (zio);
}
void
arc_tempreserve_clear(uint64_t reserve)
{
atomic_add_64(&arc_tempreserve, -reserve);
ASSERT((int64_t)arc_tempreserve >= 0);
}
int
arc_tempreserve_space(spa_t *spa, uint64_t reserve, uint64_t txg)
{
int error;
uint64_t anon_size;
if (!arc_no_grow &&
reserve > arc_c/4 &&
reserve * 4 > (2ULL << SPA_MAXBLOCKSHIFT))
arc_c = MIN(arc_c_max, reserve * 4);
/*
* Throttle when the calculated memory footprint for the TXG
* exceeds the target ARC size.
*/
if (reserve > arc_c) {
DMU_TX_STAT_BUMP(dmu_tx_memory_reserve);
return (SET_ERROR(ERESTART));
}
/*
* Don't count loaned bufs as in flight dirty data to prevent long
* network delays from blocking transactions that are ready to be
* assigned to a txg.
*/
/* assert that it has not wrapped around */
ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes, 0), >=, 0);
anon_size = MAX((int64_t)(zfs_refcount_count(&arc_anon->arcs_size) -
arc_loaned_bytes), 0);
/*
* Writes will, almost always, require additional memory allocations
* in order to compress/encrypt/etc the data. We therefore need to
* make sure that there is sufficient available memory for this.
*/
error = arc_memory_throttle(spa, reserve, txg);
if (error != 0)
return (error);
/*
* Throttle writes when the amount of dirty data in the cache
* gets too large. We try to keep the cache less than half full
* of dirty blocks so that our sync times don't grow too large.
*
* In the case of one pool being built on another pool, we want
* to make sure we don't end up throttling the lower (backing)
* pool when the upper pool is the majority contributor to dirty
* data. To insure we make forward progress during throttling, we
* also check the current pool's net dirty data and only throttle
* if it exceeds zfs_arc_pool_dirty_percent of the anonymous dirty
* data in the cache.
*
* Note: if two requests come in concurrently, we might let them
* both succeed, when one of them should fail. Not a huge deal.
*/
uint64_t total_dirty = reserve + arc_tempreserve + anon_size;
uint64_t spa_dirty_anon = spa_dirty_data(spa);
uint64_t rarc_c = arc_warm ? arc_c : arc_c_max;
if (total_dirty > rarc_c * zfs_arc_dirty_limit_percent / 100 &&
anon_size > rarc_c * zfs_arc_anon_limit_percent / 100 &&
spa_dirty_anon > anon_size * zfs_arc_pool_dirty_percent / 100) {
#ifdef ZFS_DEBUG
uint64_t meta_esize = zfs_refcount_count(
&arc_anon->arcs_esize[ARC_BUFC_METADATA]);
uint64_t data_esize =
zfs_refcount_count(&arc_anon->arcs_esize[ARC_BUFC_DATA]);
dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK "
"anon_data=%lluK tempreserve=%lluK rarc_c=%lluK\n",
(u_longlong_t)arc_tempreserve >> 10,
(u_longlong_t)meta_esize >> 10,
(u_longlong_t)data_esize >> 10,
(u_longlong_t)reserve >> 10,
(u_longlong_t)rarc_c >> 10);
#endif
DMU_TX_STAT_BUMP(dmu_tx_dirty_throttle);
return (SET_ERROR(ERESTART));
}
atomic_add_64(&arc_tempreserve, reserve);
return (0);
}
static void
arc_kstat_update_state(arc_state_t *state, kstat_named_t *size,
kstat_named_t *evict_data, kstat_named_t *evict_metadata)
{
size->value.ui64 = zfs_refcount_count(&state->arcs_size);
evict_data->value.ui64 =
zfs_refcount_count(&state->arcs_esize[ARC_BUFC_DATA]);
evict_metadata->value.ui64 =
zfs_refcount_count(&state->arcs_esize[ARC_BUFC_METADATA]);
}
static int
arc_kstat_update(kstat_t *ksp, int rw)
{
arc_stats_t *as = ksp->ks_data;
if (rw == KSTAT_WRITE)
return (SET_ERROR(EACCES));
as->arcstat_hits.value.ui64 =
wmsum_value(&arc_sums.arcstat_hits);
as->arcstat_misses.value.ui64 =
wmsum_value(&arc_sums.arcstat_misses);
as->arcstat_demand_data_hits.value.ui64 =
wmsum_value(&arc_sums.arcstat_demand_data_hits);
as->arcstat_demand_data_misses.value.ui64 =
wmsum_value(&arc_sums.arcstat_demand_data_misses);
as->arcstat_demand_metadata_hits.value.ui64 =
wmsum_value(&arc_sums.arcstat_demand_metadata_hits);
as->arcstat_demand_metadata_misses.value.ui64 =
wmsum_value(&arc_sums.arcstat_demand_metadata_misses);
as->arcstat_prefetch_data_hits.value.ui64 =
wmsum_value(&arc_sums.arcstat_prefetch_data_hits);
as->arcstat_prefetch_data_misses.value.ui64 =
wmsum_value(&arc_sums.arcstat_prefetch_data_misses);
as->arcstat_prefetch_metadata_hits.value.ui64 =
wmsum_value(&arc_sums.arcstat_prefetch_metadata_hits);
as->arcstat_prefetch_metadata_misses.value.ui64 =
wmsum_value(&arc_sums.arcstat_prefetch_metadata_misses);
as->arcstat_mru_hits.value.ui64 =
wmsum_value(&arc_sums.arcstat_mru_hits);
as->arcstat_mru_ghost_hits.value.ui64 =
wmsum_value(&arc_sums.arcstat_mru_ghost_hits);
as->arcstat_mfu_hits.value.ui64 =
wmsum_value(&arc_sums.arcstat_mfu_hits);
as->arcstat_mfu_ghost_hits.value.ui64 =
wmsum_value(&arc_sums.arcstat_mfu_ghost_hits);
as->arcstat_deleted.value.ui64 =
wmsum_value(&arc_sums.arcstat_deleted);
as->arcstat_mutex_miss.value.ui64 =
wmsum_value(&arc_sums.arcstat_mutex_miss);
as->arcstat_access_skip.value.ui64 =
wmsum_value(&arc_sums.arcstat_access_skip);
as->arcstat_evict_skip.value.ui64 =
wmsum_value(&arc_sums.arcstat_evict_skip);
as->arcstat_evict_not_enough.value.ui64 =
wmsum_value(&arc_sums.arcstat_evict_not_enough);
as->arcstat_evict_l2_cached.value.ui64 =
wmsum_value(&arc_sums.arcstat_evict_l2_cached);
as->arcstat_evict_l2_eligible.value.ui64 =
wmsum_value(&arc_sums.arcstat_evict_l2_eligible);
as->arcstat_evict_l2_eligible_mfu.value.ui64 =
wmsum_value(&arc_sums.arcstat_evict_l2_eligible_mfu);
as->arcstat_evict_l2_eligible_mru.value.ui64 =
wmsum_value(&arc_sums.arcstat_evict_l2_eligible_mru);
as->arcstat_evict_l2_ineligible.value.ui64 =
wmsum_value(&arc_sums.arcstat_evict_l2_ineligible);
as->arcstat_evict_l2_skip.value.ui64 =
wmsum_value(&arc_sums.arcstat_evict_l2_skip);
as->arcstat_hash_collisions.value.ui64 =
wmsum_value(&arc_sums.arcstat_hash_collisions);
as->arcstat_hash_chains.value.ui64 =
wmsum_value(&arc_sums.arcstat_hash_chains);
as->arcstat_size.value.ui64 =
aggsum_value(&arc_sums.arcstat_size);
as->arcstat_compressed_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_compressed_size);
as->arcstat_uncompressed_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_uncompressed_size);
as->arcstat_overhead_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_overhead_size);
as->arcstat_hdr_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_hdr_size);
as->arcstat_data_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_data_size);
as->arcstat_metadata_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_metadata_size);
as->arcstat_dbuf_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_dbuf_size);
#if defined(COMPAT_FREEBSD11)
as->arcstat_other_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_bonus_size) +
aggsum_value(&arc_sums.arcstat_dnode_size) +
wmsum_value(&arc_sums.arcstat_dbuf_size);
#endif
arc_kstat_update_state(arc_anon,
&as->arcstat_anon_size,
&as->arcstat_anon_evictable_data,
&as->arcstat_anon_evictable_metadata);
arc_kstat_update_state(arc_mru,
&as->arcstat_mru_size,
&as->arcstat_mru_evictable_data,
&as->arcstat_mru_evictable_metadata);
arc_kstat_update_state(arc_mru_ghost,
&as->arcstat_mru_ghost_size,
&as->arcstat_mru_ghost_evictable_data,
&as->arcstat_mru_ghost_evictable_metadata);
arc_kstat_update_state(arc_mfu,
&as->arcstat_mfu_size,
&as->arcstat_mfu_evictable_data,
&as->arcstat_mfu_evictable_metadata);
arc_kstat_update_state(arc_mfu_ghost,
&as->arcstat_mfu_ghost_size,
&as->arcstat_mfu_ghost_evictable_data,
&as->arcstat_mfu_ghost_evictable_metadata);
as->arcstat_dnode_size.value.ui64 =
aggsum_value(&arc_sums.arcstat_dnode_size);
as->arcstat_bonus_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_bonus_size);
as->arcstat_l2_hits.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_hits);
as->arcstat_l2_misses.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_misses);
as->arcstat_l2_prefetch_asize.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_prefetch_asize);
as->arcstat_l2_mru_asize.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_mru_asize);
as->arcstat_l2_mfu_asize.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_mfu_asize);
as->arcstat_l2_bufc_data_asize.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_bufc_data_asize);
as->arcstat_l2_bufc_metadata_asize.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_bufc_metadata_asize);
as->arcstat_l2_feeds.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_feeds);
as->arcstat_l2_rw_clash.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rw_clash);
as->arcstat_l2_read_bytes.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_read_bytes);
as->arcstat_l2_write_bytes.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_write_bytes);
as->arcstat_l2_writes_sent.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_writes_sent);
as->arcstat_l2_writes_done.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_writes_done);
as->arcstat_l2_writes_error.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_writes_error);
as->arcstat_l2_writes_lock_retry.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_writes_lock_retry);
as->arcstat_l2_evict_lock_retry.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_evict_lock_retry);
as->arcstat_l2_evict_reading.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_evict_reading);
as->arcstat_l2_evict_l1cached.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_evict_l1cached);
as->arcstat_l2_free_on_write.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_free_on_write);
as->arcstat_l2_abort_lowmem.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_abort_lowmem);
as->arcstat_l2_cksum_bad.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_cksum_bad);
as->arcstat_l2_io_error.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_io_error);
as->arcstat_l2_lsize.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_lsize);
as->arcstat_l2_psize.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_psize);
as->arcstat_l2_hdr_size.value.ui64 =
aggsum_value(&arc_sums.arcstat_l2_hdr_size);
as->arcstat_l2_log_blk_writes.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_log_blk_writes);
as->arcstat_l2_log_blk_asize.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_log_blk_asize);
as->arcstat_l2_log_blk_count.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_log_blk_count);
as->arcstat_l2_rebuild_success.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rebuild_success);
as->arcstat_l2_rebuild_abort_unsupported.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rebuild_abort_unsupported);
as->arcstat_l2_rebuild_abort_io_errors.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rebuild_abort_io_errors);
as->arcstat_l2_rebuild_abort_dh_errors.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rebuild_abort_dh_errors);
as->arcstat_l2_rebuild_abort_cksum_lb_errors.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rebuild_abort_cksum_lb_errors);
as->arcstat_l2_rebuild_abort_lowmem.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rebuild_abort_lowmem);
as->arcstat_l2_rebuild_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rebuild_size);
as->arcstat_l2_rebuild_asize.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rebuild_asize);
as->arcstat_l2_rebuild_bufs.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rebuild_bufs);
as->arcstat_l2_rebuild_bufs_precached.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rebuild_bufs_precached);
as->arcstat_l2_rebuild_log_blks.value.ui64 =
wmsum_value(&arc_sums.arcstat_l2_rebuild_log_blks);
as->arcstat_memory_throttle_count.value.ui64 =
wmsum_value(&arc_sums.arcstat_memory_throttle_count);
as->arcstat_memory_direct_count.value.ui64 =
wmsum_value(&arc_sums.arcstat_memory_direct_count);
as->arcstat_memory_indirect_count.value.ui64 =
wmsum_value(&arc_sums.arcstat_memory_indirect_count);
as->arcstat_memory_all_bytes.value.ui64 =
arc_all_memory();
as->arcstat_memory_free_bytes.value.ui64 =
arc_free_memory();
as->arcstat_memory_available_bytes.value.i64 =
arc_available_memory();
as->arcstat_prune.value.ui64 =
wmsum_value(&arc_sums.arcstat_prune);
as->arcstat_meta_used.value.ui64 =
aggsum_value(&arc_sums.arcstat_meta_used);
as->arcstat_async_upgrade_sync.value.ui64 =
wmsum_value(&arc_sums.arcstat_async_upgrade_sync);
as->arcstat_demand_hit_predictive_prefetch.value.ui64 =
wmsum_value(&arc_sums.arcstat_demand_hit_predictive_prefetch);
as->arcstat_demand_hit_prescient_prefetch.value.ui64 =
wmsum_value(&arc_sums.arcstat_demand_hit_prescient_prefetch);
as->arcstat_raw_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_raw_size);
as->arcstat_cached_only_in_progress.value.ui64 =
wmsum_value(&arc_sums.arcstat_cached_only_in_progress);
as->arcstat_abd_chunk_waste_size.value.ui64 =
wmsum_value(&arc_sums.arcstat_abd_chunk_waste_size);
return (0);
}
/*
* This function *must* return indices evenly distributed between all
* sublists of the multilist. This is needed due to how the ARC eviction
* code is laid out; arc_evict_state() assumes ARC buffers are evenly
* distributed between all sublists and uses this assumption when
* deciding which sublist to evict from and how much to evict from it.
*/
static unsigned int
arc_state_multilist_index_func(multilist_t *ml, void *obj)
{
arc_buf_hdr_t *hdr = obj;
/*
* We rely on b_dva to generate evenly distributed index
* numbers using buf_hash below. So, as an added precaution,
* let's make sure we never add empty buffers to the arc lists.
*/
ASSERT(!HDR_EMPTY(hdr));
/*
* The assumption here, is the hash value for a given
* arc_buf_hdr_t will remain constant throughout its lifetime
* (i.e. its b_spa, b_dva, and b_birth fields don't change).
* Thus, we don't need to store the header's sublist index
* on insertion, as this index can be recalculated on removal.
*
* Also, the low order bits of the hash value are thought to be
* distributed evenly. Otherwise, in the case that the multilist
* has a power of two number of sublists, each sublists' usage
- * would not be evenly distributed.
+ * would not be evenly distributed. In this context full 64bit
+ * division would be a waste of time, so limit it to 32 bits.
*/
- return (buf_hash(hdr->b_spa, &hdr->b_dva, hdr->b_birth) %
+ return ((unsigned int)buf_hash(hdr->b_spa, &hdr->b_dva, hdr->b_birth) %
multilist_get_num_sublists(ml));
}
#define WARN_IF_TUNING_IGNORED(tuning, value, do_warn) do { \
if ((do_warn) && (tuning) && ((tuning) != (value))) { \
cmn_err(CE_WARN, \
"ignoring tunable %s (using %llu instead)", \
(#tuning), (value)); \
} \
} while (0)
/*
* Called during module initialization and periodically thereafter to
* apply reasonable changes to the exposed performance tunings. Can also be
* called explicitly by param_set_arc_*() functions when ARC tunables are
* updated manually. Non-zero zfs_* values which differ from the currently set
* values will be applied.
*/
void
arc_tuning_update(boolean_t verbose)
{
uint64_t allmem = arc_all_memory();
unsigned long limit;
/* Valid range: 32M - */
if ((zfs_arc_min) && (zfs_arc_min != arc_c_min) &&
(zfs_arc_min >= 2ULL << SPA_MAXBLOCKSHIFT) &&
(zfs_arc_min <= arc_c_max)) {
arc_c_min = zfs_arc_min;
arc_c = MAX(arc_c, arc_c_min);
}
WARN_IF_TUNING_IGNORED(zfs_arc_min, arc_c_min, verbose);
/* Valid range: 64M - */
if ((zfs_arc_max) && (zfs_arc_max != arc_c_max) &&
(zfs_arc_max >= 64 << 20) && (zfs_arc_max < allmem) &&
(zfs_arc_max > arc_c_min)) {
arc_c_max = zfs_arc_max;
arc_c = MIN(arc_c, arc_c_max);
arc_p = (arc_c >> 1);
if (arc_meta_limit > arc_c_max)
arc_meta_limit = arc_c_max;
if (arc_dnode_size_limit > arc_meta_limit)
arc_dnode_size_limit = arc_meta_limit;
}
WARN_IF_TUNING_IGNORED(zfs_arc_max, arc_c_max, verbose);
/* Valid range: 16M - */
if ((zfs_arc_meta_min) && (zfs_arc_meta_min != arc_meta_min) &&
(zfs_arc_meta_min >= 1ULL << SPA_MAXBLOCKSHIFT) &&
(zfs_arc_meta_min <= arc_c_max)) {
arc_meta_min = zfs_arc_meta_min;
if (arc_meta_limit < arc_meta_min)
arc_meta_limit = arc_meta_min;
if (arc_dnode_size_limit < arc_meta_min)
arc_dnode_size_limit = arc_meta_min;
}
WARN_IF_TUNING_IGNORED(zfs_arc_meta_min, arc_meta_min, verbose);
/* Valid range: - */
limit = zfs_arc_meta_limit ? zfs_arc_meta_limit :
MIN(zfs_arc_meta_limit_percent, 100) * arc_c_max / 100;
if ((limit != arc_meta_limit) &&
(limit >= arc_meta_min) &&
(limit <= arc_c_max))
arc_meta_limit = limit;
WARN_IF_TUNING_IGNORED(zfs_arc_meta_limit, arc_meta_limit, verbose);
/* Valid range: - */
limit = zfs_arc_dnode_limit ? zfs_arc_dnode_limit :
MIN(zfs_arc_dnode_limit_percent, 100) * arc_meta_limit / 100;
if ((limit != arc_dnode_size_limit) &&
(limit >= arc_meta_min) &&
(limit <= arc_meta_limit))
arc_dnode_size_limit = limit;
WARN_IF_TUNING_IGNORED(zfs_arc_dnode_limit, arc_dnode_size_limit,
verbose);
/* Valid range: 1 - N */
if (zfs_arc_grow_retry)
arc_grow_retry = zfs_arc_grow_retry;
/* Valid range: 1 - N */
if (zfs_arc_shrink_shift) {
arc_shrink_shift = zfs_arc_shrink_shift;
arc_no_grow_shift = MIN(arc_no_grow_shift, arc_shrink_shift -1);
}
/* Valid range: 1 - N */
if (zfs_arc_p_min_shift)
arc_p_min_shift = zfs_arc_p_min_shift;
/* Valid range: 1 - N ms */
if (zfs_arc_min_prefetch_ms)
arc_min_prefetch_ms = zfs_arc_min_prefetch_ms;
/* Valid range: 1 - N ms */
if (zfs_arc_min_prescient_prefetch_ms) {
arc_min_prescient_prefetch_ms =
zfs_arc_min_prescient_prefetch_ms;
}
/* Valid range: 0 - 100 */
if ((zfs_arc_lotsfree_percent >= 0) &&
(zfs_arc_lotsfree_percent <= 100))
arc_lotsfree_percent = zfs_arc_lotsfree_percent;
WARN_IF_TUNING_IGNORED(zfs_arc_lotsfree_percent, arc_lotsfree_percent,
verbose);
/* Valid range: 0 - */
if ((zfs_arc_sys_free) && (zfs_arc_sys_free != arc_sys_free))
arc_sys_free = MIN(MAX(zfs_arc_sys_free, 0), allmem);
WARN_IF_TUNING_IGNORED(zfs_arc_sys_free, arc_sys_free, verbose);
}
static void
arc_state_init(void)
{
arc_anon = &ARC_anon;
arc_mru = &ARC_mru;
arc_mru_ghost = &ARC_mru_ghost;
arc_mfu = &ARC_mfu;
arc_mfu_ghost = &ARC_mfu_ghost;
arc_l2c_only = &ARC_l2c_only;
multilist_create(&arc_mru->arcs_list[ARC_BUFC_METADATA],
sizeof (arc_buf_hdr_t),
offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
arc_state_multilist_index_func);
multilist_create(&arc_mru->arcs_list[ARC_BUFC_DATA],
sizeof (arc_buf_hdr_t),
offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
arc_state_multilist_index_func);
multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA],
sizeof (arc_buf_hdr_t),
offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
arc_state_multilist_index_func);
multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA],
sizeof (arc_buf_hdr_t),
offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
arc_state_multilist_index_func);
multilist_create(&arc_mfu->arcs_list[ARC_BUFC_METADATA],
sizeof (arc_buf_hdr_t),
offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
arc_state_multilist_index_func);
multilist_create(&arc_mfu->arcs_list[ARC_BUFC_DATA],
sizeof (arc_buf_hdr_t),
offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
arc_state_multilist_index_func);
multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA],
sizeof (arc_buf_hdr_t),
offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
arc_state_multilist_index_func);
multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA],
sizeof (arc_buf_hdr_t),
offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
arc_state_multilist_index_func);
multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA],
sizeof (arc_buf_hdr_t),
offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
arc_state_multilist_index_func);
multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_DATA],
sizeof (arc_buf_hdr_t),
offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node),
arc_state_multilist_index_func);
zfs_refcount_create(&arc_anon->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_create(&arc_anon->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_create(&arc_mru->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_create(&arc_mru->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_create(&arc_anon->arcs_size);
zfs_refcount_create(&arc_mru->arcs_size);
zfs_refcount_create(&arc_mru_ghost->arcs_size);
zfs_refcount_create(&arc_mfu->arcs_size);
zfs_refcount_create(&arc_mfu_ghost->arcs_size);
zfs_refcount_create(&arc_l2c_only->arcs_size);
wmsum_init(&arc_sums.arcstat_hits, 0);
wmsum_init(&arc_sums.arcstat_misses, 0);
wmsum_init(&arc_sums.arcstat_demand_data_hits, 0);
wmsum_init(&arc_sums.arcstat_demand_data_misses, 0);
wmsum_init(&arc_sums.arcstat_demand_metadata_hits, 0);
wmsum_init(&arc_sums.arcstat_demand_metadata_misses, 0);
wmsum_init(&arc_sums.arcstat_prefetch_data_hits, 0);
wmsum_init(&arc_sums.arcstat_prefetch_data_misses, 0);
wmsum_init(&arc_sums.arcstat_prefetch_metadata_hits, 0);
wmsum_init(&arc_sums.arcstat_prefetch_metadata_misses, 0);
wmsum_init(&arc_sums.arcstat_mru_hits, 0);
wmsum_init(&arc_sums.arcstat_mru_ghost_hits, 0);
wmsum_init(&arc_sums.arcstat_mfu_hits, 0);
wmsum_init(&arc_sums.arcstat_mfu_ghost_hits, 0);
wmsum_init(&arc_sums.arcstat_deleted, 0);
wmsum_init(&arc_sums.arcstat_mutex_miss, 0);
wmsum_init(&arc_sums.arcstat_access_skip, 0);
wmsum_init(&arc_sums.arcstat_evict_skip, 0);
wmsum_init(&arc_sums.arcstat_evict_not_enough, 0);
wmsum_init(&arc_sums.arcstat_evict_l2_cached, 0);
wmsum_init(&arc_sums.arcstat_evict_l2_eligible, 0);
wmsum_init(&arc_sums.arcstat_evict_l2_eligible_mfu, 0);
wmsum_init(&arc_sums.arcstat_evict_l2_eligible_mru, 0);
wmsum_init(&arc_sums.arcstat_evict_l2_ineligible, 0);
wmsum_init(&arc_sums.arcstat_evict_l2_skip, 0);
wmsum_init(&arc_sums.arcstat_hash_collisions, 0);
wmsum_init(&arc_sums.arcstat_hash_chains, 0);
aggsum_init(&arc_sums.arcstat_size, 0);
wmsum_init(&arc_sums.arcstat_compressed_size, 0);
wmsum_init(&arc_sums.arcstat_uncompressed_size, 0);
wmsum_init(&arc_sums.arcstat_overhead_size, 0);
wmsum_init(&arc_sums.arcstat_hdr_size, 0);
wmsum_init(&arc_sums.arcstat_data_size, 0);
wmsum_init(&arc_sums.arcstat_metadata_size, 0);
wmsum_init(&arc_sums.arcstat_dbuf_size, 0);
aggsum_init(&arc_sums.arcstat_dnode_size, 0);
wmsum_init(&arc_sums.arcstat_bonus_size, 0);
wmsum_init(&arc_sums.arcstat_l2_hits, 0);
wmsum_init(&arc_sums.arcstat_l2_misses, 0);
wmsum_init(&arc_sums.arcstat_l2_prefetch_asize, 0);
wmsum_init(&arc_sums.arcstat_l2_mru_asize, 0);
wmsum_init(&arc_sums.arcstat_l2_mfu_asize, 0);
wmsum_init(&arc_sums.arcstat_l2_bufc_data_asize, 0);
wmsum_init(&arc_sums.arcstat_l2_bufc_metadata_asize, 0);
wmsum_init(&arc_sums.arcstat_l2_feeds, 0);
wmsum_init(&arc_sums.arcstat_l2_rw_clash, 0);
wmsum_init(&arc_sums.arcstat_l2_read_bytes, 0);
wmsum_init(&arc_sums.arcstat_l2_write_bytes, 0);
wmsum_init(&arc_sums.arcstat_l2_writes_sent, 0);
wmsum_init(&arc_sums.arcstat_l2_writes_done, 0);
wmsum_init(&arc_sums.arcstat_l2_writes_error, 0);
wmsum_init(&arc_sums.arcstat_l2_writes_lock_retry, 0);
wmsum_init(&arc_sums.arcstat_l2_evict_lock_retry, 0);
wmsum_init(&arc_sums.arcstat_l2_evict_reading, 0);
wmsum_init(&arc_sums.arcstat_l2_evict_l1cached, 0);
wmsum_init(&arc_sums.arcstat_l2_free_on_write, 0);
wmsum_init(&arc_sums.arcstat_l2_abort_lowmem, 0);
wmsum_init(&arc_sums.arcstat_l2_cksum_bad, 0);
wmsum_init(&arc_sums.arcstat_l2_io_error, 0);
wmsum_init(&arc_sums.arcstat_l2_lsize, 0);
wmsum_init(&arc_sums.arcstat_l2_psize, 0);
aggsum_init(&arc_sums.arcstat_l2_hdr_size, 0);
wmsum_init(&arc_sums.arcstat_l2_log_blk_writes, 0);
wmsum_init(&arc_sums.arcstat_l2_log_blk_asize, 0);
wmsum_init(&arc_sums.arcstat_l2_log_blk_count, 0);
wmsum_init(&arc_sums.arcstat_l2_rebuild_success, 0);
wmsum_init(&arc_sums.arcstat_l2_rebuild_abort_unsupported, 0);
wmsum_init(&arc_sums.arcstat_l2_rebuild_abort_io_errors, 0);
wmsum_init(&arc_sums.arcstat_l2_rebuild_abort_dh_errors, 0);
wmsum_init(&arc_sums.arcstat_l2_rebuild_abort_cksum_lb_errors, 0);
wmsum_init(&arc_sums.arcstat_l2_rebuild_abort_lowmem, 0);
wmsum_init(&arc_sums.arcstat_l2_rebuild_size, 0);
wmsum_init(&arc_sums.arcstat_l2_rebuild_asize, 0);
wmsum_init(&arc_sums.arcstat_l2_rebuild_bufs, 0);
wmsum_init(&arc_sums.arcstat_l2_rebuild_bufs_precached, 0);
wmsum_init(&arc_sums.arcstat_l2_rebuild_log_blks, 0);
wmsum_init(&arc_sums.arcstat_memory_throttle_count, 0);
wmsum_init(&arc_sums.arcstat_memory_direct_count, 0);
wmsum_init(&arc_sums.arcstat_memory_indirect_count, 0);
wmsum_init(&arc_sums.arcstat_prune, 0);
aggsum_init(&arc_sums.arcstat_meta_used, 0);
wmsum_init(&arc_sums.arcstat_async_upgrade_sync, 0);
wmsum_init(&arc_sums.arcstat_demand_hit_predictive_prefetch, 0);
wmsum_init(&arc_sums.arcstat_demand_hit_prescient_prefetch, 0);
wmsum_init(&arc_sums.arcstat_raw_size, 0);
wmsum_init(&arc_sums.arcstat_cached_only_in_progress, 0);
wmsum_init(&arc_sums.arcstat_abd_chunk_waste_size, 0);
arc_anon->arcs_state = ARC_STATE_ANON;
arc_mru->arcs_state = ARC_STATE_MRU;
arc_mru_ghost->arcs_state = ARC_STATE_MRU_GHOST;
arc_mfu->arcs_state = ARC_STATE_MFU;
arc_mfu_ghost->arcs_state = ARC_STATE_MFU_GHOST;
arc_l2c_only->arcs_state = ARC_STATE_L2C_ONLY;
}
static void
arc_state_fini(void)
{
zfs_refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]);
zfs_refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]);
zfs_refcount_destroy(&arc_anon->arcs_size);
zfs_refcount_destroy(&arc_mru->arcs_size);
zfs_refcount_destroy(&arc_mru_ghost->arcs_size);
zfs_refcount_destroy(&arc_mfu->arcs_size);
zfs_refcount_destroy(&arc_mfu_ghost->arcs_size);
zfs_refcount_destroy(&arc_l2c_only->arcs_size);
multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_METADATA]);
multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]);
multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_METADATA]);
multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]);
multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_DATA]);
multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA]);
multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_DATA]);
multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]);
multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA]);
multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]);
wmsum_fini(&arc_sums.arcstat_hits);
wmsum_fini(&arc_sums.arcstat_misses);
wmsum_fini(&arc_sums.arcstat_demand_data_hits);
wmsum_fini(&arc_sums.arcstat_demand_data_misses);
wmsum_fini(&arc_sums.arcstat_demand_metadata_hits);
wmsum_fini(&arc_sums.arcstat_demand_metadata_misses);
wmsum_fini(&arc_sums.arcstat_prefetch_data_hits);
wmsum_fini(&arc_sums.arcstat_prefetch_data_misses);
wmsum_fini(&arc_sums.arcstat_prefetch_metadata_hits);
wmsum_fini(&arc_sums.arcstat_prefetch_metadata_misses);
wmsum_fini(&arc_sums.arcstat_mru_hits);
wmsum_fini(&arc_sums.arcstat_mru_ghost_hits);
wmsum_fini(&arc_sums.arcstat_mfu_hits);
wmsum_fini(&arc_sums.arcstat_mfu_ghost_hits);
wmsum_fini(&arc_sums.arcstat_deleted);
wmsum_fini(&arc_sums.arcstat_mutex_miss);
wmsum_fini(&arc_sums.arcstat_access_skip);
wmsum_fini(&arc_sums.arcstat_evict_skip);
wmsum_fini(&arc_sums.arcstat_evict_not_enough);
wmsum_fini(&arc_sums.arcstat_evict_l2_cached);
wmsum_fini(&arc_sums.arcstat_evict_l2_eligible);
wmsum_fini(&arc_sums.arcstat_evict_l2_eligible_mfu);
wmsum_fini(&arc_sums.arcstat_evict_l2_eligible_mru);
wmsum_fini(&arc_sums.arcstat_evict_l2_ineligible);
wmsum_fini(&arc_sums.arcstat_evict_l2_skip);
wmsum_fini(&arc_sums.arcstat_hash_collisions);
wmsum_fini(&arc_sums.arcstat_hash_chains);
aggsum_fini(&arc_sums.arcstat_size);
wmsum_fini(&arc_sums.arcstat_compressed_size);
wmsum_fini(&arc_sums.arcstat_uncompressed_size);
wmsum_fini(&arc_sums.arcstat_overhead_size);
wmsum_fini(&arc_sums.arcstat_hdr_size);
wmsum_fini(&arc_sums.arcstat_data_size);
wmsum_fini(&arc_sums.arcstat_metadata_size);
wmsum_fini(&arc_sums.arcstat_dbuf_size);
aggsum_fini(&arc_sums.arcstat_dnode_size);
wmsum_fini(&arc_sums.arcstat_bonus_size);
wmsum_fini(&arc_sums.arcstat_l2_hits);
wmsum_fini(&arc_sums.arcstat_l2_misses);
wmsum_fini(&arc_sums.arcstat_l2_prefetch_asize);
wmsum_fini(&arc_sums.arcstat_l2_mru_asize);
wmsum_fini(&arc_sums.arcstat_l2_mfu_asize);
wmsum_fini(&arc_sums.arcstat_l2_bufc_data_asize);
wmsum_fini(&arc_sums.arcstat_l2_bufc_metadata_asize);
wmsum_fini(&arc_sums.arcstat_l2_feeds);
wmsum_fini(&arc_sums.arcstat_l2_rw_clash);
wmsum_fini(&arc_sums.arcstat_l2_read_bytes);
wmsum_fini(&arc_sums.arcstat_l2_write_bytes);
wmsum_fini(&arc_sums.arcstat_l2_writes_sent);
wmsum_fini(&arc_sums.arcstat_l2_writes_done);
wmsum_fini(&arc_sums.arcstat_l2_writes_error);
wmsum_fini(&arc_sums.arcstat_l2_writes_lock_retry);
wmsum_fini(&arc_sums.arcstat_l2_evict_lock_retry);
wmsum_fini(&arc_sums.arcstat_l2_evict_reading);
wmsum_fini(&arc_sums.arcstat_l2_evict_l1cached);
wmsum_fini(&arc_sums.arcstat_l2_free_on_write);
wmsum_fini(&arc_sums.arcstat_l2_abort_lowmem);
wmsum_fini(&arc_sums.arcstat_l2_cksum_bad);
wmsum_fini(&arc_sums.arcstat_l2_io_error);
wmsum_fini(&arc_sums.arcstat_l2_lsize);
wmsum_fini(&arc_sums.arcstat_l2_psize);
aggsum_fini(&arc_sums.arcstat_l2_hdr_size);
wmsum_fini(&arc_sums.arcstat_l2_log_blk_writes);
wmsum_fini(&arc_sums.arcstat_l2_log_blk_asize);
wmsum_fini(&arc_sums.arcstat_l2_log_blk_count);
wmsum_fini(&arc_sums.arcstat_l2_rebuild_success);
wmsum_fini(&arc_sums.arcstat_l2_rebuild_abort_unsupported);
wmsum_fini(&arc_sums.arcstat_l2_rebuild_abort_io_errors);
wmsum_fini(&arc_sums.arcstat_l2_rebuild_abort_dh_errors);
wmsum_fini(&arc_sums.arcstat_l2_rebuild_abort_cksum_lb_errors);
wmsum_fini(&arc_sums.arcstat_l2_rebuild_abort_lowmem);
wmsum_fini(&arc_sums.arcstat_l2_rebuild_size);
wmsum_fini(&arc_sums.arcstat_l2_rebuild_asize);
wmsum_fini(&arc_sums.arcstat_l2_rebuild_bufs);
wmsum_fini(&arc_sums.arcstat_l2_rebuild_bufs_precached);
wmsum_fini(&arc_sums.arcstat_l2_rebuild_log_blks);
wmsum_fini(&arc_sums.arcstat_memory_throttle_count);
wmsum_fini(&arc_sums.arcstat_memory_direct_count);
wmsum_fini(&arc_sums.arcstat_memory_indirect_count);
wmsum_fini(&arc_sums.arcstat_prune);
aggsum_fini(&arc_sums.arcstat_meta_used);
wmsum_fini(&arc_sums.arcstat_async_upgrade_sync);
wmsum_fini(&arc_sums.arcstat_demand_hit_predictive_prefetch);
wmsum_fini(&arc_sums.arcstat_demand_hit_prescient_prefetch);
wmsum_fini(&arc_sums.arcstat_raw_size);
wmsum_fini(&arc_sums.arcstat_cached_only_in_progress);
wmsum_fini(&arc_sums.arcstat_abd_chunk_waste_size);
}
uint64_t
arc_target_bytes(void)
{
return (arc_c);
}
void
arc_set_limits(uint64_t allmem)
{
/* Set min cache to 1/32 of all memory, or 32MB, whichever is more. */
arc_c_min = MAX(allmem / 32, 2ULL << SPA_MAXBLOCKSHIFT);
/* How to set default max varies by platform. */
arc_c_max = arc_default_max(arc_c_min, allmem);
}
void
arc_init(void)
{
uint64_t percent, allmem = arc_all_memory();
mutex_init(&arc_evict_lock, NULL, MUTEX_DEFAULT, NULL);
list_create(&arc_evict_waiters, sizeof (arc_evict_waiter_t),
offsetof(arc_evict_waiter_t, aew_node));
arc_min_prefetch_ms = 1000;
arc_min_prescient_prefetch_ms = 6000;
#if defined(_KERNEL)
arc_lowmem_init();
#endif
arc_set_limits(allmem);
#ifndef _KERNEL
/*
* In userland, there's only the memory pressure that we artificially
* create (see arc_available_memory()). Don't let arc_c get too
* small, because it can cause transactions to be larger than
* arc_c, causing arc_tempreserve_space() to fail.
*/
arc_c_min = MAX(arc_c_max / 2, 2ULL << SPA_MAXBLOCKSHIFT);
#endif
arc_c = arc_c_min;
arc_p = (arc_c >> 1);
/* Set min to 1/2 of arc_c_min */
arc_meta_min = 1ULL << SPA_MAXBLOCKSHIFT;
/*
* Set arc_meta_limit to a percent of arc_c_max with a floor of
* arc_meta_min, and a ceiling of arc_c_max.
*/
percent = MIN(zfs_arc_meta_limit_percent, 100);
arc_meta_limit = MAX(arc_meta_min, (percent * arc_c_max) / 100);
percent = MIN(zfs_arc_dnode_limit_percent, 100);
arc_dnode_size_limit = (percent * arc_meta_limit) / 100;
/* Apply user specified tunings */
arc_tuning_update(B_TRUE);
/* if kmem_flags are set, lets try to use less memory */
if (kmem_debugging())
arc_c = arc_c / 2;
if (arc_c < arc_c_min)
arc_c = arc_c_min;
arc_register_hotplug();
arc_state_init();
buf_init();
list_create(&arc_prune_list, sizeof (arc_prune_t),
offsetof(arc_prune_t, p_node));
mutex_init(&arc_prune_mtx, NULL, MUTEX_DEFAULT, NULL);
arc_prune_taskq = taskq_create("arc_prune", 100, defclsyspri,
boot_ncpus, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC |
TASKQ_THREADS_CPU_PCT);
arc_ksp = kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED,
sizeof (arc_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL);
if (arc_ksp != NULL) {
arc_ksp->ks_data = &arc_stats;
arc_ksp->ks_update = arc_kstat_update;
kstat_install(arc_ksp);
}
arc_evict_zthr = zthr_create("arc_evict",
arc_evict_cb_check, arc_evict_cb, NULL);
arc_reap_zthr = zthr_create_timer("arc_reap",
arc_reap_cb_check, arc_reap_cb, NULL, SEC2NSEC(1));
arc_warm = B_FALSE;
/*
* Calculate maximum amount of dirty data per pool.
*
* If it has been set by a module parameter, take that.
* Otherwise, use a percentage of physical memory defined by
* zfs_dirty_data_max_percent (default 10%) with a cap at
* zfs_dirty_data_max_max (default 4G or 25% of physical memory).
*/
#ifdef __LP64__
if (zfs_dirty_data_max_max == 0)
zfs_dirty_data_max_max = MIN(4ULL * 1024 * 1024 * 1024,
allmem * zfs_dirty_data_max_max_percent / 100);
#else
if (zfs_dirty_data_max_max == 0)
zfs_dirty_data_max_max = MIN(1ULL * 1024 * 1024 * 1024,
allmem * zfs_dirty_data_max_max_percent / 100);
#endif
if (zfs_dirty_data_max == 0) {
zfs_dirty_data_max = allmem *
zfs_dirty_data_max_percent / 100;
zfs_dirty_data_max = MIN(zfs_dirty_data_max,
zfs_dirty_data_max_max);
}
}
void
arc_fini(void)
{
arc_prune_t *p;
#ifdef _KERNEL
arc_lowmem_fini();
#endif /* _KERNEL */
/* Use B_TRUE to ensure *all* buffers are evicted */
arc_flush(NULL, B_TRUE);
if (arc_ksp != NULL) {
kstat_delete(arc_ksp);
arc_ksp = NULL;
}
taskq_wait(arc_prune_taskq);
taskq_destroy(arc_prune_taskq);
mutex_enter(&arc_prune_mtx);
while ((p = list_head(&arc_prune_list)) != NULL) {
list_remove(&arc_prune_list, p);
zfs_refcount_remove(&p->p_refcnt, &arc_prune_list);
zfs_refcount_destroy(&p->p_refcnt);
kmem_free(p, sizeof (*p));
}
mutex_exit(&arc_prune_mtx);
list_destroy(&arc_prune_list);
mutex_destroy(&arc_prune_mtx);
(void) zthr_cancel(arc_evict_zthr);
(void) zthr_cancel(arc_reap_zthr);
mutex_destroy(&arc_evict_lock);
list_destroy(&arc_evict_waiters);
/*
* Free any buffers that were tagged for destruction. This needs
* to occur before arc_state_fini() runs and destroys the aggsum
* values which are updated when freeing scatter ABDs.
*/
l2arc_do_free_on_write();
/*
* buf_fini() must proceed arc_state_fini() because buf_fin() may
* trigger the release of kmem magazines, which can callback to
* arc_space_return() which accesses aggsums freed in act_state_fini().
*/
buf_fini();
arc_state_fini();
arc_unregister_hotplug();
/*
* We destroy the zthrs after all the ARC state has been
* torn down to avoid the case of them receiving any
* wakeup() signals after they are destroyed.
*/
zthr_destroy(arc_evict_zthr);
zthr_destroy(arc_reap_zthr);
ASSERT0(arc_loaned_bytes);
}
/*
* Level 2 ARC
*
* The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk.
* It uses dedicated storage devices to hold cached data, which are populated
* using large infrequent writes. The main role of this cache is to boost
* the performance of random read workloads. The intended L2ARC devices
* include short-stroked disks, solid state disks, and other media with
* substantially faster read latency than disk.
*
* +-----------------------+
* | ARC |
* +-----------------------+
* | ^ ^
* | | |
* l2arc_feed_thread() arc_read()
* | | |
* | l2arc read |
* V | |
* +---------------+ |
* | L2ARC | |
* +---------------+ |
* | ^ |
* l2arc_write() | |
* | | |
* V | |
* +-------+ +-------+
* | vdev | | vdev |
* | cache | | cache |
* +-------+ +-------+
* +=========+ .-----.
* : L2ARC : |-_____-|
* : devices : | Disks |
* +=========+ `-_____-'
*
* Read requests are satisfied from the following sources, in order:
*
* 1) ARC
* 2) vdev cache of L2ARC devices
* 3) L2ARC devices
* 4) vdev cache of disks
* 5) disks
*
* Some L2ARC device types exhibit extremely slow write performance.
* To accommodate for this there are some significant differences between
* the L2ARC and traditional cache design:
*
* 1. There is no eviction path from the ARC to the L2ARC. Evictions from
* the ARC behave as usual, freeing buffers and placing headers on ghost
* lists. The ARC does not send buffers to the L2ARC during eviction as
* this would add inflated write latencies for all ARC memory pressure.
*
* 2. The L2ARC attempts to cache data from the ARC before it is evicted.
* It does this by periodically scanning buffers from the eviction-end of
* the MFU and MRU ARC lists, copying them to the L2ARC devices if they are
* not already there. It scans until a headroom of buffers is satisfied,
* which itself is a buffer for ARC eviction. If a compressible buffer is
* found during scanning and selected for writing to an L2ARC device, we
* temporarily boost scanning headroom during the next scan cycle to make
* sure we adapt to compression effects (which might significantly reduce
* the data volume we write to L2ARC). The thread that does this is
* l2arc_feed_thread(), illustrated below; example sizes are included to
* provide a better sense of ratio than this diagram:
*
* head --> tail
* +---------------------+----------+
* ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC
* +---------------------+----------+ | o L2ARC eligible
* ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer
* +---------------------+----------+ |
* 15.9 Gbytes ^ 32 Mbytes |
* headroom |
* l2arc_feed_thread()
* |
* l2arc write hand <--[oooo]--'
* | 8 Mbyte
* | write max
* V
* +==============================+
* L2ARC dev |####|#|###|###| |####| ... |
* +==============================+
* 32 Gbytes
*
* 3. If an ARC buffer is copied to the L2ARC but then hit instead of
* evicted, then the L2ARC has cached a buffer much sooner than it probably
* needed to, potentially wasting L2ARC device bandwidth and storage. It is
* safe to say that this is an uncommon case, since buffers at the end of
* the ARC lists have moved there due to inactivity.
*
* 4. If the ARC evicts faster than the L2ARC can maintain a headroom,
* then the L2ARC simply misses copying some buffers. This serves as a
* pressure valve to prevent heavy read workloads from both stalling the ARC
* with waits and clogging the L2ARC with writes. This also helps prevent
* the potential for the L2ARC to churn if it attempts to cache content too
* quickly, such as during backups of the entire pool.
*
* 5. After system boot and before the ARC has filled main memory, there are
* no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru
* lists can remain mostly static. Instead of searching from tail of these
* lists as pictured, the l2arc_feed_thread() will search from the list heads
* for eligible buffers, greatly increasing its chance of finding them.
*
* The L2ARC device write speed is also boosted during this time so that
* the L2ARC warms up faster. Since there have been no ARC evictions yet,
* there are no L2ARC reads, and no fear of degrading read performance
* through increased writes.
*
* 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that
* the vdev queue can aggregate them into larger and fewer writes. Each
* device is written to in a rotor fashion, sweeping writes through
* available space then repeating.
*
* 7. The L2ARC does not store dirty content. It never needs to flush
* write buffers back to disk based storage.
*
* 8. If an ARC buffer is written (and dirtied) which also exists in the
* L2ARC, the now stale L2ARC buffer is immediately dropped.
*
* The performance of the L2ARC can be tweaked by a number of tunables, which
* may be necessary for different workloads:
*
* l2arc_write_max max write bytes per interval
* l2arc_write_boost extra write bytes during device warmup
* l2arc_noprefetch skip caching prefetched buffers
* l2arc_headroom number of max device writes to precache
* l2arc_headroom_boost when we find compressed buffers during ARC
* scanning, we multiply headroom by this
* percentage factor for the next scan cycle,
* since more compressed buffers are likely to
* be present
* l2arc_feed_secs seconds between L2ARC writing
*
* Tunables may be removed or added as future performance improvements are
* integrated, and also may become zpool properties.
*
* There are three key functions that control how the L2ARC warms up:
*
* l2arc_write_eligible() check if a buffer is eligible to cache
* l2arc_write_size() calculate how much to write
* l2arc_write_interval() calculate sleep delay between writes
*
* These three functions determine what to write, how much, and how quickly
* to send writes.
*
* L2ARC persistence:
*
* When writing buffers to L2ARC, we periodically add some metadata to
* make sure we can pick them up after reboot, thus dramatically reducing
* the impact that any downtime has on the performance of storage systems
* with large caches.
*
* The implementation works fairly simply by integrating the following two
* modifications:
*
* *) When writing to the L2ARC, we occasionally write a "l2arc log block",
* which is an additional piece of metadata which describes what's been
* written. This allows us to rebuild the arc_buf_hdr_t structures of the
* main ARC buffers. There are 2 linked-lists of log blocks headed by
* dh_start_lbps[2]. We alternate which chain we append to, so they are
* time-wise and offset-wise interleaved, but that is an optimization rather
* than for correctness. The log block also includes a pointer to the
* previous block in its chain.
*
* *) We reserve SPA_MINBLOCKSIZE of space at the start of each L2ARC device
* for our header bookkeeping purposes. This contains a device header,
* which contains our top-level reference structures. We update it each
* time we write a new log block, so that we're able to locate it in the
* L2ARC device. If this write results in an inconsistent device header
* (e.g. due to power failure), we detect this by verifying the header's
* checksum and simply fail to reconstruct the L2ARC after reboot.
*
* Implementation diagram:
*
* +=== L2ARC device (not to scale) ======================================+
* | ___two newest log block pointers__.__________ |
* | / \dh_start_lbps[1] |
* | / \ \dh_start_lbps[0]|
* |.___/__. V V |
* ||L2 dev|....|lb |bufs |lb |bufs |lb |bufs |lb |bufs |lb |---(empty)---|
* || hdr| ^ /^ /^ / / |
* |+------+ ...--\-------/ \-----/--\------/ / |
* | \--------------/ \--------------/ |
* +======================================================================+
*
* As can be seen on the diagram, rather than using a simple linked list,
* we use a pair of linked lists with alternating elements. This is a
* performance enhancement due to the fact that we only find out the
* address of the next log block access once the current block has been
* completely read in. Obviously, this hurts performance, because we'd be
* keeping the device's I/O queue at only a 1 operation deep, thus
* incurring a large amount of I/O round-trip latency. Having two lists
* allows us to fetch two log blocks ahead of where we are currently
* rebuilding L2ARC buffers.
*
* On-device data structures:
*
* L2ARC device header: l2arc_dev_hdr_phys_t
* L2ARC log block: l2arc_log_blk_phys_t
*
* L2ARC reconstruction:
*
* When writing data, we simply write in the standard rotary fashion,
* evicting buffers as we go and simply writing new data over them (writing
* a new log block every now and then). This obviously means that once we
* loop around the end of the device, we will start cutting into an already
* committed log block (and its referenced data buffers), like so:
*
* current write head__ __old tail
* \ /
* V V
* <--|bufs |lb |bufs |lb | |bufs |lb |bufs |lb |-->
* ^ ^^^^^^^^^___________________________________
* | \
* <> may overwrite this blk and/or its bufs --'
*
* When importing the pool, we detect this situation and use it to stop
* our scanning process (see l2arc_rebuild).
*
* There is one significant caveat to consider when rebuilding ARC contents
* from an L2ARC device: what about invalidated buffers? Given the above
* construction, we cannot update blocks which we've already written to amend
* them to remove buffers which were invalidated. Thus, during reconstruction,
* we might be populating the cache with buffers for data that's not on the
* main pool anymore, or may have been overwritten!
*
* As it turns out, this isn't a problem. Every arc_read request includes
* both the DVA and, crucially, the birth TXG of the BP the caller is
* looking for. So even if the cache were populated by completely rotten
* blocks for data that had been long deleted and/or overwritten, we'll
* never actually return bad data from the cache, since the DVA with the
* birth TXG uniquely identify a block in space and time - once created,
* a block is immutable on disk. The worst thing we have done is wasted
* some time and memory at l2arc rebuild to reconstruct outdated ARC
* entries that will get dropped from the l2arc as it is being updated
* with new blocks.
*
* L2ARC buffers that have been evicted by l2arc_evict() ahead of the write
* hand are not restored. This is done by saving the offset (in bytes)
* l2arc_evict() has evicted to in the L2ARC device header and taking it
* into account when restoring buffers.
*/
static boolean_t
l2arc_write_eligible(uint64_t spa_guid, arc_buf_hdr_t *hdr)
{
/*
* A buffer is *not* eligible for the L2ARC if it:
* 1. belongs to a different spa.
* 2. is already cached on the L2ARC.
* 3. has an I/O in progress (it may be an incomplete read).
* 4. is flagged not eligible (zfs property).
*/
if (hdr->b_spa != spa_guid || HDR_HAS_L2HDR(hdr) ||
HDR_IO_IN_PROGRESS(hdr) || !HDR_L2CACHE(hdr))
return (B_FALSE);
return (B_TRUE);
}
static uint64_t
l2arc_write_size(l2arc_dev_t *dev)
{
uint64_t size, dev_size, tsize;
/*
* Make sure our globals have meaningful values in case the user
* altered them.
*/
size = l2arc_write_max;
if (size == 0) {
cmn_err(CE_NOTE, "Bad value for l2arc_write_max, value must "
"be greater than zero, resetting it to the default (%d)",
L2ARC_WRITE_SIZE);
size = l2arc_write_max = L2ARC_WRITE_SIZE;
}
if (arc_warm == B_FALSE)
size += l2arc_write_boost;
/*
* Make sure the write size does not exceed the size of the cache
* device. This is important in l2arc_evict(), otherwise infinite
* iteration can occur.
*/
dev_size = dev->l2ad_end - dev->l2ad_start;
tsize = size + l2arc_log_blk_overhead(size, dev);
if (dev->l2ad_vdev->vdev_has_trim && l2arc_trim_ahead > 0)
tsize += MAX(64 * 1024 * 1024,
(tsize * l2arc_trim_ahead) / 100);
if (tsize >= dev_size) {
cmn_err(CE_NOTE, "l2arc_write_max or l2arc_write_boost "
"plus the overhead of log blocks (persistent L2ARC, "
"%llu bytes) exceeds the size of the cache device "
"(guid %llu), resetting them to the default (%d)",
l2arc_log_blk_overhead(size, dev),
dev->l2ad_vdev->vdev_guid, L2ARC_WRITE_SIZE);
size = l2arc_write_max = l2arc_write_boost = L2ARC_WRITE_SIZE;
if (arc_warm == B_FALSE)
size += l2arc_write_boost;
}
return (size);
}
static clock_t
l2arc_write_interval(clock_t began, uint64_t wanted, uint64_t wrote)
{
clock_t interval, next, now;
/*
* If the ARC lists are busy, increase our write rate; if the
* lists are stale, idle back. This is achieved by checking
* how much we previously wrote - if it was more than half of
* what we wanted, schedule the next write much sooner.
*/
if (l2arc_feed_again && wrote > (wanted / 2))
interval = (hz * l2arc_feed_min_ms) / 1000;
else
interval = hz * l2arc_feed_secs;
now = ddi_get_lbolt();
next = MAX(now, MIN(now + interval, began + interval));
return (next);
}
/*
* Cycle through L2ARC devices. This is how L2ARC load balances.
* If a device is returned, this also returns holding the spa config lock.
*/
static l2arc_dev_t *
l2arc_dev_get_next(void)
{
l2arc_dev_t *first, *next = NULL;
/*
* Lock out the removal of spas (spa_namespace_lock), then removal
* of cache devices (l2arc_dev_mtx). Once a device has been selected,
* both locks will be dropped and a spa config lock held instead.
*/
mutex_enter(&spa_namespace_lock);
mutex_enter(&l2arc_dev_mtx);
/* if there are no vdevs, there is nothing to do */
if (l2arc_ndev == 0)
goto out;
first = NULL;
next = l2arc_dev_last;
do {
/* loop around the list looking for a non-faulted vdev */
if (next == NULL) {
next = list_head(l2arc_dev_list);
} else {
next = list_next(l2arc_dev_list, next);
if (next == NULL)
next = list_head(l2arc_dev_list);
}
/* if we have come back to the start, bail out */
if (first == NULL)
first = next;
else if (next == first)
break;
} while (vdev_is_dead(next->l2ad_vdev) || next->l2ad_rebuild ||
next->l2ad_trim_all);
/* if we were unable to find any usable vdevs, return NULL */
if (vdev_is_dead(next->l2ad_vdev) || next->l2ad_rebuild ||
next->l2ad_trim_all)
next = NULL;
l2arc_dev_last = next;
out:
mutex_exit(&l2arc_dev_mtx);
/*
* Grab the config lock to prevent the 'next' device from being
* removed while we are writing to it.
*/
if (next != NULL)
spa_config_enter(next->l2ad_spa, SCL_L2ARC, next, RW_READER);
mutex_exit(&spa_namespace_lock);
return (next);
}
/*
* Free buffers that were tagged for destruction.
*/
static void
l2arc_do_free_on_write(void)
{
list_t *buflist;
l2arc_data_free_t *df, *df_prev;
mutex_enter(&l2arc_free_on_write_mtx);
buflist = l2arc_free_on_write;
for (df = list_tail(buflist); df; df = df_prev) {
df_prev = list_prev(buflist, df);
ASSERT3P(df->l2df_abd, !=, NULL);
abd_free(df->l2df_abd);
list_remove(buflist, df);
kmem_free(df, sizeof (l2arc_data_free_t));
}
mutex_exit(&l2arc_free_on_write_mtx);
}
/*
* A write to a cache device has completed. Update all headers to allow
* reads from these buffers to begin.
*/
static void
l2arc_write_done(zio_t *zio)
{
l2arc_write_callback_t *cb;
l2arc_lb_abd_buf_t *abd_buf;
l2arc_lb_ptr_buf_t *lb_ptr_buf;
l2arc_dev_t *dev;
l2arc_dev_hdr_phys_t *l2dhdr;
list_t *buflist;
arc_buf_hdr_t *head, *hdr, *hdr_prev;
kmutex_t *hash_lock;
int64_t bytes_dropped = 0;
cb = zio->io_private;
ASSERT3P(cb, !=, NULL);
dev = cb->l2wcb_dev;
l2dhdr = dev->l2ad_dev_hdr;
ASSERT3P(dev, !=, NULL);
head = cb->l2wcb_head;
ASSERT3P(head, !=, NULL);
buflist = &dev->l2ad_buflist;
ASSERT3P(buflist, !=, NULL);
DTRACE_PROBE2(l2arc__iodone, zio_t *, zio,
l2arc_write_callback_t *, cb);
/*
* All writes completed, or an error was hit.
*/
top:
mutex_enter(&dev->l2ad_mtx);
for (hdr = list_prev(buflist, head); hdr; hdr = hdr_prev) {
hdr_prev = list_prev(buflist, hdr);
hash_lock = HDR_LOCK(hdr);
/*
* We cannot use mutex_enter or else we can deadlock
* with l2arc_write_buffers (due to swapping the order
* the hash lock and l2ad_mtx are taken).
*/
if (!mutex_tryenter(hash_lock)) {
/*
* Missed the hash lock. We must retry so we
* don't leave the ARC_FLAG_L2_WRITING bit set.
*/
ARCSTAT_BUMP(arcstat_l2_writes_lock_retry);
/*
* We don't want to rescan the headers we've
* already marked as having been written out, so
* we reinsert the head node so we can pick up
* where we left off.
*/
list_remove(buflist, head);
list_insert_after(buflist, hdr, head);
mutex_exit(&dev->l2ad_mtx);
/*
* We wait for the hash lock to become available
* to try and prevent busy waiting, and increase
* the chance we'll be able to acquire the lock
* the next time around.
*/
mutex_enter(hash_lock);
mutex_exit(hash_lock);
goto top;
}
/*
* We could not have been moved into the arc_l2c_only
* state while in-flight due to our ARC_FLAG_L2_WRITING
* bit being set. Let's just ensure that's being enforced.
*/
ASSERT(HDR_HAS_L1HDR(hdr));
/*
* Skipped - drop L2ARC entry and mark the header as no
* longer L2 eligibile.
*/
if (zio->io_error != 0) {
/*
* Error - drop L2ARC entry.
*/
list_remove(buflist, hdr);
arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR);
uint64_t psize = HDR_GET_PSIZE(hdr);
l2arc_hdr_arcstats_decrement(hdr);
bytes_dropped +=
vdev_psize_to_asize(dev->l2ad_vdev, psize);
(void) zfs_refcount_remove_many(&dev->l2ad_alloc,
arc_hdr_size(hdr), hdr);
}
/*
* Allow ARC to begin reads and ghost list evictions to
* this L2ARC entry.
*/
arc_hdr_clear_flags(hdr, ARC_FLAG_L2_WRITING);
mutex_exit(hash_lock);
}
/*
* Free the allocated abd buffers for writing the log blocks.
* If the zio failed reclaim the allocated space and remove the
* pointers to these log blocks from the log block pointer list
* of the L2ARC device.
*/
while ((abd_buf = list_remove_tail(&cb->l2wcb_abd_list)) != NULL) {
abd_free(abd_buf->abd);
zio_buf_free(abd_buf, sizeof (*abd_buf));
if (zio->io_error != 0) {
lb_ptr_buf = list_remove_head(&dev->l2ad_lbptr_list);
/*
* L2BLK_GET_PSIZE returns aligned size for log
* blocks.
*/
uint64_t asize =
L2BLK_GET_PSIZE((lb_ptr_buf->lb_ptr)->lbp_prop);
bytes_dropped += asize;
ARCSTAT_INCR(arcstat_l2_log_blk_asize, -asize);
ARCSTAT_BUMPDOWN(arcstat_l2_log_blk_count);
zfs_refcount_remove_many(&dev->l2ad_lb_asize, asize,
lb_ptr_buf);
zfs_refcount_remove(&dev->l2ad_lb_count, lb_ptr_buf);
kmem_free(lb_ptr_buf->lb_ptr,
sizeof (l2arc_log_blkptr_t));
kmem_free(lb_ptr_buf, sizeof (l2arc_lb_ptr_buf_t));
}
}
list_destroy(&cb->l2wcb_abd_list);
if (zio->io_error != 0) {
ARCSTAT_BUMP(arcstat_l2_writes_error);
/*
* Restore the lbps array in the header to its previous state.
* If the list of log block pointers is empty, zero out the
* log block pointers in the device header.
*/
lb_ptr_buf = list_head(&dev->l2ad_lbptr_list);
for (int i = 0; i < 2; i++) {
if (lb_ptr_buf == NULL) {
/*
* If the list is empty zero out the device
* header. Otherwise zero out the second log
* block pointer in the header.
*/
if (i == 0) {
bzero(l2dhdr, dev->l2ad_dev_hdr_asize);
} else {
bzero(&l2dhdr->dh_start_lbps[i],
sizeof (l2arc_log_blkptr_t));
}
break;
}
bcopy(lb_ptr_buf->lb_ptr, &l2dhdr->dh_start_lbps[i],
sizeof (l2arc_log_blkptr_t));
lb_ptr_buf = list_next(&dev->l2ad_lbptr_list,
lb_ptr_buf);
}
}
ARCSTAT_BUMP(arcstat_l2_writes_done);
list_remove(buflist, head);
ASSERT(!HDR_HAS_L1HDR(head));
kmem_cache_free(hdr_l2only_cache, head);
mutex_exit(&dev->l2ad_mtx);
ASSERT(dev->l2ad_vdev != NULL);
vdev_space_update(dev->l2ad_vdev, -bytes_dropped, 0, 0);
l2arc_do_free_on_write();
kmem_free(cb, sizeof (l2arc_write_callback_t));
}
static int
l2arc_untransform(zio_t *zio, l2arc_read_callback_t *cb)
{
int ret;
spa_t *spa = zio->io_spa;
arc_buf_hdr_t *hdr = cb->l2rcb_hdr;
blkptr_t *bp = zio->io_bp;
uint8_t salt[ZIO_DATA_SALT_LEN];
uint8_t iv[ZIO_DATA_IV_LEN];
uint8_t mac[ZIO_DATA_MAC_LEN];
boolean_t no_crypt = B_FALSE;
/*
* ZIL data is never be written to the L2ARC, so we don't need
* special handling for its unique MAC storage.
*/
ASSERT3U(BP_GET_TYPE(bp), !=, DMU_OT_INTENT_LOG);
ASSERT(MUTEX_HELD(HDR_LOCK(hdr)));
ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
/*
* If the data was encrypted, decrypt it now. Note that
* we must check the bp here and not the hdr, since the
* hdr does not have its encryption parameters updated
* until arc_read_done().
*/
if (BP_IS_ENCRYPTED(bp)) {
abd_t *eabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr,
B_TRUE);
zio_crypt_decode_params_bp(bp, salt, iv);
zio_crypt_decode_mac_bp(bp, mac);
ret = spa_do_crypt_abd(B_FALSE, spa, &cb->l2rcb_zb,
BP_GET_TYPE(bp), BP_GET_DEDUP(bp), BP_SHOULD_BYTESWAP(bp),
salt, iv, mac, HDR_GET_PSIZE(hdr), eabd,
hdr->b_l1hdr.b_pabd, &no_crypt);
if (ret != 0) {
arc_free_data_abd(hdr, eabd, arc_hdr_size(hdr), hdr);
goto error;
}
/*
* If we actually performed decryption, replace b_pabd
* with the decrypted data. Otherwise we can just throw
* our decryption buffer away.
*/
if (!no_crypt) {
arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd,
arc_hdr_size(hdr), hdr);
hdr->b_l1hdr.b_pabd = eabd;
zio->io_abd = eabd;
} else {
arc_free_data_abd(hdr, eabd, arc_hdr_size(hdr), hdr);
}
}
/*
* If the L2ARC block was compressed, but ARC compression
* is disabled we decompress the data into a new buffer and
* replace the existing data.
*/
if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF &&
!HDR_COMPRESSION_ENABLED(hdr)) {
abd_t *cabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr,
B_TRUE);
void *tmp = abd_borrow_buf(cabd, arc_hdr_size(hdr));
ret = zio_decompress_data(HDR_GET_COMPRESS(hdr),
hdr->b_l1hdr.b_pabd, tmp, HDR_GET_PSIZE(hdr),
HDR_GET_LSIZE(hdr), &hdr->b_complevel);
if (ret != 0) {
abd_return_buf_copy(cabd, tmp, arc_hdr_size(hdr));
arc_free_data_abd(hdr, cabd, arc_hdr_size(hdr), hdr);
goto error;
}
abd_return_buf_copy(cabd, tmp, arc_hdr_size(hdr));
arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd,
arc_hdr_size(hdr), hdr);
hdr->b_l1hdr.b_pabd = cabd;
zio->io_abd = cabd;
zio->io_size = HDR_GET_LSIZE(hdr);
}
return (0);
error:
return (ret);
}
/*
* A read to a cache device completed. Validate buffer contents before
* handing over to the regular ARC routines.
*/
static void
l2arc_read_done(zio_t *zio)
{
int tfm_error = 0;
l2arc_read_callback_t *cb = zio->io_private;
arc_buf_hdr_t *hdr;
kmutex_t *hash_lock;
boolean_t valid_cksum;
boolean_t using_rdata = (BP_IS_ENCRYPTED(&cb->l2rcb_bp) &&
(cb->l2rcb_flags & ZIO_FLAG_RAW_ENCRYPT));
ASSERT3P(zio->io_vd, !=, NULL);
ASSERT(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE);
spa_config_exit(zio->io_spa, SCL_L2ARC, zio->io_vd);
ASSERT3P(cb, !=, NULL);
hdr = cb->l2rcb_hdr;
ASSERT3P(hdr, !=, NULL);
hash_lock = HDR_LOCK(hdr);
mutex_enter(hash_lock);
ASSERT3P(hash_lock, ==, HDR_LOCK(hdr));
/*
* If the data was read into a temporary buffer,
* move it and free the buffer.
*/
if (cb->l2rcb_abd != NULL) {
ASSERT3U(arc_hdr_size(hdr), <, zio->io_size);
if (zio->io_error == 0) {
if (using_rdata) {
abd_copy(hdr->b_crypt_hdr.b_rabd,
cb->l2rcb_abd, arc_hdr_size(hdr));
} else {
abd_copy(hdr->b_l1hdr.b_pabd,
cb->l2rcb_abd, arc_hdr_size(hdr));
}
}
/*
* The following must be done regardless of whether
* there was an error:
* - free the temporary buffer
* - point zio to the real ARC buffer
* - set zio size accordingly
* These are required because zio is either re-used for
* an I/O of the block in the case of the error
* or the zio is passed to arc_read_done() and it
* needs real data.
*/
abd_free(cb->l2rcb_abd);
zio->io_size = zio->io_orig_size = arc_hdr_size(hdr);
if (using_rdata) {
ASSERT(HDR_HAS_RABD(hdr));
zio->io_abd = zio->io_orig_abd =
hdr->b_crypt_hdr.b_rabd;
} else {
ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL);
zio->io_abd = zio->io_orig_abd = hdr->b_l1hdr.b_pabd;
}
}
ASSERT3P(zio->io_abd, !=, NULL);
/*
* Check this survived the L2ARC journey.
*/
ASSERT(zio->io_abd == hdr->b_l1hdr.b_pabd ||
(HDR_HAS_RABD(hdr) && zio->io_abd == hdr->b_crypt_hdr.b_rabd));
zio->io_bp_copy = cb->l2rcb_bp; /* XXX fix in L2ARC 2.0 */
zio->io_bp = &zio->io_bp_copy; /* XXX fix in L2ARC 2.0 */
zio->io_prop.zp_complevel = hdr->b_complevel;
valid_cksum = arc_cksum_is_equal(hdr, zio);
/*
* b_rabd will always match the data as it exists on disk if it is
* being used. Therefore if we are reading into b_rabd we do not
* attempt to untransform the data.
*/
if (valid_cksum && !using_rdata)
tfm_error = l2arc_untransform(zio, cb);
if (valid_cksum && tfm_error == 0 && zio->io_error == 0 &&
!HDR_L2_EVICTED(hdr)) {
mutex_exit(hash_lock);
zio->io_private = hdr;
arc_read_done(zio);
} else {
/*
* Buffer didn't survive caching. Increment stats and
* reissue to the original storage device.
*/
if (zio->io_error != 0) {
ARCSTAT_BUMP(arcstat_l2_io_error);
} else {
zio->io_error = SET_ERROR(EIO);
}
if (!valid_cksum || tfm_error != 0)
ARCSTAT_BUMP(arcstat_l2_cksum_bad);
/*
* If there's no waiter, issue an async i/o to the primary
* storage now. If there *is* a waiter, the caller must
* issue the i/o in a context where it's OK to block.
*/
if (zio->io_waiter == NULL) {
zio_t *pio = zio_unique_parent(zio);
void *abd = (using_rdata) ?
hdr->b_crypt_hdr.b_rabd : hdr->b_l1hdr.b_pabd;
ASSERT(!pio || pio->io_child_type == ZIO_CHILD_LOGICAL);
zio = zio_read(pio, zio->io_spa, zio->io_bp,
abd, zio->io_size, arc_read_done,
hdr, zio->io_priority, cb->l2rcb_flags,
&cb->l2rcb_zb);
/*
* Original ZIO will be freed, so we need to update
* ARC header with the new ZIO pointer to be used
* by zio_change_priority() in arc_read().
*/
for (struct arc_callback *acb = hdr->b_l1hdr.b_acb;
acb != NULL; acb = acb->acb_next)
acb->acb_zio_head = zio;
mutex_exit(hash_lock);
zio_nowait(zio);
} else {
mutex_exit(hash_lock);
}
}
kmem_free(cb, sizeof (l2arc_read_callback_t));
}
/*
* This is the list priority from which the L2ARC will search for pages to
* cache. This is used within loops (0..3) to cycle through lists in the
* desired order. This order can have a significant effect on cache
* performance.
*
* Currently the metadata lists are hit first, MFU then MRU, followed by
* the data lists. This function returns a locked list, and also returns
* the lock pointer.
*/
static multilist_sublist_t *
l2arc_sublist_lock(int list_num)
{
multilist_t *ml = NULL;
unsigned int idx;
ASSERT(list_num >= 0 && list_num < L2ARC_FEED_TYPES);
switch (list_num) {
case 0:
ml = &arc_mfu->arcs_list[ARC_BUFC_METADATA];
break;
case 1:
ml = &arc_mru->arcs_list[ARC_BUFC_METADATA];
break;
case 2:
ml = &arc_mfu->arcs_list[ARC_BUFC_DATA];
break;
case 3:
ml = &arc_mru->arcs_list[ARC_BUFC_DATA];
break;
default:
return (NULL);
}
/*
* Return a randomly-selected sublist. This is acceptable
* because the caller feeds only a little bit of data for each
* call (8MB). Subsequent calls will result in different
* sublists being selected.
*/
idx = multilist_get_random_index(ml);
return (multilist_sublist_lock(ml, idx));
}
/*
* Calculates the maximum overhead of L2ARC metadata log blocks for a given
* L2ARC write size. l2arc_evict and l2arc_write_size need to include this
* overhead in processing to make sure there is enough headroom available
* when writing buffers.
*/
static inline uint64_t
l2arc_log_blk_overhead(uint64_t write_sz, l2arc_dev_t *dev)
{
if (dev->l2ad_log_entries == 0) {
return (0);
} else {
uint64_t log_entries = write_sz >> SPA_MINBLOCKSHIFT;
uint64_t log_blocks = (log_entries +
dev->l2ad_log_entries - 1) /
dev->l2ad_log_entries;
return (vdev_psize_to_asize(dev->l2ad_vdev,
sizeof (l2arc_log_blk_phys_t)) * log_blocks);
}
}
/*
* Evict buffers from the device write hand to the distance specified in
* bytes. This distance may span populated buffers, it may span nothing.
* This is clearing a region on the L2ARC device ready for writing.
* If the 'all' boolean is set, every buffer is evicted.
*/
static void
l2arc_evict(l2arc_dev_t *dev, uint64_t distance, boolean_t all)
{
list_t *buflist;
arc_buf_hdr_t *hdr, *hdr_prev;
kmutex_t *hash_lock;
uint64_t taddr;
l2arc_lb_ptr_buf_t *lb_ptr_buf, *lb_ptr_buf_prev;
vdev_t *vd = dev->l2ad_vdev;
boolean_t rerun;
buflist = &dev->l2ad_buflist;
/*
* We need to add in the worst case scenario of log block overhead.
*/
distance += l2arc_log_blk_overhead(distance, dev);
if (vd->vdev_has_trim && l2arc_trim_ahead > 0) {
/*
* Trim ahead of the write size 64MB or (l2arc_trim_ahead/100)
* times the write size, whichever is greater.
*/
distance += MAX(64 * 1024 * 1024,
(distance * l2arc_trim_ahead) / 100);
}
top:
rerun = B_FALSE;
if (dev->l2ad_hand >= (dev->l2ad_end - distance)) {
/*
* When there is no space to accommodate upcoming writes,
* evict to the end. Then bump the write and evict hands
* to the start and iterate. This iteration does not
* happen indefinitely as we make sure in
* l2arc_write_size() that when the write hand is reset,
* the write size does not exceed the end of the device.
*/
rerun = B_TRUE;
taddr = dev->l2ad_end;
} else {
taddr = dev->l2ad_hand + distance;
}
DTRACE_PROBE4(l2arc__evict, l2arc_dev_t *, dev, list_t *, buflist,
uint64_t, taddr, boolean_t, all);
if (!all) {
/*
* This check has to be placed after deciding whether to
* iterate (rerun).
*/
if (dev->l2ad_first) {
/*
* This is the first sweep through the device. There is
* nothing to evict. We have already trimmmed the
* whole device.
*/
goto out;
} else {
/*
* Trim the space to be evicted.
*/
if (vd->vdev_has_trim && dev->l2ad_evict < taddr &&
l2arc_trim_ahead > 0) {
/*
* We have to drop the spa_config lock because
* vdev_trim_range() will acquire it.
* l2ad_evict already accounts for the label
* size. To prevent vdev_trim_ranges() from
* adding it again, we subtract it from
* l2ad_evict.
*/
spa_config_exit(dev->l2ad_spa, SCL_L2ARC, dev);
vdev_trim_simple(vd,
dev->l2ad_evict - VDEV_LABEL_START_SIZE,
taddr - dev->l2ad_evict);
spa_config_enter(dev->l2ad_spa, SCL_L2ARC, dev,
RW_READER);
}
/*
* When rebuilding L2ARC we retrieve the evict hand
* from the header of the device. Of note, l2arc_evict()
* does not actually delete buffers from the cache
* device, but trimming may do so depending on the
* hardware implementation. Thus keeping track of the
* evict hand is useful.
*/
dev->l2ad_evict = MAX(dev->l2ad_evict, taddr);
}
}
retry:
mutex_enter(&dev->l2ad_mtx);
/*
* We have to account for evicted log blocks. Run vdev_space_update()
* on log blocks whose offset (in bytes) is before the evicted offset
* (in bytes) by searching in the list of pointers to log blocks
* present in the L2ARC device.
*/
for (lb_ptr_buf = list_tail(&dev->l2ad_lbptr_list); lb_ptr_buf;
lb_ptr_buf = lb_ptr_buf_prev) {
lb_ptr_buf_prev = list_prev(&dev->l2ad_lbptr_list, lb_ptr_buf);
/* L2BLK_GET_PSIZE returns aligned size for log blocks */
uint64_t asize = L2BLK_GET_PSIZE(
(lb_ptr_buf->lb_ptr)->lbp_prop);
/*
* We don't worry about log blocks left behind (ie
* lbp_payload_start < l2ad_hand) because l2arc_write_buffers()
* will never write more than l2arc_evict() evicts.
*/
if (!all && l2arc_log_blkptr_valid(dev, lb_ptr_buf->lb_ptr)) {
break;
} else {
vdev_space_update(vd, -asize, 0, 0);
ARCSTAT_INCR(arcstat_l2_log_blk_asize, -asize);
ARCSTAT_BUMPDOWN(arcstat_l2_log_blk_count);
zfs_refcount_remove_many(&dev->l2ad_lb_asize, asize,
lb_ptr_buf);
zfs_refcount_remove(&dev->l2ad_lb_count, lb_ptr_buf);
list_remove(&dev->l2ad_lbptr_list, lb_ptr_buf);
kmem_free(lb_ptr_buf->lb_ptr,
sizeof (l2arc_log_blkptr_t));
kmem_free(lb_ptr_buf, sizeof (l2arc_lb_ptr_buf_t));
}
}
for (hdr = list_tail(buflist); hdr; hdr = hdr_prev) {
hdr_prev = list_prev(buflist, hdr);
ASSERT(!HDR_EMPTY(hdr));
hash_lock = HDR_LOCK(hdr);
/*
* We cannot use mutex_enter or else we can deadlock
* with l2arc_write_buffers (due to swapping the order
* the hash lock and l2ad_mtx are taken).
*/
if (!mutex_tryenter(hash_lock)) {
/*
* Missed the hash lock. Retry.
*/
ARCSTAT_BUMP(arcstat_l2_evict_lock_retry);
mutex_exit(&dev->l2ad_mtx);
mutex_enter(hash_lock);
mutex_exit(hash_lock);
goto retry;
}
/*
* A header can't be on this list if it doesn't have L2 header.
*/
ASSERT(HDR_HAS_L2HDR(hdr));
/* Ensure this header has finished being written. */
ASSERT(!HDR_L2_WRITING(hdr));
ASSERT(!HDR_L2_WRITE_HEAD(hdr));
if (!all && (hdr->b_l2hdr.b_daddr >= dev->l2ad_evict ||
hdr->b_l2hdr.b_daddr < dev->l2ad_hand)) {
/*
* We've evicted to the target address,
* or the end of the device.
*/
mutex_exit(hash_lock);
break;
}
if (!HDR_HAS_L1HDR(hdr)) {
ASSERT(!HDR_L2_READING(hdr));
/*
* This doesn't exist in the ARC. Destroy.
* arc_hdr_destroy() will call list_remove()
* and decrement arcstat_l2_lsize.
*/
arc_change_state(arc_anon, hdr, hash_lock);
arc_hdr_destroy(hdr);
} else {
ASSERT(hdr->b_l1hdr.b_state != arc_l2c_only);
ARCSTAT_BUMP(arcstat_l2_evict_l1cached);
/*
* Invalidate issued or about to be issued
* reads, since we may be about to write
* over this location.
*/
if (HDR_L2_READING(hdr)) {
ARCSTAT_BUMP(arcstat_l2_evict_reading);
arc_hdr_set_flags(hdr, ARC_FLAG_L2_EVICTED);
}
arc_hdr_l2hdr_destroy(hdr);
}
mutex_exit(hash_lock);
}
mutex_exit(&dev->l2ad_mtx);
out:
/*
* We need to check if we evict all buffers, otherwise we may iterate
* unnecessarily.
*/
if (!all && rerun) {
/*
* Bump device hand to the device start if it is approaching the
* end. l2arc_evict() has already evicted ahead for this case.
*/
dev->l2ad_hand = dev->l2ad_start;
dev->l2ad_evict = dev->l2ad_start;
dev->l2ad_first = B_FALSE;
goto top;
}
if (!all) {
/*
* In case of cache device removal (all) the following
* assertions may be violated without functional consequences
* as the device is about to be removed.
*/
ASSERT3U(dev->l2ad_hand + distance, <, dev->l2ad_end);
if (!dev->l2ad_first)
ASSERT3U(dev->l2ad_hand, <, dev->l2ad_evict);
}
}
/*
* Handle any abd transforms that might be required for writing to the L2ARC.
* If successful, this function will always return an abd with the data
* transformed as it is on disk in a new abd of asize bytes.
*/
static int
l2arc_apply_transforms(spa_t *spa, arc_buf_hdr_t *hdr, uint64_t asize,
abd_t **abd_out)
{
int ret;
void *tmp = NULL;
abd_t *cabd = NULL, *eabd = NULL, *to_write = hdr->b_l1hdr.b_pabd;
enum zio_compress compress = HDR_GET_COMPRESS(hdr);
uint64_t psize = HDR_GET_PSIZE(hdr);
uint64_t size = arc_hdr_size(hdr);
boolean_t ismd = HDR_ISTYPE_METADATA(hdr);
boolean_t bswap = (hdr->b_l1hdr.b_byteswap != DMU_BSWAP_NUMFUNCS);
dsl_crypto_key_t *dck = NULL;
uint8_t mac[ZIO_DATA_MAC_LEN] = { 0 };
boolean_t no_crypt = B_FALSE;
ASSERT((HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF &&
!HDR_COMPRESSION_ENABLED(hdr)) ||
HDR_ENCRYPTED(hdr) || HDR_SHARED_DATA(hdr) || psize != asize);
ASSERT3U(psize, <=, asize);
/*
* If this data simply needs its own buffer, we simply allocate it
* and copy the data. This may be done to eliminate a dependency on a
* shared buffer or to reallocate the buffer to match asize.
*/
if (HDR_HAS_RABD(hdr) && asize != psize) {
ASSERT3U(asize, >=, psize);
to_write = abd_alloc_for_io(asize, ismd);
abd_copy(to_write, hdr->b_crypt_hdr.b_rabd, psize);
if (psize != asize)
abd_zero_off(to_write, psize, asize - psize);
goto out;
}
if ((compress == ZIO_COMPRESS_OFF || HDR_COMPRESSION_ENABLED(hdr)) &&
!HDR_ENCRYPTED(hdr)) {
ASSERT3U(size, ==, psize);
to_write = abd_alloc_for_io(asize, ismd);
abd_copy(to_write, hdr->b_l1hdr.b_pabd, size);
if (size != asize)
abd_zero_off(to_write, size, asize - size);
goto out;
}
if (compress != ZIO_COMPRESS_OFF && !HDR_COMPRESSION_ENABLED(hdr)) {
cabd = abd_alloc_for_io(asize, ismd);
tmp = abd_borrow_buf(cabd, asize);
psize = zio_compress_data(compress, to_write, tmp, size,
hdr->b_complevel);
if (psize >= size) {
abd_return_buf(cabd, tmp, asize);
HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_OFF);
to_write = cabd;
abd_copy(to_write, hdr->b_l1hdr.b_pabd, size);
if (size != asize)
abd_zero_off(to_write, size, asize - size);
goto encrypt;
}
ASSERT3U(psize, <=, HDR_GET_PSIZE(hdr));
if (psize < asize)
bzero((char *)tmp + psize, asize - psize);
psize = HDR_GET_PSIZE(hdr);
abd_return_buf_copy(cabd, tmp, asize);
to_write = cabd;
}
encrypt:
if (HDR_ENCRYPTED(hdr)) {
eabd = abd_alloc_for_io(asize, ismd);
/*
* If the dataset was disowned before the buffer
* made it to this point, the key to re-encrypt
* it won't be available. In this case we simply
* won't write the buffer to the L2ARC.
*/
ret = spa_keystore_lookup_key(spa, hdr->b_crypt_hdr.b_dsobj,
FTAG, &dck);
if (ret != 0)
goto error;
ret = zio_do_crypt_abd(B_TRUE, &dck->dck_key,
hdr->b_crypt_hdr.b_ot, bswap, hdr->b_crypt_hdr.b_salt,
hdr->b_crypt_hdr.b_iv, mac, psize, to_write, eabd,
&no_crypt);
if (ret != 0)
goto error;
if (no_crypt)
abd_copy(eabd, to_write, psize);
if (psize != asize)
abd_zero_off(eabd, psize, asize - psize);
/* assert that the MAC we got here matches the one we saved */
ASSERT0(bcmp(mac, hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN));
spa_keystore_dsl_key_rele(spa, dck, FTAG);
if (to_write == cabd)
abd_free(cabd);
to_write = eabd;
}
out:
ASSERT3P(to_write, !=, hdr->b_l1hdr.b_pabd);
*abd_out = to_write;
return (0);
error:
if (dck != NULL)
spa_keystore_dsl_key_rele(spa, dck, FTAG);
if (cabd != NULL)
abd_free(cabd);
if (eabd != NULL)
abd_free(eabd);
*abd_out = NULL;
return (ret);
}
static void
l2arc_blk_fetch_done(zio_t *zio)
{
l2arc_read_callback_t *cb;
cb = zio->io_private;
if (cb->l2rcb_abd != NULL)
abd_free(cb->l2rcb_abd);
kmem_free(cb, sizeof (l2arc_read_callback_t));
}
/*
* Find and write ARC buffers to the L2ARC device.
*
* An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid
* for reading until they have completed writing.
* The headroom_boost is an in-out parameter used to maintain headroom boost
* state between calls to this function.
*
* Returns the number of bytes actually written (which may be smaller than
* the delta by which the device hand has changed due to alignment and the
* writing of log blocks).
*/
static uint64_t
l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz)
{
arc_buf_hdr_t *hdr, *hdr_prev, *head;
uint64_t write_asize, write_psize, write_lsize, headroom;
boolean_t full;
l2arc_write_callback_t *cb = NULL;
zio_t *pio, *wzio;
uint64_t guid = spa_load_guid(spa);
l2arc_dev_hdr_phys_t *l2dhdr = dev->l2ad_dev_hdr;
ASSERT3P(dev->l2ad_vdev, !=, NULL);
pio = NULL;
write_lsize = write_asize = write_psize = 0;
full = B_FALSE;
head = kmem_cache_alloc(hdr_l2only_cache, KM_PUSHPAGE);
arc_hdr_set_flags(head, ARC_FLAG_L2_WRITE_HEAD | ARC_FLAG_HAS_L2HDR);
/*
* Copy buffers for L2ARC writing.
*/
for (int pass = 0; pass < L2ARC_FEED_TYPES; pass++) {
/*
* If pass == 1 or 3, we cache MRU metadata and data
* respectively.
*/
if (l2arc_mfuonly) {
if (pass == 1 || pass == 3)
continue;
}
multilist_sublist_t *mls = l2arc_sublist_lock(pass);
uint64_t passed_sz = 0;
VERIFY3P(mls, !=, NULL);
/*
* L2ARC fast warmup.
*
* Until the ARC is warm and starts to evict, read from the
* head of the ARC lists rather than the tail.
*/
if (arc_warm == B_FALSE)
hdr = multilist_sublist_head(mls);
else
hdr = multilist_sublist_tail(mls);
headroom = target_sz * l2arc_headroom;
if (zfs_compressed_arc_enabled)
headroom = (headroom * l2arc_headroom_boost) / 100;
for (; hdr; hdr = hdr_prev) {
kmutex_t *hash_lock;
abd_t *to_write = NULL;
if (arc_warm == B_FALSE)
hdr_prev = multilist_sublist_next(mls, hdr);
else
hdr_prev = multilist_sublist_prev(mls, hdr);
hash_lock = HDR_LOCK(hdr);
if (!mutex_tryenter(hash_lock)) {
/*
* Skip this buffer rather than waiting.
*/
continue;
}
passed_sz += HDR_GET_LSIZE(hdr);
if (l2arc_headroom != 0 && passed_sz > headroom) {
/*
* Searched too far.
*/
mutex_exit(hash_lock);
break;
}
if (!l2arc_write_eligible(guid, hdr)) {
mutex_exit(hash_lock);
continue;
}
/*
* We rely on the L1 portion of the header below, so
* it's invalid for this header to have been evicted out
* of the ghost cache, prior to being written out. The
* ARC_FLAG_L2_WRITING bit ensures this won't happen.
*/
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT3U(HDR_GET_PSIZE(hdr), >, 0);
ASSERT3U(arc_hdr_size(hdr), >, 0);
ASSERT(hdr->b_l1hdr.b_pabd != NULL ||
HDR_HAS_RABD(hdr));
uint64_t psize = HDR_GET_PSIZE(hdr);
uint64_t asize = vdev_psize_to_asize(dev->l2ad_vdev,
psize);
if ((write_asize + asize) > target_sz) {
full = B_TRUE;
mutex_exit(hash_lock);
break;
}
/*
* We rely on the L1 portion of the header below, so
* it's invalid for this header to have been evicted out
* of the ghost cache, prior to being written out. The
* ARC_FLAG_L2_WRITING bit ensures this won't happen.
*/
arc_hdr_set_flags(hdr, ARC_FLAG_L2_WRITING);
ASSERT(HDR_HAS_L1HDR(hdr));
ASSERT3U(HDR_GET_PSIZE(hdr), >, 0);
ASSERT(hdr->b_l1hdr.b_pabd != NULL ||
HDR_HAS_RABD(hdr));
ASSERT3U(arc_hdr_size(hdr), >, 0);
/*
* If this header has b_rabd, we can use this since it
* must always match the data exactly as it exists on
* disk. Otherwise, the L2ARC can normally use the
* hdr's data, but if we're sharing data between the
* hdr and one of its bufs, L2ARC needs its own copy of
* the data so that the ZIO below can't race with the
* buf consumer. To ensure that this copy will be
* available for the lifetime of the ZIO and be cleaned
* up afterwards, we add it to the l2arc_free_on_write
* queue. If we need to apply any transforms to the
* data (compression, encryption) we will also need the
* extra buffer.
*/
if (HDR_HAS_RABD(hdr) && psize == asize) {
to_write = hdr->b_crypt_hdr.b_rabd;
} else if ((HDR_COMPRESSION_ENABLED(hdr) ||
HDR_GET_COMPRESS(hdr) == ZIO_COMPRESS_OFF) &&
!HDR_ENCRYPTED(hdr) && !HDR_SHARED_DATA(hdr) &&
psize == asize) {
to_write = hdr->b_l1hdr.b_pabd;
} else {
int ret;
arc_buf_contents_t type = arc_buf_type(hdr);
ret = l2arc_apply_transforms(spa, hdr, asize,
&to_write);
if (ret != 0) {
arc_hdr_clear_flags(hdr,
ARC_FLAG_L2_WRITING);
mutex_exit(hash_lock);
continue;
}
l2arc_free_abd_on_write(to_write, asize, type);
}
if (pio == NULL) {
/*
* Insert a dummy header on the buflist so
* l2arc_write_done() can find where the
* write buffers begin without searching.
*/
mutex_enter(&dev->l2ad_mtx);
list_insert_head(&dev->l2ad_buflist, head);
mutex_exit(&dev->l2ad_mtx);
cb = kmem_alloc(
sizeof (l2arc_write_callback_t), KM_SLEEP);
cb->l2wcb_dev = dev;
cb->l2wcb_head = head;
/*
* Create a list to save allocated abd buffers
* for l2arc_log_blk_commit().
*/
list_create(&cb->l2wcb_abd_list,
sizeof (l2arc_lb_abd_buf_t),
offsetof(l2arc_lb_abd_buf_t, node));
pio = zio_root(spa, l2arc_write_done, cb,
ZIO_FLAG_CANFAIL);
}
hdr->b_l2hdr.b_dev = dev;
hdr->b_l2hdr.b_hits = 0;
hdr->b_l2hdr.b_daddr = dev->l2ad_hand;
hdr->b_l2hdr.b_arcs_state =
hdr->b_l1hdr.b_state->arcs_state;
arc_hdr_set_flags(hdr, ARC_FLAG_HAS_L2HDR);
mutex_enter(&dev->l2ad_mtx);
list_insert_head(&dev->l2ad_buflist, hdr);
mutex_exit(&dev->l2ad_mtx);
(void) zfs_refcount_add_many(&dev->l2ad_alloc,
arc_hdr_size(hdr), hdr);
wzio = zio_write_phys(pio, dev->l2ad_vdev,
hdr->b_l2hdr.b_daddr, asize, to_write,
ZIO_CHECKSUM_OFF, NULL, hdr,
ZIO_PRIORITY_ASYNC_WRITE,
ZIO_FLAG_CANFAIL, B_FALSE);
write_lsize += HDR_GET_LSIZE(hdr);
DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev,
zio_t *, wzio);
write_psize += psize;
write_asize += asize;
dev->l2ad_hand += asize;
l2arc_hdr_arcstats_increment(hdr);
vdev_space_update(dev->l2ad_vdev, asize, 0, 0);
mutex_exit(hash_lock);
/*
* Append buf info to current log and commit if full.
* arcstat_l2_{size,asize} kstats are updated
* internally.
*/
if (l2arc_log_blk_insert(dev, hdr))
l2arc_log_blk_commit(dev, pio, cb);
zio_nowait(wzio);
}
multilist_sublist_unlock(mls);
if (full == B_TRUE)
break;
}
/* No buffers selected for writing? */
if (pio == NULL) {
ASSERT0(write_lsize);
ASSERT(!HDR_HAS_L1HDR(head));
kmem_cache_free(hdr_l2only_cache, head);
/*
* Although we did not write any buffers l2ad_evict may
* have advanced.
*/
if (dev->l2ad_evict != l2dhdr->dh_evict)
l2arc_dev_hdr_update(dev);
return (0);
}
if (!dev->l2ad_first)
ASSERT3U(dev->l2ad_hand, <=, dev->l2ad_evict);
ASSERT3U(write_asize, <=, target_sz);
ARCSTAT_BUMP(arcstat_l2_writes_sent);
ARCSTAT_INCR(arcstat_l2_write_bytes, write_psize);
dev->l2ad_writing = B_TRUE;
(void) zio_wait(pio);
dev->l2ad_writing = B_FALSE;
/*
* Update the device header after the zio completes as
* l2arc_write_done() may have updated the memory holding the log block
* pointers in the device header.
*/
l2arc_dev_hdr_update(dev);
return (write_asize);
}
static boolean_t
l2arc_hdr_limit_reached(void)
{
int64_t s = aggsum_upper_bound(&arc_sums.arcstat_l2_hdr_size);
return (arc_reclaim_needed() || (s > arc_meta_limit * 3 / 4) ||
(s > (arc_warm ? arc_c : arc_c_max) * l2arc_meta_percent / 100));
}
/*
* This thread feeds the L2ARC at regular intervals. This is the beating
* heart of the L2ARC.
*/
/* ARGSUSED */
static void
l2arc_feed_thread(void *unused)
{
callb_cpr_t cpr;
l2arc_dev_t *dev;
spa_t *spa;
uint64_t size, wrote;
clock_t begin, next = ddi_get_lbolt();
fstrans_cookie_t cookie;
CALLB_CPR_INIT(&cpr, &l2arc_feed_thr_lock, callb_generic_cpr, FTAG);
mutex_enter(&l2arc_feed_thr_lock);
cookie = spl_fstrans_mark();
while (l2arc_thread_exit == 0) {
CALLB_CPR_SAFE_BEGIN(&cpr);
(void) cv_timedwait_idle(&l2arc_feed_thr_cv,
&l2arc_feed_thr_lock, next);
CALLB_CPR_SAFE_END(&cpr, &l2arc_feed_thr_lock);
next = ddi_get_lbolt() + hz;
/*
* Quick check for L2ARC devices.
*/
mutex_enter(&l2arc_dev_mtx);
if (l2arc_ndev == 0) {
mutex_exit(&l2arc_dev_mtx);
continue;
}
mutex_exit(&l2arc_dev_mtx);
begin = ddi_get_lbolt();
/*
* This selects the next l2arc device to write to, and in
* doing so the next spa to feed from: dev->l2ad_spa. This
* will return NULL if there are now no l2arc devices or if
* they are all faulted.
*
* If a device is returned, its spa's config lock is also
* held to prevent device removal. l2arc_dev_get_next()
* will grab and release l2arc_dev_mtx.
*/
if ((dev = l2arc_dev_get_next()) == NULL)
continue;
spa = dev->l2ad_spa;
ASSERT3P(spa, !=, NULL);
/*
* If the pool is read-only then force the feed thread to
* sleep a little longer.
*/
if (!spa_writeable(spa)) {
next = ddi_get_lbolt() + 5 * l2arc_feed_secs * hz;
spa_config_exit(spa, SCL_L2ARC, dev);
continue;
}
/*
* Avoid contributing to memory pressure.
*/
if (l2arc_hdr_limit_reached()) {
ARCSTAT_BUMP(arcstat_l2_abort_lowmem);
spa_config_exit(spa, SCL_L2ARC, dev);
continue;
}
ARCSTAT_BUMP(arcstat_l2_feeds);
size = l2arc_write_size(dev);
/*
* Evict L2ARC buffers that will be overwritten.
*/
l2arc_evict(dev, size, B_FALSE);
/*
* Write ARC buffers.
*/
wrote = l2arc_write_buffers(spa, dev, size);
/*
* Calculate interval between writes.
*/
next = l2arc_write_interval(begin, size, wrote);
spa_config_exit(spa, SCL_L2ARC, dev);
}
spl_fstrans_unmark(cookie);
l2arc_thread_exit = 0;
cv_broadcast(&l2arc_feed_thr_cv);
CALLB_CPR_EXIT(&cpr); /* drops l2arc_feed_thr_lock */
thread_exit();
}
boolean_t
l2arc_vdev_present(vdev_t *vd)
{
return (l2arc_vdev_get(vd) != NULL);
}
/*
* Returns the l2arc_dev_t associated with a particular vdev_t or NULL if
* the vdev_t isn't an L2ARC device.
*/
l2arc_dev_t *
l2arc_vdev_get(vdev_t *vd)
{
l2arc_dev_t *dev;
mutex_enter(&l2arc_dev_mtx);
for (dev = list_head(l2arc_dev_list); dev != NULL;
dev = list_next(l2arc_dev_list, dev)) {
if (dev->l2ad_vdev == vd)
break;
}
mutex_exit(&l2arc_dev_mtx);
return (dev);
}
/*
* Add a vdev for use by the L2ARC. By this point the spa has already
* validated the vdev and opened it.
*/
void
l2arc_add_vdev(spa_t *spa, vdev_t *vd)
{
l2arc_dev_t *adddev;
uint64_t l2dhdr_asize;
ASSERT(!l2arc_vdev_present(vd));
/*
* Create a new l2arc device entry.
*/
adddev = vmem_zalloc(sizeof (l2arc_dev_t), KM_SLEEP);
adddev->l2ad_spa = spa;
adddev->l2ad_vdev = vd;
/* leave extra size for an l2arc device header */
l2dhdr_asize = adddev->l2ad_dev_hdr_asize =
MAX(sizeof (*adddev->l2ad_dev_hdr), 1 << vd->vdev_ashift);
adddev->l2ad_start = VDEV_LABEL_START_SIZE + l2dhdr_asize;
adddev->l2ad_end = VDEV_LABEL_START_SIZE + vdev_get_min_asize(vd);
ASSERT3U(adddev->l2ad_start, <, adddev->l2ad_end);
adddev->l2ad_hand = adddev->l2ad_start;
adddev->l2ad_evict = adddev->l2ad_start;
adddev->l2ad_first = B_TRUE;
adddev->l2ad_writing = B_FALSE;
adddev->l2ad_trim_all = B_FALSE;
list_link_init(&adddev->l2ad_node);
adddev->l2ad_dev_hdr = kmem_zalloc(l2dhdr_asize, KM_SLEEP);
mutex_init(&adddev->l2ad_mtx, NULL, MUTEX_DEFAULT, NULL);
/*
* This is a list of all ARC buffers that are still valid on the
* device.
*/
list_create(&adddev->l2ad_buflist, sizeof (arc_buf_hdr_t),
offsetof(arc_buf_hdr_t, b_l2hdr.b_l2node));
/*
* This is a list of pointers to log blocks that are still present
* on the device.
*/
list_create(&adddev->l2ad_lbptr_list, sizeof (l2arc_lb_ptr_buf_t),
offsetof(l2arc_lb_ptr_buf_t, node));
vdev_space_update(vd, 0, 0, adddev->l2ad_end - adddev->l2ad_hand);
zfs_refcount_create(&adddev->l2ad_alloc);
zfs_refcount_create(&adddev->l2ad_lb_asize);
zfs_refcount_create(&adddev->l2ad_lb_count);
/*
* Add device to global list
*/
mutex_enter(&l2arc_dev_mtx);
list_insert_head(l2arc_dev_list, adddev);
atomic_inc_64(&l2arc_ndev);
mutex_exit(&l2arc_dev_mtx);
/*
* Decide if vdev is eligible for L2ARC rebuild
*/
l2arc_rebuild_vdev(adddev->l2ad_vdev, B_FALSE);
}
void
l2arc_rebuild_vdev(vdev_t *vd, boolean_t reopen)
{
l2arc_dev_t *dev = NULL;
l2arc_dev_hdr_phys_t *l2dhdr;
uint64_t l2dhdr_asize;
spa_t *spa;
dev = l2arc_vdev_get(vd);
ASSERT3P(dev, !=, NULL);
spa = dev->l2ad_spa;
l2dhdr = dev->l2ad_dev_hdr;
l2dhdr_asize = dev->l2ad_dev_hdr_asize;
/*
* The L2ARC has to hold at least the payload of one log block for
* them to be restored (persistent L2ARC). The payload of a log block
* depends on the amount of its log entries. We always write log blocks
* with 1022 entries. How many of them are committed or restored depends
* on the size of the L2ARC device. Thus the maximum payload of
* one log block is 1022 * SPA_MAXBLOCKSIZE = 16GB. If the L2ARC device
* is less than that, we reduce the amount of committed and restored
* log entries per block so as to enable persistence.
*/
if (dev->l2ad_end < l2arc_rebuild_blocks_min_l2size) {
dev->l2ad_log_entries = 0;
} else {
dev->l2ad_log_entries = MIN((dev->l2ad_end -
dev->l2ad_start) >> SPA_MAXBLOCKSHIFT,
L2ARC_LOG_BLK_MAX_ENTRIES);
}
/*
* Read the device header, if an error is returned do not rebuild L2ARC.
*/
if (l2arc_dev_hdr_read(dev) == 0 && dev->l2ad_log_entries > 0) {
/*
* If we are onlining a cache device (vdev_reopen) that was
* still present (l2arc_vdev_present()) and rebuild is enabled,
* we should evict all ARC buffers and pointers to log blocks
* and reclaim their space before restoring its contents to
* L2ARC.
*/
if (reopen) {
if (!l2arc_rebuild_enabled) {
return;
} else {
l2arc_evict(dev, 0, B_TRUE);
/* start a new log block */
dev->l2ad_log_ent_idx = 0;
dev->l2ad_log_blk_payload_asize = 0;
dev->l2ad_log_blk_payload_start = 0;
}
}
/*
* Just mark the device as pending for a rebuild. We won't
* be starting a rebuild in line here as it would block pool
* import. Instead spa_load_impl will hand that off to an
* async task which will call l2arc_spa_rebuild_start.
*/
dev->l2ad_rebuild = B_TRUE;
} else if (spa_writeable(spa)) {
/*
* In this case TRIM the whole device if l2arc_trim_ahead > 0,
* otherwise create a new header. We zero out the memory holding
* the header to reset dh_start_lbps. If we TRIM the whole
* device the new header will be written by
* vdev_trim_l2arc_thread() at the end of the TRIM to update the
* trim_state in the header too. When reading the header, if
* trim_state is not VDEV_TRIM_COMPLETE and l2arc_trim_ahead > 0
* we opt to TRIM the whole device again.
*/
if (l2arc_trim_ahead > 0) {
dev->l2ad_trim_all = B_TRUE;
} else {
bzero(l2dhdr, l2dhdr_asize);
l2arc_dev_hdr_update(dev);
}
}
}
/*
* Remove a vdev from the L2ARC.
*/
void
l2arc_remove_vdev(vdev_t *vd)
{
l2arc_dev_t *remdev = NULL;
/*
* Find the device by vdev
*/
remdev = l2arc_vdev_get(vd);
ASSERT3P(remdev, !=, NULL);
/*
* Cancel any ongoing or scheduled rebuild.
*/
mutex_enter(&l2arc_rebuild_thr_lock);
if (remdev->l2ad_rebuild_began == B_TRUE) {
remdev->l2ad_rebuild_cancel = B_TRUE;
while (remdev->l2ad_rebuild == B_TRUE)
cv_wait(&l2arc_rebuild_thr_cv, &l2arc_rebuild_thr_lock);
}
mutex_exit(&l2arc_rebuild_thr_lock);
/*
* Remove device from global list
*/
mutex_enter(&l2arc_dev_mtx);
list_remove(l2arc_dev_list, remdev);
l2arc_dev_last = NULL; /* may have been invalidated */
atomic_dec_64(&l2arc_ndev);
mutex_exit(&l2arc_dev_mtx);
/*
* Clear all buflists and ARC references. L2ARC device flush.
*/
l2arc_evict(remdev, 0, B_TRUE);
list_destroy(&remdev->l2ad_buflist);
ASSERT(list_is_empty(&remdev->l2ad_lbptr_list));
list_destroy(&remdev->l2ad_lbptr_list);
mutex_destroy(&remdev->l2ad_mtx);
zfs_refcount_destroy(&remdev->l2ad_alloc);
zfs_refcount_destroy(&remdev->l2ad_lb_asize);
zfs_refcount_destroy(&remdev->l2ad_lb_count);
kmem_free(remdev->l2ad_dev_hdr, remdev->l2ad_dev_hdr_asize);
vmem_free(remdev, sizeof (l2arc_dev_t));
}
void
l2arc_init(void)
{
l2arc_thread_exit = 0;
l2arc_ndev = 0;
mutex_init(&l2arc_feed_thr_lock, NULL, MUTEX_DEFAULT, NULL);
cv_init(&l2arc_feed_thr_cv, NULL, CV_DEFAULT, NULL);
mutex_init(&l2arc_rebuild_thr_lock, NULL, MUTEX_DEFAULT, NULL);
cv_init(&l2arc_rebuild_thr_cv, NULL, CV_DEFAULT, NULL);
mutex_init(&l2arc_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&l2arc_free_on_write_mtx, NULL, MUTEX_DEFAULT, NULL);
l2arc_dev_list = &L2ARC_dev_list;
l2arc_free_on_write = &L2ARC_free_on_write;
list_create(l2arc_dev_list, sizeof (l2arc_dev_t),
offsetof(l2arc_dev_t, l2ad_node));
list_create(l2arc_free_on_write, sizeof (l2arc_data_free_t),
offsetof(l2arc_data_free_t, l2df_list_node));
}
void
l2arc_fini(void)
{
mutex_destroy(&l2arc_feed_thr_lock);
cv_destroy(&l2arc_feed_thr_cv);
mutex_destroy(&l2arc_rebuild_thr_lock);
cv_destroy(&l2arc_rebuild_thr_cv);
mutex_destroy(&l2arc_dev_mtx);
mutex_destroy(&l2arc_free_on_write_mtx);
list_destroy(l2arc_dev_list);
list_destroy(l2arc_free_on_write);
}
void
l2arc_start(void)
{
if (!(spa_mode_global & SPA_MODE_WRITE))
return;
(void) thread_create(NULL, 0, l2arc_feed_thread, NULL, 0, &p0,
TS_RUN, defclsyspri);
}
void
l2arc_stop(void)
{
if (!(spa_mode_global & SPA_MODE_WRITE))
return;
mutex_enter(&l2arc_feed_thr_lock);
cv_signal(&l2arc_feed_thr_cv); /* kick thread out of startup */
l2arc_thread_exit = 1;
while (l2arc_thread_exit != 0)
cv_wait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock);
mutex_exit(&l2arc_feed_thr_lock);
}
/*
* Punches out rebuild threads for the L2ARC devices in a spa. This should
* be called after pool import from the spa async thread, since starting
* these threads directly from spa_import() will make them part of the
* "zpool import" context and delay process exit (and thus pool import).
*/
void
l2arc_spa_rebuild_start(spa_t *spa)
{
ASSERT(MUTEX_HELD(&spa_namespace_lock));
/*
* Locate the spa's l2arc devices and kick off rebuild threads.
*/
for (int i = 0; i < spa->spa_l2cache.sav_count; i++) {
l2arc_dev_t *dev =
l2arc_vdev_get(spa->spa_l2cache.sav_vdevs[i]);
if (dev == NULL) {
/* Don't attempt a rebuild if the vdev is UNAVAIL */
continue;
}
mutex_enter(&l2arc_rebuild_thr_lock);
if (dev->l2ad_rebuild && !dev->l2ad_rebuild_cancel) {
dev->l2ad_rebuild_began = B_TRUE;
(void) thread_create(NULL, 0, l2arc_dev_rebuild_thread,
dev, 0, &p0, TS_RUN, minclsyspri);
}
mutex_exit(&l2arc_rebuild_thr_lock);
}
}
/*
* Main entry point for L2ARC rebuilding.
*/
static void
l2arc_dev_rebuild_thread(void *arg)
{
l2arc_dev_t *dev = arg;
VERIFY(!dev->l2ad_rebuild_cancel);
VERIFY(dev->l2ad_rebuild);
(void) l2arc_rebuild(dev);
mutex_enter(&l2arc_rebuild_thr_lock);
dev->l2ad_rebuild_began = B_FALSE;
dev->l2ad_rebuild = B_FALSE;
mutex_exit(&l2arc_rebuild_thr_lock);
thread_exit();
}
/*
* This function implements the actual L2ARC metadata rebuild. It:
* starts reading the log block chain and restores each block's contents
* to memory (reconstructing arc_buf_hdr_t's).
*
* Operation stops under any of the following conditions:
*
* 1) We reach the end of the log block chain.
* 2) We encounter *any* error condition (cksum errors, io errors)
*/
static int
l2arc_rebuild(l2arc_dev_t *dev)
{
vdev_t *vd = dev->l2ad_vdev;
spa_t *spa = vd->vdev_spa;
int err = 0;
l2arc_dev_hdr_phys_t *l2dhdr = dev->l2ad_dev_hdr;
l2arc_log_blk_phys_t *this_lb, *next_lb;
zio_t *this_io = NULL, *next_io = NULL;
l2arc_log_blkptr_t lbps[2];
l2arc_lb_ptr_buf_t *lb_ptr_buf;
boolean_t lock_held;
this_lb = vmem_zalloc(sizeof (*this_lb), KM_SLEEP);
next_lb = vmem_zalloc(sizeof (*next_lb), KM_SLEEP);
/*
* We prevent device removal while issuing reads to the device,
* then during the rebuilding phases we drop this lock again so
* that a spa_unload or device remove can be initiated - this is
* safe, because the spa will signal us to stop before removing
* our device and wait for us to stop.
*/
spa_config_enter(spa, SCL_L2ARC, vd, RW_READER);
lock_held = B_TRUE;
/*
* Retrieve the persistent L2ARC device state.
* L2BLK_GET_PSIZE returns aligned size for log blocks.
*/
dev->l2ad_evict = MAX(l2dhdr->dh_evict, dev->l2ad_start);
dev->l2ad_hand = MAX(l2dhdr->dh_start_lbps[0].lbp_daddr +
L2BLK_GET_PSIZE((&l2dhdr->dh_start_lbps[0])->lbp_prop),
dev->l2ad_start);
dev->l2ad_first = !!(l2dhdr->dh_flags & L2ARC_DEV_HDR_EVICT_FIRST);
vd->vdev_trim_action_time = l2dhdr->dh_trim_action_time;
vd->vdev_trim_state = l2dhdr->dh_trim_state;
/*
* In case the zfs module parameter l2arc_rebuild_enabled is false
* we do not start the rebuild process.
*/
if (!l2arc_rebuild_enabled)
goto out;
/* Prepare the rebuild process */
bcopy(l2dhdr->dh_start_lbps, lbps, sizeof (lbps));
/* Start the rebuild process */
for (;;) {
if (!l2arc_log_blkptr_valid(dev, &lbps[0]))
break;
if ((err = l2arc_log_blk_read(dev, &lbps[0], &lbps[1],
this_lb, next_lb, this_io, &next_io)) != 0)
goto out;
/*
* Our memory pressure valve. If the system is running low
* on memory, rather than swamping memory with new ARC buf
* hdrs, we opt not to rebuild the L2ARC. At this point,
* however, we have already set up our L2ARC dev to chain in
* new metadata log blocks, so the user may choose to offline/
* online the L2ARC dev at a later time (or re-import the pool)
* to reconstruct it (when there's less memory pressure).
*/
if (l2arc_hdr_limit_reached()) {
ARCSTAT_BUMP(arcstat_l2_rebuild_abort_lowmem);
cmn_err(CE_NOTE, "System running low on memory, "
"aborting L2ARC rebuild.");
err = SET_ERROR(ENOMEM);
goto out;
}
spa_config_exit(spa, SCL_L2ARC, vd);
lock_held = B_FALSE;
/*
* Now that we know that the next_lb checks out alright, we
* can start reconstruction from this log block.
* L2BLK_GET_PSIZE returns aligned size for log blocks.
*/
uint64_t asize = L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
l2arc_log_blk_restore(dev, this_lb, asize);
/*
* log block restored, include its pointer in the list of
* pointers to log blocks present in the L2ARC device.
*/
lb_ptr_buf = kmem_zalloc(sizeof (l2arc_lb_ptr_buf_t), KM_SLEEP);
lb_ptr_buf->lb_ptr = kmem_zalloc(sizeof (l2arc_log_blkptr_t),
KM_SLEEP);
bcopy(&lbps[0], lb_ptr_buf->lb_ptr,
sizeof (l2arc_log_blkptr_t));
mutex_enter(&dev->l2ad_mtx);
list_insert_tail(&dev->l2ad_lbptr_list, lb_ptr_buf);
ARCSTAT_INCR(arcstat_l2_log_blk_asize, asize);
ARCSTAT_BUMP(arcstat_l2_log_blk_count);
zfs_refcount_add_many(&dev->l2ad_lb_asize, asize, lb_ptr_buf);
zfs_refcount_add(&dev->l2ad_lb_count, lb_ptr_buf);
mutex_exit(&dev->l2ad_mtx);
vdev_space_update(vd, asize, 0, 0);
/*
* Protection against loops of log blocks:
*
* l2ad_hand l2ad_evict
* V V
* l2ad_start |=======================================| l2ad_end
* -----|||----|||---|||----|||
* (3) (2) (1) (0)
* ---|||---|||----|||---|||
* (7) (6) (5) (4)
*
* In this situation the pointer of log block (4) passes
* l2arc_log_blkptr_valid() but the log block should not be
* restored as it is overwritten by the payload of log block
* (0). Only log blocks (0)-(3) should be restored. We check
* whether l2ad_evict lies in between the payload starting
* offset of the next log block (lbps[1].lbp_payload_start)
* and the payload starting offset of the present log block
* (lbps[0].lbp_payload_start). If true and this isn't the
* first pass, we are looping from the beginning and we should
* stop.
*/
if (l2arc_range_check_overlap(lbps[1].lbp_payload_start,
lbps[0].lbp_payload_start, dev->l2ad_evict) &&
!dev->l2ad_first)
goto out;
cond_resched();
for (;;) {
mutex_enter(&l2arc_rebuild_thr_lock);
if (dev->l2ad_rebuild_cancel) {
dev->l2ad_rebuild = B_FALSE;
cv_signal(&l2arc_rebuild_thr_cv);
mutex_exit(&l2arc_rebuild_thr_lock);
err = SET_ERROR(ECANCELED);
goto out;
}
mutex_exit(&l2arc_rebuild_thr_lock);
if (spa_config_tryenter(spa, SCL_L2ARC, vd,
RW_READER)) {
lock_held = B_TRUE;
break;
}
/*
* L2ARC config lock held by somebody in writer,
* possibly due to them trying to remove us. They'll
* likely to want us to shut down, so after a little
* delay, we check l2ad_rebuild_cancel and retry
* the lock again.
*/
delay(1);
}
/*
* Continue with the next log block.
*/
lbps[0] = lbps[1];
lbps[1] = this_lb->lb_prev_lbp;
PTR_SWAP(this_lb, next_lb);
this_io = next_io;
next_io = NULL;
}
if (this_io != NULL)
l2arc_log_blk_fetch_abort(this_io);
out:
if (next_io != NULL)
l2arc_log_blk_fetch_abort(next_io);
vmem_free(this_lb, sizeof (*this_lb));
vmem_free(next_lb, sizeof (*next_lb));
if (!l2arc_rebuild_enabled) {
spa_history_log_internal(spa, "L2ARC rebuild", NULL,
"disabled");
} else if (err == 0 && zfs_refcount_count(&dev->l2ad_lb_count) > 0) {
ARCSTAT_BUMP(arcstat_l2_rebuild_success);
spa_history_log_internal(spa, "L2ARC rebuild", NULL,
"successful, restored %llu blocks",
(u_longlong_t)zfs_refcount_count(&dev->l2ad_lb_count));
} else if (err == 0 && zfs_refcount_count(&dev->l2ad_lb_count) == 0) {
/*
* No error but also nothing restored, meaning the lbps array
* in the device header points to invalid/non-present log
* blocks. Reset the header.
*/
spa_history_log_internal(spa, "L2ARC rebuild", NULL,
"no valid log blocks");
bzero(l2dhdr, dev->l2ad_dev_hdr_asize);
l2arc_dev_hdr_update(dev);
} else if (err == ECANCELED) {
/*
* In case the rebuild was canceled do not log to spa history
* log as the pool may be in the process of being removed.
*/
zfs_dbgmsg("L2ARC rebuild aborted, restored %llu blocks",
(u_longlong_t)zfs_refcount_count(&dev->l2ad_lb_count));
} else if (err != 0) {
spa_history_log_internal(spa, "L2ARC rebuild", NULL,
"aborted, restored %llu blocks",
(u_longlong_t)zfs_refcount_count(&dev->l2ad_lb_count));
}
if (lock_held)
spa_config_exit(spa, SCL_L2ARC, vd);
return (err);
}
/*
* Attempts to read the device header on the provided L2ARC device and writes
* it to `hdr'. On success, this function returns 0, otherwise the appropriate
* error code is returned.
*/
static int
l2arc_dev_hdr_read(l2arc_dev_t *dev)
{
int err;
uint64_t guid;
l2arc_dev_hdr_phys_t *l2dhdr = dev->l2ad_dev_hdr;
const uint64_t l2dhdr_asize = dev->l2ad_dev_hdr_asize;
abd_t *abd;
guid = spa_guid(dev->l2ad_vdev->vdev_spa);
abd = abd_get_from_buf(l2dhdr, l2dhdr_asize);
err = zio_wait(zio_read_phys(NULL, dev->l2ad_vdev,
VDEV_LABEL_START_SIZE, l2dhdr_asize, abd,
ZIO_CHECKSUM_LABEL, NULL, NULL, ZIO_PRIORITY_SYNC_READ,
ZIO_FLAG_DONT_CACHE | ZIO_FLAG_CANFAIL |
ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY |
ZIO_FLAG_SPECULATIVE, B_FALSE));
abd_free(abd);
if (err != 0) {
ARCSTAT_BUMP(arcstat_l2_rebuild_abort_dh_errors);
zfs_dbgmsg("L2ARC IO error (%d) while reading device header, "
"vdev guid: %llu", err,
(u_longlong_t)dev->l2ad_vdev->vdev_guid);
return (err);
}
if (l2dhdr->dh_magic == BSWAP_64(L2ARC_DEV_HDR_MAGIC))
byteswap_uint64_array(l2dhdr, sizeof (*l2dhdr));
if (l2dhdr->dh_magic != L2ARC_DEV_HDR_MAGIC ||
l2dhdr->dh_spa_guid != guid ||
l2dhdr->dh_vdev_guid != dev->l2ad_vdev->vdev_guid ||
l2dhdr->dh_version != L2ARC_PERSISTENT_VERSION ||
l2dhdr->dh_log_entries != dev->l2ad_log_entries ||
l2dhdr->dh_end != dev->l2ad_end ||
!l2arc_range_check_overlap(dev->l2ad_start, dev->l2ad_end,
l2dhdr->dh_evict) ||
(l2dhdr->dh_trim_state != VDEV_TRIM_COMPLETE &&
l2arc_trim_ahead > 0)) {
/*
* Attempt to rebuild a device containing no actual dev hdr
* or containing a header from some other pool or from another
* version of persistent L2ARC.
*/
ARCSTAT_BUMP(arcstat_l2_rebuild_abort_unsupported);
return (SET_ERROR(ENOTSUP));
}
return (0);
}
/*
* Reads L2ARC log blocks from storage and validates their contents.
*
* This function implements a simple fetcher to make sure that while
* we're processing one buffer the L2ARC is already fetching the next
* one in the chain.
*
* The arguments this_lp and next_lp point to the current and next log block
* address in the block chain. Similarly, this_lb and next_lb hold the
* l2arc_log_blk_phys_t's of the current and next L2ARC blk.
*
* The `this_io' and `next_io' arguments are used for block fetching.
* When issuing the first blk IO during rebuild, you should pass NULL for
* `this_io'. This function will then issue a sync IO to read the block and
* also issue an async IO to fetch the next block in the block chain. The
* fetched IO is returned in `next_io'. On subsequent calls to this
* function, pass the value returned in `next_io' from the previous call
* as `this_io' and a fresh `next_io' pointer to hold the next fetch IO.
* Prior to the call, you should initialize your `next_io' pointer to be
* NULL. If no fetch IO was issued, the pointer is left set at NULL.
*
* On success, this function returns 0, otherwise it returns an appropriate
* error code. On error the fetching IO is aborted and cleared before
* returning from this function. Therefore, if we return `success', the
* caller can assume that we have taken care of cleanup of fetch IOs.
*/
static int
l2arc_log_blk_read(l2arc_dev_t *dev,
const l2arc_log_blkptr_t *this_lbp, const l2arc_log_blkptr_t *next_lbp,
l2arc_log_blk_phys_t *this_lb, l2arc_log_blk_phys_t *next_lb,
zio_t *this_io, zio_t **next_io)
{
int err = 0;
zio_cksum_t cksum;
abd_t *abd = NULL;
uint64_t asize;
ASSERT(this_lbp != NULL && next_lbp != NULL);
ASSERT(this_lb != NULL && next_lb != NULL);
ASSERT(next_io != NULL && *next_io == NULL);
ASSERT(l2arc_log_blkptr_valid(dev, this_lbp));
/*
* Check to see if we have issued the IO for this log block in a
* previous run. If not, this is the first call, so issue it now.
*/
if (this_io == NULL) {
this_io = l2arc_log_blk_fetch(dev->l2ad_vdev, this_lbp,
this_lb);
}
/*
* Peek to see if we can start issuing the next IO immediately.
*/
if (l2arc_log_blkptr_valid(dev, next_lbp)) {
/*
* Start issuing IO for the next log block early - this
* should help keep the L2ARC device busy while we
* decompress and restore this log block.
*/
*next_io = l2arc_log_blk_fetch(dev->l2ad_vdev, next_lbp,
next_lb);
}
/* Wait for the IO to read this log block to complete */
if ((err = zio_wait(this_io)) != 0) {
ARCSTAT_BUMP(arcstat_l2_rebuild_abort_io_errors);
zfs_dbgmsg("L2ARC IO error (%d) while reading log block, "
"offset: %llu, vdev guid: %llu", err,
(u_longlong_t)this_lbp->lbp_daddr,
(u_longlong_t)dev->l2ad_vdev->vdev_guid);
goto cleanup;
}
/*
* Make sure the buffer checks out.
* L2BLK_GET_PSIZE returns aligned size for log blocks.
*/
asize = L2BLK_GET_PSIZE((this_lbp)->lbp_prop);
fletcher_4_native(this_lb, asize, NULL, &cksum);
if (!ZIO_CHECKSUM_EQUAL(cksum, this_lbp->lbp_cksum)) {
ARCSTAT_BUMP(arcstat_l2_rebuild_abort_cksum_lb_errors);
zfs_dbgmsg("L2ARC log block cksum failed, offset: %llu, "
"vdev guid: %llu, l2ad_hand: %llu, l2ad_evict: %llu",
(u_longlong_t)this_lbp->lbp_daddr,
(u_longlong_t)dev->l2ad_vdev->vdev_guid,
(u_longlong_t)dev->l2ad_hand,
(u_longlong_t)dev->l2ad_evict);
err = SET_ERROR(ECKSUM);
goto cleanup;
}
/* Now we can take our time decoding this buffer */
switch (L2BLK_GET_COMPRESS((this_lbp)->lbp_prop)) {
case ZIO_COMPRESS_OFF:
break;
case ZIO_COMPRESS_LZ4:
abd = abd_alloc_for_io(asize, B_TRUE);
abd_copy_from_buf_off(abd, this_lb, 0, asize);
if ((err = zio_decompress_data(
L2BLK_GET_COMPRESS((this_lbp)->lbp_prop),
abd, this_lb, asize, sizeof (*this_lb), NULL)) != 0) {
err = SET_ERROR(EINVAL);
goto cleanup;
}
break;
default:
err = SET_ERROR(EINVAL);
goto cleanup;
}
if (this_lb->lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC))
byteswap_uint64_array(this_lb, sizeof (*this_lb));
if (this_lb->lb_magic != L2ARC_LOG_BLK_MAGIC) {
err = SET_ERROR(EINVAL);
goto cleanup;
}
cleanup:
/* Abort an in-flight fetch I/O in case of error */
if (err != 0 && *next_io != NULL) {
l2arc_log_blk_fetch_abort(*next_io);
*next_io = NULL;
}
if (abd != NULL)
abd_free(abd);
return (err);
}
/*
* Restores the payload of a log block to ARC. This creates empty ARC hdr
* entries which only contain an l2arc hdr, essentially restoring the
* buffers to their L2ARC evicted state. This function also updates space
* usage on the L2ARC vdev to make sure it tracks restored buffers.
*/
static void
l2arc_log_blk_restore(l2arc_dev_t *dev, const l2arc_log_blk_phys_t *lb,
uint64_t lb_asize)
{
uint64_t size = 0, asize = 0;
uint64_t log_entries = dev->l2ad_log_entries;
/*
* Usually arc_adapt() is called only for data, not headers, but
* since we may allocate significant amount of memory here, let ARC
* grow its arc_c.
*/
arc_adapt(log_entries * HDR_L2ONLY_SIZE, arc_l2c_only);
for (int i = log_entries - 1; i >= 0; i--) {
/*
* Restore goes in the reverse temporal direction to preserve
* correct temporal ordering of buffers in the l2ad_buflist.
* l2arc_hdr_restore also does a list_insert_tail instead of
* list_insert_head on the l2ad_buflist:
*
* LIST l2ad_buflist LIST
* HEAD <------ (time) ------ TAIL
* direction +-----+-----+-----+-----+-----+ direction
* of l2arc <== | buf | buf | buf | buf | buf | ===> of rebuild
* fill +-----+-----+-----+-----+-----+
* ^ ^
* | |
* | |
* l2arc_feed_thread l2arc_rebuild
* will place new bufs here restores bufs here
*
* During l2arc_rebuild() the device is not used by
* l2arc_feed_thread() as dev->l2ad_rebuild is set to true.
*/
size += L2BLK_GET_LSIZE((&lb->lb_entries[i])->le_prop);
asize += vdev_psize_to_asize(dev->l2ad_vdev,
L2BLK_GET_PSIZE((&lb->lb_entries[i])->le_prop));
l2arc_hdr_restore(&lb->lb_entries[i], dev);
}
/*
* Record rebuild stats:
* size Logical size of restored buffers in the L2ARC
* asize Aligned size of restored buffers in the L2ARC
*/
ARCSTAT_INCR(arcstat_l2_rebuild_size, size);
ARCSTAT_INCR(arcstat_l2_rebuild_asize, asize);
ARCSTAT_INCR(arcstat_l2_rebuild_bufs, log_entries);
ARCSTAT_F_AVG(arcstat_l2_log_blk_avg_asize, lb_asize);
ARCSTAT_F_AVG(arcstat_l2_data_to_meta_ratio, asize / lb_asize);
ARCSTAT_BUMP(arcstat_l2_rebuild_log_blks);
}
/*
* Restores a single ARC buf hdr from a log entry. The ARC buffer is put
* into a state indicating that it has been evicted to L2ARC.
*/
static void
l2arc_hdr_restore(const l2arc_log_ent_phys_t *le, l2arc_dev_t *dev)
{
arc_buf_hdr_t *hdr, *exists;
kmutex_t *hash_lock;
arc_buf_contents_t type = L2BLK_GET_TYPE((le)->le_prop);
uint64_t asize;
/*
* Do all the allocation before grabbing any locks, this lets us
* sleep if memory is full and we don't have to deal with failed
* allocations.
*/
hdr = arc_buf_alloc_l2only(L2BLK_GET_LSIZE((le)->le_prop), type,
dev, le->le_dva, le->le_daddr,
L2BLK_GET_PSIZE((le)->le_prop), le->le_birth,
L2BLK_GET_COMPRESS((le)->le_prop), le->le_complevel,
L2BLK_GET_PROTECTED((le)->le_prop),
L2BLK_GET_PREFETCH((le)->le_prop),
L2BLK_GET_STATE((le)->le_prop));
asize = vdev_psize_to_asize(dev->l2ad_vdev,
L2BLK_GET_PSIZE((le)->le_prop));
/*
* vdev_space_update() has to be called before arc_hdr_destroy() to
* avoid underflow since the latter also calls vdev_space_update().
*/
l2arc_hdr_arcstats_increment(hdr);
vdev_space_update(dev->l2ad_vdev, asize, 0, 0);
mutex_enter(&dev->l2ad_mtx);
list_insert_tail(&dev->l2ad_buflist, hdr);
(void) zfs_refcount_add_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr);
mutex_exit(&dev->l2ad_mtx);
exists = buf_hash_insert(hdr, &hash_lock);
if (exists) {
/* Buffer was already cached, no need to restore it. */
arc_hdr_destroy(hdr);
/*
* If the buffer is already cached, check whether it has
* L2ARC metadata. If not, enter them and update the flag.
* This is important is case of onlining a cache device, since
* we previously evicted all L2ARC metadata from ARC.
*/
if (!HDR_HAS_L2HDR(exists)) {
arc_hdr_set_flags(exists, ARC_FLAG_HAS_L2HDR);
exists->b_l2hdr.b_dev = dev;
exists->b_l2hdr.b_daddr = le->le_daddr;
exists->b_l2hdr.b_arcs_state =
L2BLK_GET_STATE((le)->le_prop);
mutex_enter(&dev->l2ad_mtx);
list_insert_tail(&dev->l2ad_buflist, exists);
(void) zfs_refcount_add_many(&dev->l2ad_alloc,
arc_hdr_size(exists), exists);
mutex_exit(&dev->l2ad_mtx);
l2arc_hdr_arcstats_increment(exists);
vdev_space_update(dev->l2ad_vdev, asize, 0, 0);
}
ARCSTAT_BUMP(arcstat_l2_rebuild_bufs_precached);
}
mutex_exit(hash_lock);
}
/*
* Starts an asynchronous read IO to read a log block. This is used in log
* block reconstruction to start reading the next block before we are done
* decoding and reconstructing the current block, to keep the l2arc device
* nice and hot with read IO to process.
* The returned zio will contain a newly allocated memory buffers for the IO
* data which should then be freed by the caller once the zio is no longer
* needed (i.e. due to it having completed). If you wish to abort this
* zio, you should do so using l2arc_log_blk_fetch_abort, which takes
* care of disposing of the allocated buffers correctly.
*/
static zio_t *
l2arc_log_blk_fetch(vdev_t *vd, const l2arc_log_blkptr_t *lbp,
l2arc_log_blk_phys_t *lb)
{
uint32_t asize;
zio_t *pio;
l2arc_read_callback_t *cb;
/* L2BLK_GET_PSIZE returns aligned size for log blocks */
asize = L2BLK_GET_PSIZE((lbp)->lbp_prop);
ASSERT(asize <= sizeof (l2arc_log_blk_phys_t));
cb = kmem_zalloc(sizeof (l2arc_read_callback_t), KM_SLEEP);
cb->l2rcb_abd = abd_get_from_buf(lb, asize);
pio = zio_root(vd->vdev_spa, l2arc_blk_fetch_done, cb,
ZIO_FLAG_DONT_CACHE | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE |
ZIO_FLAG_DONT_RETRY);
(void) zio_nowait(zio_read_phys(pio, vd, lbp->lbp_daddr, asize,
cb->l2rcb_abd, ZIO_CHECKSUM_OFF, NULL, NULL,
ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_DONT_CACHE | ZIO_FLAG_CANFAIL |
ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY, B_FALSE));
return (pio);
}
/*
* Aborts a zio returned from l2arc_log_blk_fetch and frees the data
* buffers allocated for it.
*/
static void
l2arc_log_blk_fetch_abort(zio_t *zio)
{
(void) zio_wait(zio);
}
/*
* Creates a zio to update the device header on an l2arc device.
*/
void
l2arc_dev_hdr_update(l2arc_dev_t *dev)
{
l2arc_dev_hdr_phys_t *l2dhdr = dev->l2ad_dev_hdr;
const uint64_t l2dhdr_asize = dev->l2ad_dev_hdr_asize;
abd_t *abd;
int err;
VERIFY(spa_config_held(dev->l2ad_spa, SCL_STATE_ALL, RW_READER));
l2dhdr->dh_magic = L2ARC_DEV_HDR_MAGIC;
l2dhdr->dh_version = L2ARC_PERSISTENT_VERSION;
l2dhdr->dh_spa_guid = spa_guid(dev->l2ad_vdev->vdev_spa);
l2dhdr->dh_vdev_guid = dev->l2ad_vdev->vdev_guid;
l2dhdr->dh_log_entries = dev->l2ad_log_entries;
l2dhdr->dh_evict = dev->l2ad_evict;
l2dhdr->dh_start = dev->l2ad_start;
l2dhdr->dh_end = dev->l2ad_end;
l2dhdr->dh_lb_asize = zfs_refcount_count(&dev->l2ad_lb_asize);
l2dhdr->dh_lb_count = zfs_refcount_count(&dev->l2ad_lb_count);
l2dhdr->dh_flags = 0;
l2dhdr->dh_trim_action_time = dev->l2ad_vdev->vdev_trim_action_time;
l2dhdr->dh_trim_state = dev->l2ad_vdev->vdev_trim_state;
if (dev->l2ad_first)
l2dhdr->dh_flags |= L2ARC_DEV_HDR_EVICT_FIRST;
abd = abd_get_from_buf(l2dhdr, l2dhdr_asize);
err = zio_wait(zio_write_phys(NULL, dev->l2ad_vdev,
VDEV_LABEL_START_SIZE, l2dhdr_asize, abd, ZIO_CHECKSUM_LABEL, NULL,
NULL, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_CANFAIL, B_FALSE));
abd_free(abd);
if (err != 0) {
zfs_dbgmsg("L2ARC IO error (%d) while writing device header, "
"vdev guid: %llu", err,
(u_longlong_t)dev->l2ad_vdev->vdev_guid);
}
}
/*
* Commits a log block to the L2ARC device. This routine is invoked from
* l2arc_write_buffers when the log block fills up.
* This function allocates some memory to temporarily hold the serialized
* buffer to be written. This is then released in l2arc_write_done.
*/
static void
l2arc_log_blk_commit(l2arc_dev_t *dev, zio_t *pio, l2arc_write_callback_t *cb)
{
l2arc_log_blk_phys_t *lb = &dev->l2ad_log_blk;
l2arc_dev_hdr_phys_t *l2dhdr = dev->l2ad_dev_hdr;
uint64_t psize, asize;
zio_t *wzio;
l2arc_lb_abd_buf_t *abd_buf;
uint8_t *tmpbuf;
l2arc_lb_ptr_buf_t *lb_ptr_buf;
VERIFY3S(dev->l2ad_log_ent_idx, ==, dev->l2ad_log_entries);
tmpbuf = zio_buf_alloc(sizeof (*lb));
abd_buf = zio_buf_alloc(sizeof (*abd_buf));
abd_buf->abd = abd_get_from_buf(lb, sizeof (*lb));
lb_ptr_buf = kmem_zalloc(sizeof (l2arc_lb_ptr_buf_t), KM_SLEEP);
lb_ptr_buf->lb_ptr = kmem_zalloc(sizeof (l2arc_log_blkptr_t), KM_SLEEP);
/* link the buffer into the block chain */
lb->lb_prev_lbp = l2dhdr->dh_start_lbps[1];
lb->lb_magic = L2ARC_LOG_BLK_MAGIC;
/*
* l2arc_log_blk_commit() may be called multiple times during a single
* l2arc_write_buffers() call. Save the allocated abd buffers in a list
* so we can free them in l2arc_write_done() later on.
*/
list_insert_tail(&cb->l2wcb_abd_list, abd_buf);
/* try to compress the buffer */
psize = zio_compress_data(ZIO_COMPRESS_LZ4,
abd_buf->abd, tmpbuf, sizeof (*lb), 0);
/* a log block is never entirely zero */
ASSERT(psize != 0);
asize = vdev_psize_to_asize(dev->l2ad_vdev, psize);
ASSERT(asize <= sizeof (*lb));
/*
* Update the start log block pointer in the device header to point
* to the log block we're about to write.
*/
l2dhdr->dh_start_lbps[1] = l2dhdr->dh_start_lbps[0];
l2dhdr->dh_start_lbps[0].lbp_daddr = dev->l2ad_hand;
l2dhdr->dh_start_lbps[0].lbp_payload_asize =
dev->l2ad_log_blk_payload_asize;
l2dhdr->dh_start_lbps[0].lbp_payload_start =
dev->l2ad_log_blk_payload_start;
_NOTE(CONSTCOND)
L2BLK_SET_LSIZE(
(&l2dhdr->dh_start_lbps[0])->lbp_prop, sizeof (*lb));
L2BLK_SET_PSIZE(
(&l2dhdr->dh_start_lbps[0])->lbp_prop, asize);
L2BLK_SET_CHECKSUM(
(&l2dhdr->dh_start_lbps[0])->lbp_prop,
ZIO_CHECKSUM_FLETCHER_4);
if (asize < sizeof (*lb)) {
/* compression succeeded */
bzero(tmpbuf + psize, asize - psize);
L2BLK_SET_COMPRESS(
(&l2dhdr->dh_start_lbps[0])->lbp_prop,
ZIO_COMPRESS_LZ4);
} else {
/* compression failed */
bcopy(lb, tmpbuf, sizeof (*lb));
L2BLK_SET_COMPRESS(
(&l2dhdr->dh_start_lbps[0])->lbp_prop,
ZIO_COMPRESS_OFF);
}
/* checksum what we're about to write */
fletcher_4_native(tmpbuf, asize, NULL,
&l2dhdr->dh_start_lbps[0].lbp_cksum);
abd_free(abd_buf->abd);
/* perform the write itself */
abd_buf->abd = abd_get_from_buf(tmpbuf, sizeof (*lb));
abd_take_ownership_of_buf(abd_buf->abd, B_TRUE);
wzio = zio_write_phys(pio, dev->l2ad_vdev, dev->l2ad_hand,
asize, abd_buf->abd, ZIO_CHECKSUM_OFF, NULL, NULL,
ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_CANFAIL, B_FALSE);
DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev, zio_t *, wzio);
(void) zio_nowait(wzio);
dev->l2ad_hand += asize;
/*
* Include the committed log block's pointer in the list of pointers
* to log blocks present in the L2ARC device.
*/
bcopy(&l2dhdr->dh_start_lbps[0], lb_ptr_buf->lb_ptr,
sizeof (l2arc_log_blkptr_t));
mutex_enter(&dev->l2ad_mtx);
list_insert_head(&dev->l2ad_lbptr_list, lb_ptr_buf);
ARCSTAT_INCR(arcstat_l2_log_blk_asize, asize);
ARCSTAT_BUMP(arcstat_l2_log_blk_count);
zfs_refcount_add_many(&dev->l2ad_lb_asize, asize, lb_ptr_buf);
zfs_refcount_add(&dev->l2ad_lb_count, lb_ptr_buf);
mutex_exit(&dev->l2ad_mtx);
vdev_space_update(dev->l2ad_vdev, asize, 0, 0);
/* bump the kstats */
ARCSTAT_INCR(arcstat_l2_write_bytes, asize);
ARCSTAT_BUMP(arcstat_l2_log_blk_writes);
ARCSTAT_F_AVG(arcstat_l2_log_blk_avg_asize, asize);
ARCSTAT_F_AVG(arcstat_l2_data_to_meta_ratio,
dev->l2ad_log_blk_payload_asize / asize);
/* start a new log block */
dev->l2ad_log_ent_idx = 0;
dev->l2ad_log_blk_payload_asize = 0;
dev->l2ad_log_blk_payload_start = 0;
}
/*
* Validates an L2ARC log block address to make sure that it can be read
* from the provided L2ARC device.
*/
boolean_t
l2arc_log_blkptr_valid(l2arc_dev_t *dev, const l2arc_log_blkptr_t *lbp)
{
/* L2BLK_GET_PSIZE returns aligned size for log blocks */
uint64_t asize = L2BLK_GET_PSIZE((lbp)->lbp_prop);
uint64_t end = lbp->lbp_daddr + asize - 1;
uint64_t start = lbp->lbp_payload_start;
boolean_t evicted = B_FALSE;
/*
* A log block is valid if all of the following conditions are true:
* - it fits entirely (including its payload) between l2ad_start and
* l2ad_end
* - it has a valid size
* - neither the log block itself nor part of its payload was evicted
* by l2arc_evict():
*
* l2ad_hand l2ad_evict
* | | lbp_daddr
* | start | | end
* | | | | |
* V V V V V
* l2ad_start ============================================ l2ad_end
* --------------------------||||
* ^ ^
* | log block
* payload
*/
evicted =
l2arc_range_check_overlap(start, end, dev->l2ad_hand) ||
l2arc_range_check_overlap(start, end, dev->l2ad_evict) ||
l2arc_range_check_overlap(dev->l2ad_hand, dev->l2ad_evict, start) ||
l2arc_range_check_overlap(dev->l2ad_hand, dev->l2ad_evict, end);
return (start >= dev->l2ad_start && end <= dev->l2ad_end &&
asize > 0 && asize <= sizeof (l2arc_log_blk_phys_t) &&
(!evicted || dev->l2ad_first));
}
/*
* Inserts ARC buffer header `hdr' into the current L2ARC log block on
* the device. The buffer being inserted must be present in L2ARC.
* Returns B_TRUE if the L2ARC log block is full and needs to be committed
* to L2ARC, or B_FALSE if it still has room for more ARC buffers.
*/
static boolean_t
l2arc_log_blk_insert(l2arc_dev_t *dev, const arc_buf_hdr_t *hdr)
{
l2arc_log_blk_phys_t *lb = &dev->l2ad_log_blk;
l2arc_log_ent_phys_t *le;
if (dev->l2ad_log_entries == 0)
return (B_FALSE);
int index = dev->l2ad_log_ent_idx++;
ASSERT3S(index, <, dev->l2ad_log_entries);
ASSERT(HDR_HAS_L2HDR(hdr));
le = &lb->lb_entries[index];
bzero(le, sizeof (*le));
le->le_dva = hdr->b_dva;
le->le_birth = hdr->b_birth;
le->le_daddr = hdr->b_l2hdr.b_daddr;
if (index == 0)
dev->l2ad_log_blk_payload_start = le->le_daddr;
L2BLK_SET_LSIZE((le)->le_prop, HDR_GET_LSIZE(hdr));
L2BLK_SET_PSIZE((le)->le_prop, HDR_GET_PSIZE(hdr));
L2BLK_SET_COMPRESS((le)->le_prop, HDR_GET_COMPRESS(hdr));
le->le_complevel = hdr->b_complevel;
L2BLK_SET_TYPE((le)->le_prop, hdr->b_type);
L2BLK_SET_PROTECTED((le)->le_prop, !!(HDR_PROTECTED(hdr)));
L2BLK_SET_PREFETCH((le)->le_prop, !!(HDR_PREFETCH(hdr)));
L2BLK_SET_STATE((le)->le_prop, hdr->b_l1hdr.b_state->arcs_state);
dev->l2ad_log_blk_payload_asize += vdev_psize_to_asize(dev->l2ad_vdev,
HDR_GET_PSIZE(hdr));
return (dev->l2ad_log_ent_idx == dev->l2ad_log_entries);
}
/*
* Checks whether a given L2ARC device address sits in a time-sequential
* range. The trick here is that the L2ARC is a rotary buffer, so we can't
* just do a range comparison, we need to handle the situation in which the
* range wraps around the end of the L2ARC device. Arguments:
* bottom -- Lower end of the range to check (written to earlier).
* top -- Upper end of the range to check (written to later).
* check -- The address for which we want to determine if it sits in
* between the top and bottom.
*
* The 3-way conditional below represents the following cases:
*
* bottom < top : Sequentially ordered case:
* --------+-------------------+
* | (overlap here?) |
* L2ARC dev V V
* |---------------============--------------|
*
* bottom > top: Looped-around case:
* --------+------------------+
* | (overlap here?) |
* L2ARC dev V V
* |===============---------------===========|
* ^ ^
* | (or here?) |
* +---------------+---------
*
* top == bottom : Just a single address comparison.
*/
boolean_t
l2arc_range_check_overlap(uint64_t bottom, uint64_t top, uint64_t check)
{
if (bottom < top)
return (bottom <= check && check <= top);
else if (bottom > top)
return (check <= top || bottom <= check);
else
return (check == top);
}
EXPORT_SYMBOL(arc_buf_size);
EXPORT_SYMBOL(arc_write);
EXPORT_SYMBOL(arc_read);
EXPORT_SYMBOL(arc_buf_info);
EXPORT_SYMBOL(arc_getbuf_func);
EXPORT_SYMBOL(arc_add_prune_callback);
EXPORT_SYMBOL(arc_remove_prune_callback);
/* BEGIN CSTYLED */
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, min, param_set_arc_long,
param_get_long, ZMOD_RW, "Min arc size");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, max, param_set_arc_long,
param_get_long, ZMOD_RW, "Max arc size");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, meta_limit, param_set_arc_long,
param_get_long, ZMOD_RW, "Metadata limit for arc size");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, meta_limit_percent,
param_set_arc_long, param_get_long, ZMOD_RW,
"Percent of arc size for arc meta limit");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, meta_min, param_set_arc_long,
param_get_long, ZMOD_RW, "Min arc metadata");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_prune, INT, ZMOD_RW,
"Meta objects to scan for prune");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_adjust_restarts, INT, ZMOD_RW,
"Limit number of restarts in arc_evict_meta");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_strategy, INT, ZMOD_RW,
"Meta reclaim strategy");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, grow_retry, param_set_arc_int,
param_get_int, ZMOD_RW, "Seconds before growing arc size");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, p_dampener_disable, INT, ZMOD_RW,
"Disable arc_p adapt dampener");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, shrink_shift, param_set_arc_int,
param_get_int, ZMOD_RW, "log2(fraction of arc to reclaim)");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, pc_percent, UINT, ZMOD_RW,
"Percent of pagecache to reclaim arc to");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, p_min_shift, param_set_arc_int,
param_get_int, ZMOD_RW, "arc_c shift to calc min/max arc_p");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, average_blocksize, INT, ZMOD_RD,
"Target average block size");
ZFS_MODULE_PARAM(zfs, zfs_, compressed_arc_enabled, INT, ZMOD_RW,
"Disable compressed arc buffers");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, min_prefetch_ms, param_set_arc_int,
param_get_int, ZMOD_RW, "Min life of prefetch block in ms");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, min_prescient_prefetch_ms,
param_set_arc_int, param_get_int, ZMOD_RW,
"Min life of prescient prefetched block in ms");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, write_max, ULONG, ZMOD_RW,
"Max write bytes per interval");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, write_boost, ULONG, ZMOD_RW,
"Extra write bytes during device warmup");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, headroom, ULONG, ZMOD_RW,
"Number of max device writes to precache");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, headroom_boost, ULONG, ZMOD_RW,
"Compressed l2arc_headroom multiplier");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, trim_ahead, ULONG, ZMOD_RW,
"TRIM ahead L2ARC write size multiplier");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, feed_secs, ULONG, ZMOD_RW,
"Seconds between L2ARC writing");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, feed_min_ms, ULONG, ZMOD_RW,
"Min feed interval in milliseconds");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, noprefetch, INT, ZMOD_RW,
"Skip caching prefetched buffers");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, feed_again, INT, ZMOD_RW,
"Turbo L2ARC warmup");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, norw, INT, ZMOD_RW,
"No reads during writes");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, meta_percent, INT, ZMOD_RW,
"Percent of ARC size allowed for L2ARC-only headers");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, rebuild_enabled, INT, ZMOD_RW,
"Rebuild the L2ARC when importing a pool");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, rebuild_blocks_min_l2size, ULONG, ZMOD_RW,
"Min size in bytes to write rebuild log blocks in L2ARC");
ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, mfuonly, INT, ZMOD_RW,
"Cache only MFU data from ARC into L2ARC");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, lotsfree_percent, param_set_arc_int,
param_get_int, ZMOD_RW, "System free memory I/O throttle in bytes");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, sys_free, param_set_arc_long,
param_get_long, ZMOD_RW, "System free memory target size in bytes");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, dnode_limit, param_set_arc_long,
param_get_long, ZMOD_RW, "Minimum bytes of dnodes in arc");
ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, dnode_limit_percent,
param_set_arc_long, param_get_long, ZMOD_RW,
"Percent of ARC meta buffers for dnodes");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, dnode_reduce_percent, ULONG, ZMOD_RW,
"Percentage of excess dnodes to try to unpin");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, eviction_pct, INT, ZMOD_RW,
"When full, ARC allocation waits for eviction of this % of alloc size");
ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, evict_batch_limit, INT, ZMOD_RW,
"The number of headers to evict per sublist before moving to the next");
/* END CSTYLED */
diff --git a/sys/contrib/openzfs/module/zfs/dbuf.c b/sys/contrib/openzfs/module/zfs/dbuf.c
index f9bcd9313f0a..9ce091b80dcb 100644
--- a/sys/contrib/openzfs/module/zfs/dbuf.c
+++ b/sys/contrib/openzfs/module/zfs/dbuf.c
@@ -1,5053 +1,5054 @@
/*
* 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 2011 Nexenta Systems, Inc. All rights reserved.
* Copyright (c) 2012, 2020 by Delphix. All rights reserved.
* Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
* Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
* Copyright (c) 2019, Klara Inc.
* Copyright (c) 2019, Allan Jude
*/
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
kstat_t *dbuf_ksp;
typedef struct dbuf_stats {
/*
* Various statistics about the size of the dbuf cache.
*/
kstat_named_t cache_count;
kstat_named_t cache_size_bytes;
kstat_named_t cache_size_bytes_max;
/*
* Statistics regarding the bounds on the dbuf cache size.
*/
kstat_named_t cache_target_bytes;
kstat_named_t cache_lowater_bytes;
kstat_named_t cache_hiwater_bytes;
/*
* Total number of dbuf cache evictions that have occurred.
*/
kstat_named_t cache_total_evicts;
/*
* The distribution of dbuf levels in the dbuf cache and
* the total size of all dbufs at each level.
*/
kstat_named_t cache_levels[DN_MAX_LEVELS];
kstat_named_t cache_levels_bytes[DN_MAX_LEVELS];
/*
* Statistics about the dbuf hash table.
*/
kstat_named_t hash_hits;
kstat_named_t hash_misses;
kstat_named_t hash_collisions;
kstat_named_t hash_elements;
kstat_named_t hash_elements_max;
/*
* Number of sublists containing more than one dbuf in the dbuf
* hash table. Keep track of the longest hash chain.
*/
kstat_named_t hash_chains;
kstat_named_t hash_chain_max;
/*
* Number of times a dbuf_create() discovers that a dbuf was
* already created and in the dbuf hash table.
*/
kstat_named_t hash_insert_race;
/*
* Statistics about the size of the metadata dbuf cache.
*/
kstat_named_t metadata_cache_count;
kstat_named_t metadata_cache_size_bytes;
kstat_named_t metadata_cache_size_bytes_max;
/*
* For diagnostic purposes, this is incremented whenever we can't add
* something to the metadata cache because it's full, and instead put
* the data in the regular dbuf cache.
*/
kstat_named_t metadata_cache_overflow;
} dbuf_stats_t;
dbuf_stats_t dbuf_stats = {
{ "cache_count", KSTAT_DATA_UINT64 },
{ "cache_size_bytes", KSTAT_DATA_UINT64 },
{ "cache_size_bytes_max", KSTAT_DATA_UINT64 },
{ "cache_target_bytes", KSTAT_DATA_UINT64 },
{ "cache_lowater_bytes", KSTAT_DATA_UINT64 },
{ "cache_hiwater_bytes", KSTAT_DATA_UINT64 },
{ "cache_total_evicts", KSTAT_DATA_UINT64 },
{ { "cache_levels_N", KSTAT_DATA_UINT64 } },
{ { "cache_levels_bytes_N", KSTAT_DATA_UINT64 } },
{ "hash_hits", KSTAT_DATA_UINT64 },
{ "hash_misses", KSTAT_DATA_UINT64 },
{ "hash_collisions", KSTAT_DATA_UINT64 },
{ "hash_elements", KSTAT_DATA_UINT64 },
{ "hash_elements_max", KSTAT_DATA_UINT64 },
{ "hash_chains", KSTAT_DATA_UINT64 },
{ "hash_chain_max", KSTAT_DATA_UINT64 },
{ "hash_insert_race", KSTAT_DATA_UINT64 },
{ "metadata_cache_count", KSTAT_DATA_UINT64 },
{ "metadata_cache_size_bytes", KSTAT_DATA_UINT64 },
{ "metadata_cache_size_bytes_max", KSTAT_DATA_UINT64 },
{ "metadata_cache_overflow", KSTAT_DATA_UINT64 }
};
struct {
wmsum_t cache_count;
wmsum_t cache_total_evicts;
wmsum_t cache_levels[DN_MAX_LEVELS];
wmsum_t cache_levels_bytes[DN_MAX_LEVELS];
wmsum_t hash_hits;
wmsum_t hash_misses;
wmsum_t hash_collisions;
wmsum_t hash_chains;
wmsum_t hash_insert_race;
wmsum_t metadata_cache_count;
wmsum_t metadata_cache_overflow;
} dbuf_sums;
#define DBUF_STAT_INCR(stat, val) \
wmsum_add(&dbuf_sums.stat, val);
#define DBUF_STAT_DECR(stat, val) \
DBUF_STAT_INCR(stat, -(val));
#define DBUF_STAT_BUMP(stat) \
DBUF_STAT_INCR(stat, 1);
#define DBUF_STAT_BUMPDOWN(stat) \
DBUF_STAT_INCR(stat, -1);
#define DBUF_STAT_MAX(stat, v) { \
uint64_t _m; \
while ((v) > (_m = dbuf_stats.stat.value.ui64) && \
(_m != atomic_cas_64(&dbuf_stats.stat.value.ui64, _m, (v))))\
continue; \
}
static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx);
static void dbuf_sync_leaf_verify_bonus_dnode(dbuf_dirty_record_t *dr);
static int dbuf_read_verify_dnode_crypt(dmu_buf_impl_t *db, uint32_t flags);
extern inline void dmu_buf_init_user(dmu_buf_user_t *dbu,
dmu_buf_evict_func_t *evict_func_sync,
dmu_buf_evict_func_t *evict_func_async,
dmu_buf_t **clear_on_evict_dbufp);
/*
* Global data structures and functions for the dbuf cache.
*/
static kmem_cache_t *dbuf_kmem_cache;
static taskq_t *dbu_evict_taskq;
static kthread_t *dbuf_cache_evict_thread;
static kmutex_t dbuf_evict_lock;
static kcondvar_t dbuf_evict_cv;
static boolean_t dbuf_evict_thread_exit;
/*
* There are two dbuf caches; each dbuf can only be in one of them at a time.
*
* 1. Cache of metadata dbufs, to help make read-heavy administrative commands
* from /sbin/zfs run faster. The "metadata cache" specifically stores dbufs
* that represent the metadata that describes filesystems/snapshots/
* bookmarks/properties/etc. We only evict from this cache when we export a
* pool, to short-circuit as much I/O as possible for all administrative
* commands that need the metadata. There is no eviction policy for this
* cache, because we try to only include types in it which would occupy a
* very small amount of space per object but create a large impact on the
* performance of these commands. Instead, after it reaches a maximum size
* (which should only happen on very small memory systems with a very large
* number of filesystem objects), we stop taking new dbufs into the
* metadata cache, instead putting them in the normal dbuf cache.
*
* 2. LRU cache of dbufs. The dbuf cache maintains a list of dbufs that
* are not currently held but have been recently released. These dbufs
* are not eligible for arc eviction until they are aged out of the cache.
* Dbufs that are aged out of the cache will be immediately destroyed and
* become eligible for arc eviction.
*
* Dbufs are added to these caches once the last hold is released. If a dbuf is
* later accessed and still exists in the dbuf cache, then it will be removed
* from the cache and later re-added to the head of the cache.
*
* If a given dbuf meets the requirements for the metadata cache, it will go
* there, otherwise it will be considered for the generic LRU dbuf cache. The
* caches and the refcounts tracking their sizes are stored in an array indexed
* by those caches' matching enum values (from dbuf_cached_state_t).
*/
typedef struct dbuf_cache {
multilist_t cache;
zfs_refcount_t size ____cacheline_aligned;
} dbuf_cache_t;
dbuf_cache_t dbuf_caches[DB_CACHE_MAX];
/* Size limits for the caches */
unsigned long dbuf_cache_max_bytes = ULONG_MAX;
unsigned long dbuf_metadata_cache_max_bytes = ULONG_MAX;
/* Set the default sizes of the caches to log2 fraction of arc size */
int dbuf_cache_shift = 5;
int dbuf_metadata_cache_shift = 6;
static unsigned long dbuf_cache_target_bytes(void);
static unsigned long dbuf_metadata_cache_target_bytes(void);
/*
* The LRU dbuf cache uses a three-stage eviction policy:
* - A low water marker designates when the dbuf eviction thread
* should stop evicting from the dbuf cache.
* - When we reach the maximum size (aka mid water mark), we
* signal the eviction thread to run.
* - The high water mark indicates when the eviction thread
* is unable to keep up with the incoming load and eviction must
* happen in the context of the calling thread.
*
* The dbuf cache:
* (max size)
* low water mid water hi water
* +----------------------------------------+----------+----------+
* | | | |
* | | | |
* | | | |
* | | | |
* +----------------------------------------+----------+----------+
* stop signal evict
* evicting eviction directly
* thread
*
* The high and low water marks indicate the operating range for the eviction
* thread. The low water mark is, by default, 90% of the total size of the
* cache and the high water mark is at 110% (both of these percentages can be
* changed by setting dbuf_cache_lowater_pct and dbuf_cache_hiwater_pct,
* respectively). The eviction thread will try to ensure that the cache remains
* within this range by waking up every second and checking if the cache is
* above the low water mark. The thread can also be woken up by callers adding
* elements into the cache if the cache is larger than the mid water (i.e max
* cache size). Once the eviction thread is woken up and eviction is required,
* it will continue evicting buffers until it's able to reduce the cache size
* to the low water mark. If the cache size continues to grow and hits the high
* water mark, then callers adding elements to the cache will begin to evict
* directly from the cache until the cache is no longer above the high water
* mark.
*/
/*
* The percentage above and below the maximum cache size.
*/
uint_t dbuf_cache_hiwater_pct = 10;
uint_t dbuf_cache_lowater_pct = 10;
/* ARGSUSED */
static int
dbuf_cons(void *vdb, void *unused, int kmflag)
{
dmu_buf_impl_t *db = vdb;
bzero(db, sizeof (dmu_buf_impl_t));
mutex_init(&db->db_mtx, NULL, MUTEX_DEFAULT, NULL);
rw_init(&db->db_rwlock, NULL, RW_DEFAULT, NULL);
cv_init(&db->db_changed, NULL, CV_DEFAULT, NULL);
multilist_link_init(&db->db_cache_link);
zfs_refcount_create(&db->db_holds);
return (0);
}
/* ARGSUSED */
static void
dbuf_dest(void *vdb, void *unused)
{
dmu_buf_impl_t *db = vdb;
mutex_destroy(&db->db_mtx);
rw_destroy(&db->db_rwlock);
cv_destroy(&db->db_changed);
ASSERT(!multilist_link_active(&db->db_cache_link));
zfs_refcount_destroy(&db->db_holds);
}
/*
* dbuf hash table routines
*/
static dbuf_hash_table_t dbuf_hash_table;
/*
* We use Cityhash for this. It's fast, and has good hash properties without
* requiring any large static buffers.
*/
static uint64_t
dbuf_hash(void *os, uint64_t obj, uint8_t lvl, uint64_t blkid)
{
return (cityhash4((uintptr_t)os, obj, (uint64_t)lvl, blkid));
}
#define DTRACE_SET_STATE(db, why) \
DTRACE_PROBE2(dbuf__state_change, dmu_buf_impl_t *, db, \
const char *, why)
#define DBUF_EQUAL(dbuf, os, obj, level, blkid) \
((dbuf)->db.db_object == (obj) && \
(dbuf)->db_objset == (os) && \
(dbuf)->db_level == (level) && \
(dbuf)->db_blkid == (blkid))
dmu_buf_impl_t *
dbuf_find(objset_t *os, uint64_t obj, uint8_t level, uint64_t blkid)
{
dbuf_hash_table_t *h = &dbuf_hash_table;
uint64_t hv;
uint64_t idx;
dmu_buf_impl_t *db;
hv = dbuf_hash(os, obj, level, blkid);
idx = hv & h->hash_table_mask;
mutex_enter(DBUF_HASH_MUTEX(h, idx));
for (db = h->hash_table[idx]; db != NULL; db = db->db_hash_next) {
if (DBUF_EQUAL(db, os, obj, level, blkid)) {
mutex_enter(&db->db_mtx);
if (db->db_state != DB_EVICTING) {
mutex_exit(DBUF_HASH_MUTEX(h, idx));
return (db);
}
mutex_exit(&db->db_mtx);
}
}
mutex_exit(DBUF_HASH_MUTEX(h, idx));
return (NULL);
}
static dmu_buf_impl_t *
dbuf_find_bonus(objset_t *os, uint64_t object)
{
dnode_t *dn;
dmu_buf_impl_t *db = NULL;
if (dnode_hold(os, object, FTAG, &dn) == 0) {
rw_enter(&dn->dn_struct_rwlock, RW_READER);
if (dn->dn_bonus != NULL) {
db = dn->dn_bonus;
mutex_enter(&db->db_mtx);
}
rw_exit(&dn->dn_struct_rwlock);
dnode_rele(dn, FTAG);
}
return (db);
}
/*
* Insert an entry into the hash table. If there is already an element
* equal to elem in the hash table, then the already existing element
* will be returned and the new element will not be inserted.
* Otherwise returns NULL.
*/
static dmu_buf_impl_t *
dbuf_hash_insert(dmu_buf_impl_t *db)
{
dbuf_hash_table_t *h = &dbuf_hash_table;
objset_t *os = db->db_objset;
uint64_t obj = db->db.db_object;
int level = db->db_level;
uint64_t blkid, hv, idx;
dmu_buf_impl_t *dbf;
uint32_t i;
blkid = db->db_blkid;
hv = dbuf_hash(os, obj, level, blkid);
idx = hv & h->hash_table_mask;
mutex_enter(DBUF_HASH_MUTEX(h, idx));
for (dbf = h->hash_table[idx], i = 0; dbf != NULL;
dbf = dbf->db_hash_next, i++) {
if (DBUF_EQUAL(dbf, os, obj, level, blkid)) {
mutex_enter(&dbf->db_mtx);
if (dbf->db_state != DB_EVICTING) {
mutex_exit(DBUF_HASH_MUTEX(h, idx));
return (dbf);
}
mutex_exit(&dbf->db_mtx);
}
}
if (i > 0) {
DBUF_STAT_BUMP(hash_collisions);
if (i == 1)
DBUF_STAT_BUMP(hash_chains);
DBUF_STAT_MAX(hash_chain_max, i);
}
mutex_enter(&db->db_mtx);
db->db_hash_next = h->hash_table[idx];
h->hash_table[idx] = db;
mutex_exit(DBUF_HASH_MUTEX(h, idx));
uint64_t he = atomic_inc_64_nv(&dbuf_stats.hash_elements.value.ui64);
DBUF_STAT_MAX(hash_elements_max, he);
return (NULL);
}
/*
* This returns whether this dbuf should be stored in the metadata cache, which
* is based on whether it's from one of the dnode types that store data related
* to traversing dataset hierarchies.
*/
static boolean_t
dbuf_include_in_metadata_cache(dmu_buf_impl_t *db)
{
DB_DNODE_ENTER(db);
dmu_object_type_t type = DB_DNODE(db)->dn_type;
DB_DNODE_EXIT(db);
/* Check if this dbuf is one of the types we care about */
if (DMU_OT_IS_METADATA_CACHED(type)) {
/* If we hit this, then we set something up wrong in dmu_ot */
ASSERT(DMU_OT_IS_METADATA(type));
/*
* Sanity check for small-memory systems: don't allocate too
* much memory for this purpose.
*/
if (zfs_refcount_count(
&dbuf_caches[DB_DBUF_METADATA_CACHE].size) >
dbuf_metadata_cache_target_bytes()) {
DBUF_STAT_BUMP(metadata_cache_overflow);
return (B_FALSE);
}
return (B_TRUE);
}
return (B_FALSE);
}
/*
* Remove an entry from the hash table. It must be in the EVICTING state.
*/
static void
dbuf_hash_remove(dmu_buf_impl_t *db)
{
dbuf_hash_table_t *h = &dbuf_hash_table;
uint64_t hv, idx;
dmu_buf_impl_t *dbf, **dbp;
hv = dbuf_hash(db->db_objset, db->db.db_object,
db->db_level, db->db_blkid);
idx = hv & h->hash_table_mask;
/*
* We mustn't hold db_mtx to maintain lock ordering:
* DBUF_HASH_MUTEX > db_mtx.
*/
ASSERT(zfs_refcount_is_zero(&db->db_holds));
ASSERT(db->db_state == DB_EVICTING);
ASSERT(!MUTEX_HELD(&db->db_mtx));
mutex_enter(DBUF_HASH_MUTEX(h, idx));
dbp = &h->hash_table[idx];
while ((dbf = *dbp) != db) {
dbp = &dbf->db_hash_next;
ASSERT(dbf != NULL);
}
*dbp = db->db_hash_next;
db->db_hash_next = NULL;
if (h->hash_table[idx] &&
h->hash_table[idx]->db_hash_next == NULL)
DBUF_STAT_BUMPDOWN(hash_chains);
mutex_exit(DBUF_HASH_MUTEX(h, idx));
atomic_dec_64(&dbuf_stats.hash_elements.value.ui64);
}
typedef enum {
DBVU_EVICTING,
DBVU_NOT_EVICTING
} dbvu_verify_type_t;
static void
dbuf_verify_user(dmu_buf_impl_t *db, dbvu_verify_type_t verify_type)
{
#ifdef ZFS_DEBUG
int64_t holds;
if (db->db_user == NULL)
return;
/* Only data blocks support the attachment of user data. */
ASSERT(db->db_level == 0);
/* Clients must resolve a dbuf before attaching user data. */
ASSERT(db->db.db_data != NULL);
ASSERT3U(db->db_state, ==, DB_CACHED);
holds = zfs_refcount_count(&db->db_holds);
if (verify_type == DBVU_EVICTING) {
/*
* Immediate eviction occurs when holds == dirtycnt.
* For normal eviction buffers, holds is zero on
* eviction, except when dbuf_fix_old_data() calls
* dbuf_clear_data(). However, the hold count can grow
* during eviction even though db_mtx is held (see
* dmu_bonus_hold() for an example), so we can only
* test the generic invariant that holds >= dirtycnt.
*/
ASSERT3U(holds, >=, db->db_dirtycnt);
} else {
if (db->db_user_immediate_evict == TRUE)
ASSERT3U(holds, >=, db->db_dirtycnt);
else
ASSERT3U(holds, >, 0);
}
#endif
}
static void
dbuf_evict_user(dmu_buf_impl_t *db)
{
dmu_buf_user_t *dbu = db->db_user;
ASSERT(MUTEX_HELD(&db->db_mtx));
if (dbu == NULL)
return;
dbuf_verify_user(db, DBVU_EVICTING);
db->db_user = NULL;
#ifdef ZFS_DEBUG
if (dbu->dbu_clear_on_evict_dbufp != NULL)
*dbu->dbu_clear_on_evict_dbufp = NULL;
#endif
/*
* There are two eviction callbacks - one that we call synchronously
* and one that we invoke via a taskq. The async one is useful for
* avoiding lock order reversals and limiting stack depth.
*
* Note that if we have a sync callback but no async callback,
* it's likely that the sync callback will free the structure
* containing the dbu. In that case we need to take care to not
* dereference dbu after calling the sync evict func.
*/
boolean_t has_async = (dbu->dbu_evict_func_async != NULL);
if (dbu->dbu_evict_func_sync != NULL)
dbu->dbu_evict_func_sync(dbu);
if (has_async) {
taskq_dispatch_ent(dbu_evict_taskq, dbu->dbu_evict_func_async,
dbu, 0, &dbu->dbu_tqent);
}
}
boolean_t
dbuf_is_metadata(dmu_buf_impl_t *db)
{
/*
* Consider indirect blocks and spill blocks to be meta data.
*/
if (db->db_level > 0 || db->db_blkid == DMU_SPILL_BLKID) {
return (B_TRUE);
} else {
boolean_t is_metadata;
DB_DNODE_ENTER(db);
is_metadata = DMU_OT_IS_METADATA(DB_DNODE(db)->dn_type);
DB_DNODE_EXIT(db);
return (is_metadata);
}
}
/*
* This function *must* return indices evenly distributed between all
* sublists of the multilist. This is needed due to how the dbuf eviction
* code is laid out; dbuf_evict_thread() assumes dbufs are evenly
* distributed between all sublists and uses this assumption when
* deciding which sublist to evict from and how much to evict from it.
*/
static unsigned int
dbuf_cache_multilist_index_func(multilist_t *ml, void *obj)
{
dmu_buf_impl_t *db = obj;
/*
* The assumption here, is the hash value for a given
* dmu_buf_impl_t will remain constant throughout it's lifetime
* (i.e. it's objset, object, level and blkid fields don't change).
* Thus, we don't need to store the dbuf's sublist index
* on insertion, as this index can be recalculated on removal.
*
* Also, the low order bits of the hash value are thought to be
* distributed evenly. Otherwise, in the case that the multilist
* has a power of two number of sublists, each sublists' usage
- * would not be evenly distributed.
+ * would not be evenly distributed. In this context full 64bit
+ * division would be a waste of time, so limit it to 32 bits.
*/
- return (dbuf_hash(db->db_objset, db->db.db_object,
+ return ((unsigned int)dbuf_hash(db->db_objset, db->db.db_object,
db->db_level, db->db_blkid) %
multilist_get_num_sublists(ml));
}
/*
* The target size of the dbuf cache can grow with the ARC target,
* unless limited by the tunable dbuf_cache_max_bytes.
*/
static inline unsigned long
dbuf_cache_target_bytes(void)
{
return (MIN(dbuf_cache_max_bytes,
arc_target_bytes() >> dbuf_cache_shift));
}
/*
* The target size of the dbuf metadata cache can grow with the ARC target,
* unless limited by the tunable dbuf_metadata_cache_max_bytes.
*/
static inline unsigned long
dbuf_metadata_cache_target_bytes(void)
{
return (MIN(dbuf_metadata_cache_max_bytes,
arc_target_bytes() >> dbuf_metadata_cache_shift));
}
static inline uint64_t
dbuf_cache_hiwater_bytes(void)
{
uint64_t dbuf_cache_target = dbuf_cache_target_bytes();
return (dbuf_cache_target +
(dbuf_cache_target * dbuf_cache_hiwater_pct) / 100);
}
static inline uint64_t
dbuf_cache_lowater_bytes(void)
{
uint64_t dbuf_cache_target = dbuf_cache_target_bytes();
return (dbuf_cache_target -
(dbuf_cache_target * dbuf_cache_lowater_pct) / 100);
}
static inline boolean_t
dbuf_cache_above_lowater(void)
{
return (zfs_refcount_count(&dbuf_caches[DB_DBUF_CACHE].size) >
dbuf_cache_lowater_bytes());
}
/*
* Evict the oldest eligible dbuf from the dbuf cache.
*/
static void
dbuf_evict_one(void)
{
int idx = multilist_get_random_index(&dbuf_caches[DB_DBUF_CACHE].cache);
multilist_sublist_t *mls = multilist_sublist_lock(
&dbuf_caches[DB_DBUF_CACHE].cache, idx);
ASSERT(!MUTEX_HELD(&dbuf_evict_lock));
dmu_buf_impl_t *db = multilist_sublist_tail(mls);
while (db != NULL && mutex_tryenter(&db->db_mtx) == 0) {
db = multilist_sublist_prev(mls, db);
}
DTRACE_PROBE2(dbuf__evict__one, dmu_buf_impl_t *, db,
multilist_sublist_t *, mls);
if (db != NULL) {
multilist_sublist_remove(mls, db);
multilist_sublist_unlock(mls);
(void) zfs_refcount_remove_many(
&dbuf_caches[DB_DBUF_CACHE].size, db->db.db_size, db);
DBUF_STAT_BUMPDOWN(cache_levels[db->db_level]);
DBUF_STAT_BUMPDOWN(cache_count);
DBUF_STAT_DECR(cache_levels_bytes[db->db_level],
db->db.db_size);
ASSERT3U(db->db_caching_status, ==, DB_DBUF_CACHE);
db->db_caching_status = DB_NO_CACHE;
dbuf_destroy(db);
DBUF_STAT_BUMP(cache_total_evicts);
} else {
multilist_sublist_unlock(mls);
}
}
/*
* The dbuf evict thread is responsible for aging out dbufs from the
* cache. Once the cache has reached it's maximum size, dbufs are removed
* and destroyed. The eviction thread will continue running until the size
* of the dbuf cache is at or below the maximum size. Once the dbuf is aged
* out of the cache it is destroyed and becomes eligible for arc eviction.
*/
/* ARGSUSED */
static void
dbuf_evict_thread(void *unused)
{
callb_cpr_t cpr;
CALLB_CPR_INIT(&cpr, &dbuf_evict_lock, callb_generic_cpr, FTAG);
mutex_enter(&dbuf_evict_lock);
while (!dbuf_evict_thread_exit) {
while (!dbuf_cache_above_lowater() && !dbuf_evict_thread_exit) {
CALLB_CPR_SAFE_BEGIN(&cpr);
(void) cv_timedwait_idle_hires(&dbuf_evict_cv,
&dbuf_evict_lock, SEC2NSEC(1), MSEC2NSEC(1), 0);
CALLB_CPR_SAFE_END(&cpr, &dbuf_evict_lock);
}
mutex_exit(&dbuf_evict_lock);
/*
* Keep evicting as long as we're above the low water mark
* for the cache. We do this without holding the locks to
* minimize lock contention.
*/
while (dbuf_cache_above_lowater() && !dbuf_evict_thread_exit) {
dbuf_evict_one();
}
mutex_enter(&dbuf_evict_lock);
}
dbuf_evict_thread_exit = B_FALSE;
cv_broadcast(&dbuf_evict_cv);
CALLB_CPR_EXIT(&cpr); /* drops dbuf_evict_lock */
thread_exit();
}
/*
* Wake up the dbuf eviction thread if the dbuf cache is at its max size.
* If the dbuf cache is at its high water mark, then evict a dbuf from the
* dbuf cache using the callers context.
*/
static void
dbuf_evict_notify(uint64_t size)
{
/*
* We check if we should evict without holding the dbuf_evict_lock,
* because it's OK to occasionally make the wrong decision here,
* and grabbing the lock results in massive lock contention.
*/
if (size > dbuf_cache_target_bytes()) {
if (size > dbuf_cache_hiwater_bytes())
dbuf_evict_one();
cv_signal(&dbuf_evict_cv);
}
}
static int
dbuf_kstat_update(kstat_t *ksp, int rw)
{
dbuf_stats_t *ds = ksp->ks_data;
if (rw == KSTAT_WRITE)
return (SET_ERROR(EACCES));
ds->cache_count.value.ui64 =
wmsum_value(&dbuf_sums.cache_count);
ds->cache_size_bytes.value.ui64 =
zfs_refcount_count(&dbuf_caches[DB_DBUF_CACHE].size);
ds->cache_target_bytes.value.ui64 = dbuf_cache_target_bytes();
ds->cache_hiwater_bytes.value.ui64 = dbuf_cache_hiwater_bytes();
ds->cache_lowater_bytes.value.ui64 = dbuf_cache_lowater_bytes();
ds->cache_total_evicts.value.ui64 =
wmsum_value(&dbuf_sums.cache_total_evicts);
for (int i = 0; i < DN_MAX_LEVELS; i++) {
ds->cache_levels[i].value.ui64 =
wmsum_value(&dbuf_sums.cache_levels[i]);
ds->cache_levels_bytes[i].value.ui64 =
wmsum_value(&dbuf_sums.cache_levels_bytes[i]);
}
ds->hash_hits.value.ui64 =
wmsum_value(&dbuf_sums.hash_hits);
ds->hash_misses.value.ui64 =
wmsum_value(&dbuf_sums.hash_misses);
ds->hash_collisions.value.ui64 =
wmsum_value(&dbuf_sums.hash_collisions);
ds->hash_chains.value.ui64 =
wmsum_value(&dbuf_sums.hash_chains);
ds->hash_insert_race.value.ui64 =
wmsum_value(&dbuf_sums.hash_insert_race);
ds->metadata_cache_count.value.ui64 =
wmsum_value(&dbuf_sums.metadata_cache_count);
ds->metadata_cache_size_bytes.value.ui64 = zfs_refcount_count(
&dbuf_caches[DB_DBUF_METADATA_CACHE].size);
ds->metadata_cache_overflow.value.ui64 =
wmsum_value(&dbuf_sums.metadata_cache_overflow);
return (0);
}
void
dbuf_init(void)
{
uint64_t hsize = 1ULL << 16;
dbuf_hash_table_t *h = &dbuf_hash_table;
int i;
/*
* The hash table is big enough to fill all of physical memory
* with an average block size of zfs_arc_average_blocksize (default 8K).
* By default, the table will take up
* totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers).
*/
while (hsize * zfs_arc_average_blocksize < physmem * PAGESIZE)
hsize <<= 1;
retry:
h->hash_table_mask = hsize - 1;
#if defined(_KERNEL)
/*
* Large allocations which do not require contiguous pages
* should be using vmem_alloc() in the linux kernel
*/
h->hash_table = vmem_zalloc(hsize * sizeof (void *), KM_SLEEP);
#else
h->hash_table = kmem_zalloc(hsize * sizeof (void *), KM_NOSLEEP);
#endif
if (h->hash_table == NULL) {
/* XXX - we should really return an error instead of assert */
ASSERT(hsize > (1ULL << 10));
hsize >>= 1;
goto retry;
}
dbuf_kmem_cache = kmem_cache_create("dmu_buf_impl_t",
sizeof (dmu_buf_impl_t),
0, dbuf_cons, dbuf_dest, NULL, NULL, NULL, 0);
for (i = 0; i < DBUF_MUTEXES; i++)
mutex_init(&h->hash_mutexes[i], NULL, MUTEX_DEFAULT, NULL);
dbuf_stats_init(h);
/*
* All entries are queued via taskq_dispatch_ent(), so min/maxalloc
* configuration is not required.
*/
dbu_evict_taskq = taskq_create("dbu_evict", 1, defclsyspri, 0, 0, 0);
for (dbuf_cached_state_t dcs = 0; dcs < DB_CACHE_MAX; dcs++) {
multilist_create(&dbuf_caches[dcs].cache,
sizeof (dmu_buf_impl_t),
offsetof(dmu_buf_impl_t, db_cache_link),
dbuf_cache_multilist_index_func);
zfs_refcount_create(&dbuf_caches[dcs].size);
}
dbuf_evict_thread_exit = B_FALSE;
mutex_init(&dbuf_evict_lock, NULL, MUTEX_DEFAULT, NULL);
cv_init(&dbuf_evict_cv, NULL, CV_DEFAULT, NULL);
dbuf_cache_evict_thread = thread_create(NULL, 0, dbuf_evict_thread,
NULL, 0, &p0, TS_RUN, minclsyspri);
wmsum_init(&dbuf_sums.cache_count, 0);
wmsum_init(&dbuf_sums.cache_total_evicts, 0);
for (i = 0; i < DN_MAX_LEVELS; i++) {
wmsum_init(&dbuf_sums.cache_levels[i], 0);
wmsum_init(&dbuf_sums.cache_levels_bytes[i], 0);
}
wmsum_init(&dbuf_sums.hash_hits, 0);
wmsum_init(&dbuf_sums.hash_misses, 0);
wmsum_init(&dbuf_sums.hash_collisions, 0);
wmsum_init(&dbuf_sums.hash_chains, 0);
wmsum_init(&dbuf_sums.hash_insert_race, 0);
wmsum_init(&dbuf_sums.metadata_cache_count, 0);
wmsum_init(&dbuf_sums.metadata_cache_overflow, 0);
dbuf_ksp = kstat_create("zfs", 0, "dbufstats", "misc",
KSTAT_TYPE_NAMED, sizeof (dbuf_stats) / sizeof (kstat_named_t),
KSTAT_FLAG_VIRTUAL);
if (dbuf_ksp != NULL) {
for (i = 0; i < DN_MAX_LEVELS; i++) {
snprintf(dbuf_stats.cache_levels[i].name,
KSTAT_STRLEN, "cache_level_%d", i);
dbuf_stats.cache_levels[i].data_type =
KSTAT_DATA_UINT64;
snprintf(dbuf_stats.cache_levels_bytes[i].name,
KSTAT_STRLEN, "cache_level_%d_bytes", i);
dbuf_stats.cache_levels_bytes[i].data_type =
KSTAT_DATA_UINT64;
}
dbuf_ksp->ks_data = &dbuf_stats;
dbuf_ksp->ks_update = dbuf_kstat_update;
kstat_install(dbuf_ksp);
}
}
void
dbuf_fini(void)
{
dbuf_hash_table_t *h = &dbuf_hash_table;
int i;
dbuf_stats_destroy();
for (i = 0; i < DBUF_MUTEXES; i++)
mutex_destroy(&h->hash_mutexes[i]);
#if defined(_KERNEL)
/*
* Large allocations which do not require contiguous pages
* should be using vmem_free() in the linux kernel
*/
vmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *));
#else
kmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *));
#endif
kmem_cache_destroy(dbuf_kmem_cache);
taskq_destroy(dbu_evict_taskq);
mutex_enter(&dbuf_evict_lock);
dbuf_evict_thread_exit = B_TRUE;
while (dbuf_evict_thread_exit) {
cv_signal(&dbuf_evict_cv);
cv_wait(&dbuf_evict_cv, &dbuf_evict_lock);
}
mutex_exit(&dbuf_evict_lock);
mutex_destroy(&dbuf_evict_lock);
cv_destroy(&dbuf_evict_cv);
for (dbuf_cached_state_t dcs = 0; dcs < DB_CACHE_MAX; dcs++) {
zfs_refcount_destroy(&dbuf_caches[dcs].size);
multilist_destroy(&dbuf_caches[dcs].cache);
}
if (dbuf_ksp != NULL) {
kstat_delete(dbuf_ksp);
dbuf_ksp = NULL;
}
wmsum_fini(&dbuf_sums.cache_count);
wmsum_fini(&dbuf_sums.cache_total_evicts);
for (i = 0; i < DN_MAX_LEVELS; i++) {
wmsum_fini(&dbuf_sums.cache_levels[i]);
wmsum_fini(&dbuf_sums.cache_levels_bytes[i]);
}
wmsum_fini(&dbuf_sums.hash_hits);
wmsum_fini(&dbuf_sums.hash_misses);
wmsum_fini(&dbuf_sums.hash_collisions);
wmsum_fini(&dbuf_sums.hash_chains);
wmsum_fini(&dbuf_sums.hash_insert_race);
wmsum_fini(&dbuf_sums.metadata_cache_count);
wmsum_fini(&dbuf_sums.metadata_cache_overflow);
}
/*
* Other stuff.
*/
#ifdef ZFS_DEBUG
static void
dbuf_verify(dmu_buf_impl_t *db)
{
dnode_t *dn;
dbuf_dirty_record_t *dr;
uint32_t txg_prev;
ASSERT(MUTEX_HELD(&db->db_mtx));
if (!(zfs_flags & ZFS_DEBUG_DBUF_VERIFY))
return;
ASSERT(db->db_objset != NULL);
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
if (dn == NULL) {
ASSERT(db->db_parent == NULL);
ASSERT(db->db_blkptr == NULL);
} else {
ASSERT3U(db->db.db_object, ==, dn->dn_object);
ASSERT3P(db->db_objset, ==, dn->dn_objset);
ASSERT3U(db->db_level, <, dn->dn_nlevels);
ASSERT(db->db_blkid == DMU_BONUS_BLKID ||
db->db_blkid == DMU_SPILL_BLKID ||
!avl_is_empty(&dn->dn_dbufs));
}
if (db->db_blkid == DMU_BONUS_BLKID) {
ASSERT(dn != NULL);
ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen);
ASSERT3U(db->db.db_offset, ==, DMU_BONUS_BLKID);
} else if (db->db_blkid == DMU_SPILL_BLKID) {
ASSERT(dn != NULL);
ASSERT0(db->db.db_offset);
} else {
ASSERT3U(db->db.db_offset, ==, db->db_blkid * db->db.db_size);
}
if ((dr = list_head(&db->db_dirty_records)) != NULL) {
ASSERT(dr->dr_dbuf == db);
txg_prev = dr->dr_txg;
for (dr = list_next(&db->db_dirty_records, dr); dr != NULL;
dr = list_next(&db->db_dirty_records, dr)) {
ASSERT(dr->dr_dbuf == db);
ASSERT(txg_prev > dr->dr_txg);
txg_prev = dr->dr_txg;
}
}
/*
* We can't assert that db_size matches dn_datablksz because it
* can be momentarily different when another thread is doing
* dnode_set_blksz().
*/
if (db->db_level == 0 && db->db.db_object == DMU_META_DNODE_OBJECT) {
dr = db->db_data_pending;
/*
* It should only be modified in syncing context, so
* make sure we only have one copy of the data.
*/
ASSERT(dr == NULL || dr->dt.dl.dr_data == db->db_buf);
}
/* verify db->db_blkptr */
if (db->db_blkptr) {
if (db->db_parent == dn->dn_dbuf) {
/* db is pointed to by the dnode */
/* ASSERT3U(db->db_blkid, <, dn->dn_nblkptr); */
if (DMU_OBJECT_IS_SPECIAL(db->db.db_object))
ASSERT(db->db_parent == NULL);
else
ASSERT(db->db_parent != NULL);
if (db->db_blkid != DMU_SPILL_BLKID)
ASSERT3P(db->db_blkptr, ==,
&dn->dn_phys->dn_blkptr[db->db_blkid]);
} else {
/* db is pointed to by an indirect block */
int epb __maybe_unused = db->db_parent->db.db_size >>
SPA_BLKPTRSHIFT;
ASSERT3U(db->db_parent->db_level, ==, db->db_level+1);
ASSERT3U(db->db_parent->db.db_object, ==,
db->db.db_object);
/*
* dnode_grow_indblksz() can make this fail if we don't
* have the parent's rwlock. XXX indblksz no longer
* grows. safe to do this now?
*/
if (RW_LOCK_HELD(&db->db_parent->db_rwlock)) {
ASSERT3P(db->db_blkptr, ==,
((blkptr_t *)db->db_parent->db.db_data +
db->db_blkid % epb));
}
}
}
if ((db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr)) &&
(db->db_buf == NULL || db->db_buf->b_data) &&
db->db.db_data && db->db_blkid != DMU_BONUS_BLKID &&
db->db_state != DB_FILL && !dn->dn_free_txg) {
/*
* If the blkptr isn't set but they have nonzero data,
* it had better be dirty, otherwise we'll lose that
* data when we evict this buffer.
*
* There is an exception to this rule for indirect blocks; in
* this case, if the indirect block is a hole, we fill in a few
* fields on each of the child blocks (importantly, birth time)
* to prevent hole birth times from being lost when you
* partially fill in a hole.
*/
if (db->db_dirtycnt == 0) {
if (db->db_level == 0) {
uint64_t *buf = db->db.db_data;
int i;
for (i = 0; i < db->db.db_size >> 3; i++) {
ASSERT(buf[i] == 0);
}
} else {
blkptr_t *bps = db->db.db_data;
ASSERT3U(1 << DB_DNODE(db)->dn_indblkshift, ==,
db->db.db_size);
/*
* We want to verify that all the blkptrs in the
* indirect block are holes, but we may have
* automatically set up a few fields for them.
* We iterate through each blkptr and verify
* they only have those fields set.
*/
for (int i = 0;
i < db->db.db_size / sizeof (blkptr_t);
i++) {
blkptr_t *bp = &bps[i];
ASSERT(ZIO_CHECKSUM_IS_ZERO(
&bp->blk_cksum));
ASSERT(
DVA_IS_EMPTY(&bp->blk_dva[0]) &&
DVA_IS_EMPTY(&bp->blk_dva[1]) &&
DVA_IS_EMPTY(&bp->blk_dva[2]));
ASSERT0(bp->blk_fill);
ASSERT0(bp->blk_pad[0]);
ASSERT0(bp->blk_pad[1]);
ASSERT(!BP_IS_EMBEDDED(bp));
ASSERT(BP_IS_HOLE(bp));
ASSERT0(bp->blk_phys_birth);
}
}
}
}
DB_DNODE_EXIT(db);
}
#endif
static void
dbuf_clear_data(dmu_buf_impl_t *db)
{
ASSERT(MUTEX_HELD(&db->db_mtx));
dbuf_evict_user(db);
ASSERT3P(db->db_buf, ==, NULL);
db->db.db_data = NULL;
if (db->db_state != DB_NOFILL) {
db->db_state = DB_UNCACHED;
DTRACE_SET_STATE(db, "clear data");
}
}
static void
dbuf_set_data(dmu_buf_impl_t *db, arc_buf_t *buf)
{
ASSERT(MUTEX_HELD(&db->db_mtx));
ASSERT(buf != NULL);
db->db_buf = buf;
ASSERT(buf->b_data != NULL);
db->db.db_data = buf->b_data;
}
static arc_buf_t *
dbuf_alloc_arcbuf(dmu_buf_impl_t *db)
{
spa_t *spa = db->db_objset->os_spa;
return (arc_alloc_buf(spa, db, DBUF_GET_BUFC_TYPE(db), db->db.db_size));
}
/*
* Loan out an arc_buf for read. Return the loaned arc_buf.
*/
arc_buf_t *
dbuf_loan_arcbuf(dmu_buf_impl_t *db)
{
arc_buf_t *abuf;
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
mutex_enter(&db->db_mtx);
if (arc_released(db->db_buf) || zfs_refcount_count(&db->db_holds) > 1) {
int blksz = db->db.db_size;
spa_t *spa = db->db_objset->os_spa;
mutex_exit(&db->db_mtx);
abuf = arc_loan_buf(spa, B_FALSE, blksz);
bcopy(db->db.db_data, abuf->b_data, blksz);
} else {
abuf = db->db_buf;
arc_loan_inuse_buf(abuf, db);
db->db_buf = NULL;
dbuf_clear_data(db);
mutex_exit(&db->db_mtx);
}
return (abuf);
}
/*
* Calculate which level n block references the data at the level 0 offset
* provided.
*/
uint64_t
dbuf_whichblock(const dnode_t *dn, const int64_t level, const uint64_t offset)
{
if (dn->dn_datablkshift != 0 && dn->dn_indblkshift != 0) {
/*
* The level n blkid is equal to the level 0 blkid divided by
* the number of level 0s in a level n block.
*
* The level 0 blkid is offset >> datablkshift =
* offset / 2^datablkshift.
*
* The number of level 0s in a level n is the number of block
* pointers in an indirect block, raised to the power of level.
* This is 2^(indblkshift - SPA_BLKPTRSHIFT)^level =
* 2^(level*(indblkshift - SPA_BLKPTRSHIFT)).
*
* Thus, the level n blkid is: offset /
* ((2^datablkshift)*(2^(level*(indblkshift-SPA_BLKPTRSHIFT))))
* = offset / 2^(datablkshift + level *
* (indblkshift - SPA_BLKPTRSHIFT))
* = offset >> (datablkshift + level *
* (indblkshift - SPA_BLKPTRSHIFT))
*/
const unsigned exp = dn->dn_datablkshift +
level * (dn->dn_indblkshift - SPA_BLKPTRSHIFT);
if (exp >= 8 * sizeof (offset)) {
/* This only happens on the highest indirection level */
ASSERT3U(level, ==, dn->dn_nlevels - 1);
return (0);
}
ASSERT3U(exp, <, 8 * sizeof (offset));
return (offset >> exp);
} else {
ASSERT3U(offset, <, dn->dn_datablksz);
return (0);
}
}
/*
* This function is used to lock the parent of the provided dbuf. This should be
* used when modifying or reading db_blkptr.
*/
db_lock_type_t
dmu_buf_lock_parent(dmu_buf_impl_t *db, krw_t rw, void *tag)
{
enum db_lock_type ret = DLT_NONE;
if (db->db_parent != NULL) {
rw_enter(&db->db_parent->db_rwlock, rw);
ret = DLT_PARENT;
} else if (dmu_objset_ds(db->db_objset) != NULL) {
rrw_enter(&dmu_objset_ds(db->db_objset)->ds_bp_rwlock, rw,
tag);
ret = DLT_OBJSET;
}
/*
* We only return a DLT_NONE lock when it's the top-most indirect block
* of the meta-dnode of the MOS.
*/
return (ret);
}
/*
* We need to pass the lock type in because it's possible that the block will
* move from being the topmost indirect block in a dnode (and thus, have no
* parent) to not the top-most via an indirection increase. This would cause a
* panic if we didn't pass the lock type in.
*/
void
dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type, void *tag)
{
if (type == DLT_PARENT)
rw_exit(&db->db_parent->db_rwlock);
else if (type == DLT_OBJSET)
rrw_exit(&dmu_objset_ds(db->db_objset)->ds_bp_rwlock, tag);
}
static void
dbuf_read_done(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp,
arc_buf_t *buf, void *vdb)
{
dmu_buf_impl_t *db = vdb;
mutex_enter(&db->db_mtx);
ASSERT3U(db->db_state, ==, DB_READ);
/*
* All reads are synchronous, so we must have a hold on the dbuf
*/
ASSERT(zfs_refcount_count(&db->db_holds) > 0);
ASSERT(db->db_buf == NULL);
ASSERT(db->db.db_data == NULL);
if (buf == NULL) {
/* i/o error */
ASSERT(zio == NULL || zio->io_error != 0);
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
ASSERT3P(db->db_buf, ==, NULL);
db->db_state = DB_UNCACHED;
DTRACE_SET_STATE(db, "i/o error");
} else if (db->db_level == 0 && db->db_freed_in_flight) {
/* freed in flight */
ASSERT(zio == NULL || zio->io_error == 0);
arc_release(buf, db);
bzero(buf->b_data, db->db.db_size);
arc_buf_freeze(buf);
db->db_freed_in_flight = FALSE;
dbuf_set_data(db, buf);
db->db_state = DB_CACHED;
DTRACE_SET_STATE(db, "freed in flight");
} else {
/* success */
ASSERT(zio == NULL || zio->io_error == 0);
dbuf_set_data(db, buf);
db->db_state = DB_CACHED;
DTRACE_SET_STATE(db, "successful read");
}
cv_broadcast(&db->db_changed);
dbuf_rele_and_unlock(db, NULL, B_FALSE);
}
/*
* Shortcut for performing reads on bonus dbufs. Returns
* an error if we fail to verify the dnode associated with
* a decrypted block. Otherwise success.
*/
static int
dbuf_read_bonus(dmu_buf_impl_t *db, dnode_t *dn, uint32_t flags)
{
int bonuslen, max_bonuslen, err;
err = dbuf_read_verify_dnode_crypt(db, flags);
if (err)
return (err);
bonuslen = MIN(dn->dn_bonuslen, dn->dn_phys->dn_bonuslen);
max_bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots);
ASSERT(MUTEX_HELD(&db->db_mtx));
ASSERT(DB_DNODE_HELD(db));
ASSERT3U(bonuslen, <=, db->db.db_size);
db->db.db_data = kmem_alloc(max_bonuslen, KM_SLEEP);
arc_space_consume(max_bonuslen, ARC_SPACE_BONUS);
if (bonuslen < max_bonuslen)
bzero(db->db.db_data, max_bonuslen);
if (bonuslen)
bcopy(DN_BONUS(dn->dn_phys), db->db.db_data, bonuslen);
db->db_state = DB_CACHED;
DTRACE_SET_STATE(db, "bonus buffer filled");
return (0);
}
static void
dbuf_handle_indirect_hole(dmu_buf_impl_t *db, dnode_t *dn)
{
blkptr_t *bps = db->db.db_data;
uint32_t indbs = 1ULL << dn->dn_indblkshift;
int n_bps = indbs >> SPA_BLKPTRSHIFT;
for (int i = 0; i < n_bps; i++) {
blkptr_t *bp = &bps[i];
ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==, indbs);
BP_SET_LSIZE(bp, BP_GET_LEVEL(db->db_blkptr) == 1 ?
dn->dn_datablksz : BP_GET_LSIZE(db->db_blkptr));
BP_SET_TYPE(bp, BP_GET_TYPE(db->db_blkptr));
BP_SET_LEVEL(bp, BP_GET_LEVEL(db->db_blkptr) - 1);
BP_SET_BIRTH(bp, db->db_blkptr->blk_birth, 0);
}
}
/*
* Handle reads on dbufs that are holes, if necessary. This function
* requires that the dbuf's mutex is held. Returns success (0) if action
* was taken, ENOENT if no action was taken.
*/
static int
dbuf_read_hole(dmu_buf_impl_t *db, dnode_t *dn, uint32_t flags)
{
ASSERT(MUTEX_HELD(&db->db_mtx));
int is_hole = db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr);
/*
* For level 0 blocks only, if the above check fails:
* Recheck BP_IS_HOLE() after dnode_block_freed() in case dnode_sync()
* processes the delete record and clears the bp while we are waiting
* for the dn_mtx (resulting in a "no" from block_freed).
*/
if (!is_hole && db->db_level == 0) {
is_hole = dnode_block_freed(dn, db->db_blkid) ||
BP_IS_HOLE(db->db_blkptr);
}
if (is_hole) {
dbuf_set_data(db, dbuf_alloc_arcbuf(db));
bzero(db->db.db_data, db->db.db_size);
if (db->db_blkptr != NULL && db->db_level > 0 &&
BP_IS_HOLE(db->db_blkptr) &&
db->db_blkptr->blk_birth != 0) {
dbuf_handle_indirect_hole(db, dn);
}
db->db_state = DB_CACHED;
DTRACE_SET_STATE(db, "hole read satisfied");
return (0);
}
return (ENOENT);
}
/*
* This function ensures that, when doing a decrypting read of a block,
* we make sure we have decrypted the dnode associated with it. We must do
* this so that we ensure we are fully authenticating the checksum-of-MACs
* tree from the root of the objset down to this block. Indirect blocks are
* always verified against their secure checksum-of-MACs assuming that the
* dnode containing them is correct. Now that we are doing a decrypting read,
* we can be sure that the key is loaded and verify that assumption. This is
* especially important considering that we always read encrypted dnode
* blocks as raw data (without verifying their MACs) to start, and
* decrypt / authenticate them when we need to read an encrypted bonus buffer.
*/
static int
dbuf_read_verify_dnode_crypt(dmu_buf_impl_t *db, uint32_t flags)
{
int err = 0;
objset_t *os = db->db_objset;
arc_buf_t *dnode_abuf;
dnode_t *dn;
zbookmark_phys_t zb;
ASSERT(MUTEX_HELD(&db->db_mtx));
if (!os->os_encrypted || os->os_raw_receive ||
(flags & DB_RF_NO_DECRYPT) != 0)
return (0);
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
dnode_abuf = (dn->dn_dbuf != NULL) ? dn->dn_dbuf->db_buf : NULL;
if (dnode_abuf == NULL || !arc_is_encrypted(dnode_abuf)) {
DB_DNODE_EXIT(db);
return (0);
}
SET_BOOKMARK(&zb, dmu_objset_id(os),
DMU_META_DNODE_OBJECT, 0, dn->dn_dbuf->db_blkid);
err = arc_untransform(dnode_abuf, os->os_spa, &zb, B_TRUE);
/*
* An error code of EACCES tells us that the key is still not
* available. This is ok if we are only reading authenticated
* (and therefore non-encrypted) blocks.
*/
if (err == EACCES && ((db->db_blkid != DMU_BONUS_BLKID &&
!DMU_OT_IS_ENCRYPTED(dn->dn_type)) ||
(db->db_blkid == DMU_BONUS_BLKID &&
!DMU_OT_IS_ENCRYPTED(dn->dn_bonustype))))
err = 0;
DB_DNODE_EXIT(db);
return (err);
}
/*
* Drops db_mtx and the parent lock specified by dblt and tag before
* returning.
*/
static int
dbuf_read_impl(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags,
db_lock_type_t dblt, void *tag)
{
dnode_t *dn;
zbookmark_phys_t zb;
uint32_t aflags = ARC_FLAG_NOWAIT;
int err, zio_flags;
err = zio_flags = 0;
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
ASSERT(!zfs_refcount_is_zero(&db->db_holds));
ASSERT(MUTEX_HELD(&db->db_mtx));
ASSERT(db->db_state == DB_UNCACHED);
ASSERT(db->db_buf == NULL);
ASSERT(db->db_parent == NULL ||
RW_LOCK_HELD(&db->db_parent->db_rwlock));
if (db->db_blkid == DMU_BONUS_BLKID) {
err = dbuf_read_bonus(db, dn, flags);
goto early_unlock;
}
err = dbuf_read_hole(db, dn, flags);
if (err == 0)
goto early_unlock;
/*
* Any attempt to read a redacted block should result in an error. This
* will never happen under normal conditions, but can be useful for
* debugging purposes.
*/
if (BP_IS_REDACTED(db->db_blkptr)) {
ASSERT(dsl_dataset_feature_is_active(
db->db_objset->os_dsl_dataset,
SPA_FEATURE_REDACTED_DATASETS));
err = SET_ERROR(EIO);
goto early_unlock;
}
SET_BOOKMARK(&zb, dmu_objset_id(db->db_objset),
db->db.db_object, db->db_level, db->db_blkid);
/*
* All bps of an encrypted os should have the encryption bit set.
* If this is not true it indicates tampering and we report an error.
*/
if (db->db_objset->os_encrypted && !BP_USES_CRYPT(db->db_blkptr)) {
spa_log_error(db->db_objset->os_spa, &zb);
zfs_panic_recover("unencrypted block in encrypted "
"object set %llu", dmu_objset_id(db->db_objset));
err = SET_ERROR(EIO);
goto early_unlock;
}
err = dbuf_read_verify_dnode_crypt(db, flags);
if (err != 0)
goto early_unlock;
DB_DNODE_EXIT(db);
db->db_state = DB_READ;
DTRACE_SET_STATE(db, "read issued");
mutex_exit(&db->db_mtx);
if (DBUF_IS_L2CACHEABLE(db))
aflags |= ARC_FLAG_L2CACHE;
dbuf_add_ref(db, NULL);
zio_flags = (flags & DB_RF_CANFAIL) ?
ZIO_FLAG_CANFAIL : ZIO_FLAG_MUSTSUCCEED;
if ((flags & DB_RF_NO_DECRYPT) && BP_IS_PROTECTED(db->db_blkptr))
zio_flags |= ZIO_FLAG_RAW;
/*
* The zio layer will copy the provided blkptr later, but we need to
* do this now so that we can release the parent's rwlock. We have to
* do that now so that if dbuf_read_done is called synchronously (on
* an l1 cache hit) we don't acquire the db_mtx while holding the
* parent's rwlock, which would be a lock ordering violation.
*/
blkptr_t bp = *db->db_blkptr;
dmu_buf_unlock_parent(db, dblt, tag);
(void) arc_read(zio, db->db_objset->os_spa, &bp,
dbuf_read_done, db, ZIO_PRIORITY_SYNC_READ, zio_flags,
&aflags, &zb);
return (err);
early_unlock:
DB_DNODE_EXIT(db);
mutex_exit(&db->db_mtx);
dmu_buf_unlock_parent(db, dblt, tag);
return (err);
}
/*
* This is our just-in-time copy function. It makes a copy of buffers that
* have been modified in a previous transaction group before we access them in
* the current active group.
*
* This function is used in three places: when we are dirtying a buffer for the
* first time in a txg, when we are freeing a range in a dnode that includes
* this buffer, and when we are accessing a buffer which was received compressed
* and later referenced in a WRITE_BYREF record.
*
* Note that when we are called from dbuf_free_range() we do not put a hold on
* the buffer, we just traverse the active dbuf list for the dnode.
*/
static void
dbuf_fix_old_data(dmu_buf_impl_t *db, uint64_t txg)
{
dbuf_dirty_record_t *dr = list_head(&db->db_dirty_records);
ASSERT(MUTEX_HELD(&db->db_mtx));
ASSERT(db->db.db_data != NULL);
ASSERT(db->db_level == 0);
ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT);
if (dr == NULL ||
(dr->dt.dl.dr_data !=
((db->db_blkid == DMU_BONUS_BLKID) ? db->db.db_data : db->db_buf)))
return;
/*
* If the last dirty record for this dbuf has not yet synced
* and its referencing the dbuf data, either:
* reset the reference to point to a new copy,
* or (if there a no active holders)
* just null out the current db_data pointer.
*/
ASSERT3U(dr->dr_txg, >=, txg - 2);
if (db->db_blkid == DMU_BONUS_BLKID) {
dnode_t *dn = DB_DNODE(db);
int bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots);
dr->dt.dl.dr_data = kmem_alloc(bonuslen, KM_SLEEP);
arc_space_consume(bonuslen, ARC_SPACE_BONUS);
bcopy(db->db.db_data, dr->dt.dl.dr_data, bonuslen);
} else if (zfs_refcount_count(&db->db_holds) > db->db_dirtycnt) {
dnode_t *dn = DB_DNODE(db);
int size = arc_buf_size(db->db_buf);
arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db);
spa_t *spa = db->db_objset->os_spa;
enum zio_compress compress_type =
arc_get_compression(db->db_buf);
uint8_t complevel = arc_get_complevel(db->db_buf);
if (arc_is_encrypted(db->db_buf)) {
boolean_t byteorder;
uint8_t salt[ZIO_DATA_SALT_LEN];
uint8_t iv[ZIO_DATA_IV_LEN];
uint8_t mac[ZIO_DATA_MAC_LEN];
arc_get_raw_params(db->db_buf, &byteorder, salt,
iv, mac);
dr->dt.dl.dr_data = arc_alloc_raw_buf(spa, db,
dmu_objset_id(dn->dn_objset), byteorder, salt, iv,
mac, dn->dn_type, size, arc_buf_lsize(db->db_buf),
compress_type, complevel);
} else if (compress_type != ZIO_COMPRESS_OFF) {
ASSERT3U(type, ==, ARC_BUFC_DATA);
dr->dt.dl.dr_data = arc_alloc_compressed_buf(spa, db,
size, arc_buf_lsize(db->db_buf), compress_type,
complevel);
} else {
dr->dt.dl.dr_data = arc_alloc_buf(spa, db, type, size);
}
bcopy(db->db.db_data, dr->dt.dl.dr_data->b_data, size);
} else {
db->db_buf = NULL;
dbuf_clear_data(db);
}
}
int
dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags)
{
int err = 0;
boolean_t prefetch;
dnode_t *dn;
/*
* We don't have to hold the mutex to check db_state because it
* can't be freed while we have a hold on the buffer.
*/
ASSERT(!zfs_refcount_is_zero(&db->db_holds));
if (db->db_state == DB_NOFILL)
return (SET_ERROR(EIO));
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
prefetch = db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID &&
(flags & DB_RF_NOPREFETCH) == 0 && dn != NULL &&
DBUF_IS_CACHEABLE(db);
mutex_enter(&db->db_mtx);
if (db->db_state == DB_CACHED) {
spa_t *spa = dn->dn_objset->os_spa;
/*
* Ensure that this block's dnode has been decrypted if
* the caller has requested decrypted data.
*/
err = dbuf_read_verify_dnode_crypt(db, flags);
/*
* If the arc buf is compressed or encrypted and the caller
* requested uncompressed data, we need to untransform it
* before returning. We also call arc_untransform() on any
* unauthenticated blocks, which will verify their MAC if
* the key is now available.
*/
if (err == 0 && db->db_buf != NULL &&
(flags & DB_RF_NO_DECRYPT) == 0 &&
(arc_is_encrypted(db->db_buf) ||
arc_is_unauthenticated(db->db_buf) ||
arc_get_compression(db->db_buf) != ZIO_COMPRESS_OFF)) {
zbookmark_phys_t zb;
SET_BOOKMARK(&zb, dmu_objset_id(db->db_objset),
db->db.db_object, db->db_level, db->db_blkid);
dbuf_fix_old_data(db, spa_syncing_txg(spa));
err = arc_untransform(db->db_buf, spa, &zb, B_FALSE);
dbuf_set_data(db, db->db_buf);
}
mutex_exit(&db->db_mtx);
if (err == 0 && prefetch) {
dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE,
B_FALSE, flags & DB_RF_HAVESTRUCT);
}
DB_DNODE_EXIT(db);
DBUF_STAT_BUMP(hash_hits);
} else if (db->db_state == DB_UNCACHED) {
spa_t *spa = dn->dn_objset->os_spa;
boolean_t need_wait = B_FALSE;
db_lock_type_t dblt = dmu_buf_lock_parent(db, RW_READER, FTAG);
if (zio == NULL &&
db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr)) {
zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
need_wait = B_TRUE;
}
err = dbuf_read_impl(db, zio, flags, dblt, FTAG);
/*
* dbuf_read_impl has dropped db_mtx and our parent's rwlock
* for us
*/
if (!err && prefetch) {
dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE,
db->db_state != DB_CACHED,
flags & DB_RF_HAVESTRUCT);
}
DB_DNODE_EXIT(db);
DBUF_STAT_BUMP(hash_misses);
/*
* If we created a zio_root we must execute it to avoid
* leaking it, even if it isn't attached to any work due
* to an error in dbuf_read_impl().
*/
if (need_wait) {
if (err == 0)
err = zio_wait(zio);
else
VERIFY0(zio_wait(zio));
}
} else {
/*
* Another reader came in while the dbuf was in flight
* between UNCACHED and CACHED. Either a writer will finish
* writing the buffer (sending the dbuf to CACHED) or the
* first reader's request will reach the read_done callback
* and send the dbuf to CACHED. Otherwise, a failure
* occurred and the dbuf went to UNCACHED.
*/
mutex_exit(&db->db_mtx);
if (prefetch) {
dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE,
B_TRUE, flags & DB_RF_HAVESTRUCT);
}
DB_DNODE_EXIT(db);
DBUF_STAT_BUMP(hash_misses);
/* Skip the wait per the caller's request. */
if ((flags & DB_RF_NEVERWAIT) == 0) {
mutex_enter(&db->db_mtx);
while (db->db_state == DB_READ ||
db->db_state == DB_FILL) {
ASSERT(db->db_state == DB_READ ||
(flags & DB_RF_HAVESTRUCT) == 0);
DTRACE_PROBE2(blocked__read, dmu_buf_impl_t *,
db, zio_t *, zio);
cv_wait(&db->db_changed, &db->db_mtx);
}
if (db->db_state == DB_UNCACHED)
err = SET_ERROR(EIO);
mutex_exit(&db->db_mtx);
}
}
return (err);
}
static void
dbuf_noread(dmu_buf_impl_t *db)
{
ASSERT(!zfs_refcount_is_zero(&db->db_holds));
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
mutex_enter(&db->db_mtx);
while (db->db_state == DB_READ || db->db_state == DB_FILL)
cv_wait(&db->db_changed, &db->db_mtx);
if (db->db_state == DB_UNCACHED) {
ASSERT(db->db_buf == NULL);
ASSERT(db->db.db_data == NULL);
dbuf_set_data(db, dbuf_alloc_arcbuf(db));
db->db_state = DB_FILL;
DTRACE_SET_STATE(db, "assigning filled buffer");
} else if (db->db_state == DB_NOFILL) {
dbuf_clear_data(db);
} else {
ASSERT3U(db->db_state, ==, DB_CACHED);
}
mutex_exit(&db->db_mtx);
}
void
dbuf_unoverride(dbuf_dirty_record_t *dr)
{
dmu_buf_impl_t *db = dr->dr_dbuf;
blkptr_t *bp = &dr->dt.dl.dr_overridden_by;
uint64_t txg = dr->dr_txg;
ASSERT(MUTEX_HELD(&db->db_mtx));
/*
* This assert is valid because dmu_sync() expects to be called by
* a zilog's get_data while holding a range lock. This call only
* comes from dbuf_dirty() callers who must also hold a range lock.
*/
ASSERT(dr->dt.dl.dr_override_state != DR_IN_DMU_SYNC);
ASSERT(db->db_level == 0);
if (db->db_blkid == DMU_BONUS_BLKID ||
dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN)
return;
ASSERT(db->db_data_pending != dr);
/* free this block */
if (!BP_IS_HOLE(bp) && !dr->dt.dl.dr_nopwrite)
zio_free(db->db_objset->os_spa, txg, bp);
dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
dr->dt.dl.dr_nopwrite = B_FALSE;
dr->dt.dl.dr_has_raw_params = B_FALSE;
/*
* Release the already-written buffer, so we leave it in
* a consistent dirty state. Note that all callers are
* modifying the buffer, so they will immediately do
* another (redundant) arc_release(). Therefore, leave
* the buf thawed to save the effort of freezing &
* immediately re-thawing it.
*/
arc_release(dr->dt.dl.dr_data, db);
}
/*
* Evict (if its unreferenced) or clear (if its referenced) any level-0
* data blocks in the free range, so that any future readers will find
* empty blocks.
*/
void
dbuf_free_range(dnode_t *dn, uint64_t start_blkid, uint64_t end_blkid,
dmu_tx_t *tx)
{
dmu_buf_impl_t *db_search;
dmu_buf_impl_t *db, *db_next;
uint64_t txg = tx->tx_txg;
avl_index_t where;
dbuf_dirty_record_t *dr;
if (end_blkid > dn->dn_maxblkid &&
!(start_blkid == DMU_SPILL_BLKID || end_blkid == DMU_SPILL_BLKID))
end_blkid = dn->dn_maxblkid;
dprintf_dnode(dn, "start=%llu end=%llu\n", (u_longlong_t)start_blkid,
(u_longlong_t)end_blkid);
db_search = kmem_alloc(sizeof (dmu_buf_impl_t), KM_SLEEP);
db_search->db_level = 0;
db_search->db_blkid = start_blkid;
db_search->db_state = DB_SEARCH;
mutex_enter(&dn->dn_dbufs_mtx);
db = avl_find(&dn->dn_dbufs, db_search, &where);
ASSERT3P(db, ==, NULL);
db = avl_nearest(&dn->dn_dbufs, where, AVL_AFTER);
for (; db != NULL; db = db_next) {
db_next = AVL_NEXT(&dn->dn_dbufs, db);
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
if (db->db_level != 0 || db->db_blkid > end_blkid) {
break;
}
ASSERT3U(db->db_blkid, >=, start_blkid);
/* found a level 0 buffer in the range */
mutex_enter(&db->db_mtx);
if (dbuf_undirty(db, tx)) {
/* mutex has been dropped and dbuf destroyed */
continue;
}
if (db->db_state == DB_UNCACHED ||
db->db_state == DB_NOFILL ||
db->db_state == DB_EVICTING) {
ASSERT(db->db.db_data == NULL);
mutex_exit(&db->db_mtx);
continue;
}
if (db->db_state == DB_READ || db->db_state == DB_FILL) {
/* will be handled in dbuf_read_done or dbuf_rele */
db->db_freed_in_flight = TRUE;
mutex_exit(&db->db_mtx);
continue;
}
if (zfs_refcount_count(&db->db_holds) == 0) {
ASSERT(db->db_buf);
dbuf_destroy(db);
continue;
}
/* The dbuf is referenced */
dr = list_head(&db->db_dirty_records);
if (dr != NULL) {
if (dr->dr_txg == txg) {
/*
* This buffer is "in-use", re-adjust the file
* size to reflect that this buffer may
* contain new data when we sync.
*/
if (db->db_blkid != DMU_SPILL_BLKID &&
db->db_blkid > dn->dn_maxblkid)
dn->dn_maxblkid = db->db_blkid;
dbuf_unoverride(dr);
} else {
/*
* This dbuf is not dirty in the open context.
* Either uncache it (if its not referenced in
* the open context) or reset its contents to
* empty.
*/
dbuf_fix_old_data(db, txg);
}
}
/* clear the contents if its cached */
if (db->db_state == DB_CACHED) {
ASSERT(db->db.db_data != NULL);
arc_release(db->db_buf, db);
rw_enter(&db->db_rwlock, RW_WRITER);
bzero(db->db.db_data, db->db.db_size);
rw_exit(&db->db_rwlock);
arc_buf_freeze(db->db_buf);
}
mutex_exit(&db->db_mtx);
}
kmem_free(db_search, sizeof (dmu_buf_impl_t));
mutex_exit(&dn->dn_dbufs_mtx);
}
void
dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx)
{
arc_buf_t *buf, *old_buf;
dbuf_dirty_record_t *dr;
int osize = db->db.db_size;
arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db);
dnode_t *dn;
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
/*
* XXX we should be doing a dbuf_read, checking the return
* value and returning that up to our callers
*/
dmu_buf_will_dirty(&db->db, tx);
/* create the data buffer for the new block */
buf = arc_alloc_buf(dn->dn_objset->os_spa, db, type, size);
/* copy old block data to the new block */
old_buf = db->db_buf;
bcopy(old_buf->b_data, buf->b_data, MIN(osize, size));
/* zero the remainder */
if (size > osize)
bzero((uint8_t *)buf->b_data + osize, size - osize);
mutex_enter(&db->db_mtx);
dbuf_set_data(db, buf);
arc_buf_destroy(old_buf, db);
db->db.db_size = size;
dr = list_head(&db->db_dirty_records);
/* dirty record added by dmu_buf_will_dirty() */
VERIFY(dr != NULL);
if (db->db_level == 0)
dr->dt.dl.dr_data = buf;
ASSERT3U(dr->dr_txg, ==, tx->tx_txg);
ASSERT3U(dr->dr_accounted, ==, osize);
dr->dr_accounted = size;
mutex_exit(&db->db_mtx);
dmu_objset_willuse_space(dn->dn_objset, size - osize, tx);
DB_DNODE_EXIT(db);
}
void
dbuf_release_bp(dmu_buf_impl_t *db)
{
objset_t *os __maybe_unused = db->db_objset;
ASSERT(dsl_pool_sync_context(dmu_objset_pool(os)));
ASSERT(arc_released(os->os_phys_buf) ||
list_link_active(&os->os_dsl_dataset->ds_synced_link));
ASSERT(db->db_parent == NULL || arc_released(db->db_parent->db_buf));
(void) arc_release(db->db_buf, db);
}
/*
* We already have a dirty record for this TXG, and we are being
* dirtied again.
*/
static void
dbuf_redirty(dbuf_dirty_record_t *dr)
{
dmu_buf_impl_t *db = dr->dr_dbuf;
ASSERT(MUTEX_HELD(&db->db_mtx));
if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID) {
/*
* If this buffer has already been written out,
* we now need to reset its state.
*/
dbuf_unoverride(dr);
if (db->db.db_object != DMU_META_DNODE_OBJECT &&
db->db_state != DB_NOFILL) {
/* Already released on initial dirty, so just thaw. */
ASSERT(arc_released(db->db_buf));
arc_buf_thaw(db->db_buf);
}
}
}
dbuf_dirty_record_t *
dbuf_dirty_lightweight(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx)
{
rw_enter(&dn->dn_struct_rwlock, RW_READER);
IMPLY(dn->dn_objset->os_raw_receive, dn->dn_maxblkid >= blkid);
dnode_new_blkid(dn, blkid, tx, B_TRUE, B_FALSE);
ASSERT(dn->dn_maxblkid >= blkid);
dbuf_dirty_record_t *dr = kmem_zalloc(sizeof (*dr), KM_SLEEP);
list_link_init(&dr->dr_dirty_node);
list_link_init(&dr->dr_dbuf_node);
dr->dr_dnode = dn;
dr->dr_txg = tx->tx_txg;
dr->dt.dll.dr_blkid = blkid;
dr->dr_accounted = dn->dn_datablksz;
/*
* There should not be any dbuf for the block that we're dirtying.
* Otherwise the buffer contents could be inconsistent between the
* dbuf and the lightweight dirty record.
*/
ASSERT3P(NULL, ==, dbuf_find(dn->dn_objset, dn->dn_object, 0, blkid));
mutex_enter(&dn->dn_mtx);
int txgoff = tx->tx_txg & TXG_MASK;
if (dn->dn_free_ranges[txgoff] != NULL) {
range_tree_clear(dn->dn_free_ranges[txgoff], blkid, 1);
}
if (dn->dn_nlevels == 1) {
ASSERT3U(blkid, <, dn->dn_nblkptr);
list_insert_tail(&dn->dn_dirty_records[txgoff], dr);
mutex_exit(&dn->dn_mtx);
rw_exit(&dn->dn_struct_rwlock);
dnode_setdirty(dn, tx);
} else {
mutex_exit(&dn->dn_mtx);
int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
dmu_buf_impl_t *parent_db = dbuf_hold_level(dn,
1, blkid >> epbs, FTAG);
rw_exit(&dn->dn_struct_rwlock);
if (parent_db == NULL) {
kmem_free(dr, sizeof (*dr));
return (NULL);
}
int err = dbuf_read(parent_db, NULL,
(DB_RF_NOPREFETCH | DB_RF_CANFAIL));
if (err != 0) {
dbuf_rele(parent_db, FTAG);
kmem_free(dr, sizeof (*dr));
return (NULL);
}
dbuf_dirty_record_t *parent_dr = dbuf_dirty(parent_db, tx);
dbuf_rele(parent_db, FTAG);
mutex_enter(&parent_dr->dt.di.dr_mtx);
ASSERT3U(parent_dr->dr_txg, ==, tx->tx_txg);
list_insert_tail(&parent_dr->dt.di.dr_children, dr);
mutex_exit(&parent_dr->dt.di.dr_mtx);
dr->dr_parent = parent_dr;
}
dmu_objset_willuse_space(dn->dn_objset, dr->dr_accounted, tx);
return (dr);
}
dbuf_dirty_record_t *
dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
{
dnode_t *dn;
objset_t *os;
dbuf_dirty_record_t *dr, *dr_next, *dr_head;
int txgoff = tx->tx_txg & TXG_MASK;
boolean_t drop_struct_rwlock = B_FALSE;
ASSERT(tx->tx_txg != 0);
ASSERT(!zfs_refcount_is_zero(&db->db_holds));
DMU_TX_DIRTY_BUF(tx, db);
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
/*
* Shouldn't dirty a regular buffer in syncing context. Private
* objects may be dirtied in syncing context, but only if they
* were already pre-dirtied in open context.
*/
#ifdef ZFS_DEBUG
if (dn->dn_objset->os_dsl_dataset != NULL) {
rrw_enter(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock,
RW_READER, FTAG);
}
ASSERT(!dmu_tx_is_syncing(tx) ||
BP_IS_HOLE(dn->dn_objset->os_rootbp) ||
DMU_OBJECT_IS_SPECIAL(dn->dn_object) ||
dn->dn_objset->os_dsl_dataset == NULL);
if (dn->dn_objset->os_dsl_dataset != NULL)
rrw_exit(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, FTAG);
#endif
/*
* We make this assert for private objects as well, but after we
* check if we're already dirty. They are allowed to re-dirty
* in syncing context.
*/
ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT ||
dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx ==
(dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN));
mutex_enter(&db->db_mtx);
/*
* XXX make this true for indirects too? The problem is that
* transactions created with dmu_tx_create_assigned() from
* syncing context don't bother holding ahead.
*/
ASSERT(db->db_level != 0 ||
db->db_state == DB_CACHED || db->db_state == DB_FILL ||
db->db_state == DB_NOFILL);
mutex_enter(&dn->dn_mtx);
dnode_set_dirtyctx(dn, tx, db);
if (tx->tx_txg > dn->dn_dirty_txg)
dn->dn_dirty_txg = tx->tx_txg;
mutex_exit(&dn->dn_mtx);
if (db->db_blkid == DMU_SPILL_BLKID)
dn->dn_have_spill = B_TRUE;
/*
* If this buffer is already dirty, we're done.
*/
dr_head = list_head(&db->db_dirty_records);
ASSERT(dr_head == NULL || dr_head->dr_txg <= tx->tx_txg ||
db->db.db_object == DMU_META_DNODE_OBJECT);
dr_next = dbuf_find_dirty_lte(db, tx->tx_txg);
if (dr_next && dr_next->dr_txg == tx->tx_txg) {
DB_DNODE_EXIT(db);
dbuf_redirty(dr_next);
mutex_exit(&db->db_mtx);
return (dr_next);
}
/*
* Only valid if not already dirty.
*/
ASSERT(dn->dn_object == 0 ||
dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx ==
(dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN));
ASSERT3U(dn->dn_nlevels, >, db->db_level);
/*
* We should only be dirtying in syncing context if it's the
* mos or we're initializing the os or it's a special object.
* However, we are allowed to dirty in syncing context provided
* we already dirtied it in open context. Hence we must make
* this assertion only if we're not already dirty.
*/
os = dn->dn_objset;
VERIFY3U(tx->tx_txg, <=, spa_final_dirty_txg(os->os_spa));
#ifdef ZFS_DEBUG
if (dn->dn_objset->os_dsl_dataset != NULL)
rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_READER, FTAG);
ASSERT(!dmu_tx_is_syncing(tx) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) ||
os->os_dsl_dataset == NULL || BP_IS_HOLE(os->os_rootbp));
if (dn->dn_objset->os_dsl_dataset != NULL)
rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG);
#endif
ASSERT(db->db.db_size != 0);
dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size);
if (db->db_blkid != DMU_BONUS_BLKID) {
dmu_objset_willuse_space(os, db->db.db_size, tx);
}
/*
* If this buffer is dirty in an old transaction group we need
* to make a copy of it so that the changes we make in this
* transaction group won't leak out when we sync the older txg.
*/
dr = kmem_zalloc(sizeof (dbuf_dirty_record_t), KM_SLEEP);
list_link_init(&dr->dr_dirty_node);
list_link_init(&dr->dr_dbuf_node);
dr->dr_dnode = dn;
if (db->db_level == 0) {
void *data_old = db->db_buf;
if (db->db_state != DB_NOFILL) {
if (db->db_blkid == DMU_BONUS_BLKID) {
dbuf_fix_old_data(db, tx->tx_txg);
data_old = db->db.db_data;
} else if (db->db.db_object != DMU_META_DNODE_OBJECT) {
/*
* Release the data buffer from the cache so
* that we can modify it without impacting
* possible other users of this cached data
* block. Note that indirect blocks and
* private objects are not released until the
* syncing state (since they are only modified
* then).
*/
arc_release(db->db_buf, db);
dbuf_fix_old_data(db, tx->tx_txg);
data_old = db->db_buf;
}
ASSERT(data_old != NULL);
}
dr->dt.dl.dr_data = data_old;
} else {
mutex_init(&dr->dt.di.dr_mtx, NULL, MUTEX_NOLOCKDEP, NULL);
list_create(&dr->dt.di.dr_children,
sizeof (dbuf_dirty_record_t),
offsetof(dbuf_dirty_record_t, dr_dirty_node));
}
if (db->db_blkid != DMU_BONUS_BLKID)
dr->dr_accounted = db->db.db_size;
dr->dr_dbuf = db;
dr->dr_txg = tx->tx_txg;
list_insert_before(&db->db_dirty_records, dr_next, dr);
/*
* We could have been freed_in_flight between the dbuf_noread
* and dbuf_dirty. We win, as though the dbuf_noread() had
* happened after the free.
*/
if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID &&
db->db_blkid != DMU_SPILL_BLKID) {
mutex_enter(&dn->dn_mtx);
if (dn->dn_free_ranges[txgoff] != NULL) {
range_tree_clear(dn->dn_free_ranges[txgoff],
db->db_blkid, 1);
}
mutex_exit(&dn->dn_mtx);
db->db_freed_in_flight = FALSE;
}
/*
* This buffer is now part of this txg
*/
dbuf_add_ref(db, (void *)(uintptr_t)tx->tx_txg);
db->db_dirtycnt += 1;
ASSERT3U(db->db_dirtycnt, <=, 3);
mutex_exit(&db->db_mtx);
if (db->db_blkid == DMU_BONUS_BLKID ||
db->db_blkid == DMU_SPILL_BLKID) {
mutex_enter(&dn->dn_mtx);
ASSERT(!list_link_active(&dr->dr_dirty_node));
list_insert_tail(&dn->dn_dirty_records[txgoff], dr);
mutex_exit(&dn->dn_mtx);
dnode_setdirty(dn, tx);
DB_DNODE_EXIT(db);
return (dr);
}
if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) {
rw_enter(&dn->dn_struct_rwlock, RW_READER);
drop_struct_rwlock = B_TRUE;
}
/*
* If we are overwriting a dedup BP, then unless it is snapshotted,
* when we get to syncing context we will need to decrement its
* refcount in the DDT. Prefetch the relevant DDT block so that
* syncing context won't have to wait for the i/o.
*/
if (db->db_blkptr != NULL) {
db_lock_type_t dblt = dmu_buf_lock_parent(db, RW_READER, FTAG);
ddt_prefetch(os->os_spa, db->db_blkptr);
dmu_buf_unlock_parent(db, dblt, FTAG);
}
/*
* We need to hold the dn_struct_rwlock to make this assertion,
* because it protects dn_phys / dn_next_nlevels from changing.
*/
ASSERT((dn->dn_phys->dn_nlevels == 0 && db->db_level == 0) ||
dn->dn_phys->dn_nlevels > db->db_level ||
dn->dn_next_nlevels[txgoff] > db->db_level ||
dn->dn_next_nlevels[(tx->tx_txg-1) & TXG_MASK] > db->db_level ||
dn->dn_next_nlevels[(tx->tx_txg-2) & TXG_MASK] > db->db_level);
if (db->db_level == 0) {
ASSERT(!db->db_objset->os_raw_receive ||
dn->dn_maxblkid >= db->db_blkid);
dnode_new_blkid(dn, db->db_blkid, tx,
drop_struct_rwlock, B_FALSE);
ASSERT(dn->dn_maxblkid >= db->db_blkid);
}
if (db->db_level+1 < dn->dn_nlevels) {
dmu_buf_impl_t *parent = db->db_parent;
dbuf_dirty_record_t *di;
int parent_held = FALSE;
if (db->db_parent == NULL || db->db_parent == dn->dn_dbuf) {
int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
parent = dbuf_hold_level(dn, db->db_level + 1,
db->db_blkid >> epbs, FTAG);
ASSERT(parent != NULL);
parent_held = TRUE;
}
if (drop_struct_rwlock)
rw_exit(&dn->dn_struct_rwlock);
ASSERT3U(db->db_level + 1, ==, parent->db_level);
di = dbuf_dirty(parent, tx);
if (parent_held)
dbuf_rele(parent, FTAG);
mutex_enter(&db->db_mtx);
/*
* Since we've dropped the mutex, it's possible that
* dbuf_undirty() might have changed this out from under us.
*/
if (list_head(&db->db_dirty_records) == dr ||
dn->dn_object == DMU_META_DNODE_OBJECT) {
mutex_enter(&di->dt.di.dr_mtx);
ASSERT3U(di->dr_txg, ==, tx->tx_txg);
ASSERT(!list_link_active(&dr->dr_dirty_node));
list_insert_tail(&di->dt.di.dr_children, dr);
mutex_exit(&di->dt.di.dr_mtx);
dr->dr_parent = di;
}
mutex_exit(&db->db_mtx);
} else {
ASSERT(db->db_level + 1 == dn->dn_nlevels);
ASSERT(db->db_blkid < dn->dn_nblkptr);
ASSERT(db->db_parent == NULL || db->db_parent == dn->dn_dbuf);
mutex_enter(&dn->dn_mtx);
ASSERT(!list_link_active(&dr->dr_dirty_node));
list_insert_tail(&dn->dn_dirty_records[txgoff], dr);
mutex_exit(&dn->dn_mtx);
if (drop_struct_rwlock)
rw_exit(&dn->dn_struct_rwlock);
}
dnode_setdirty(dn, tx);
DB_DNODE_EXIT(db);
return (dr);
}
static void
dbuf_undirty_bonus(dbuf_dirty_record_t *dr)
{
dmu_buf_impl_t *db = dr->dr_dbuf;
if (dr->dt.dl.dr_data != db->db.db_data) {
struct dnode *dn = dr->dr_dnode;
int max_bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots);
kmem_free(dr->dt.dl.dr_data, max_bonuslen);
arc_space_return(max_bonuslen, ARC_SPACE_BONUS);
}
db->db_data_pending = NULL;
ASSERT(list_next(&db->db_dirty_records, dr) == NULL);
list_remove(&db->db_dirty_records, dr);
if (dr->dr_dbuf->db_level != 0) {
mutex_destroy(&dr->dt.di.dr_mtx);
list_destroy(&dr->dt.di.dr_children);
}
kmem_free(dr, sizeof (dbuf_dirty_record_t));
ASSERT3U(db->db_dirtycnt, >, 0);
db->db_dirtycnt -= 1;
}
/*
* Undirty a buffer in the transaction group referenced by the given
* transaction. Return whether this evicted the dbuf.
*/
static boolean_t
dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
{
uint64_t txg = tx->tx_txg;
ASSERT(txg != 0);
/*
* Due to our use of dn_nlevels below, this can only be called
* in open context, unless we are operating on the MOS.
* From syncing context, dn_nlevels may be different from the
* dn_nlevels used when dbuf was dirtied.
*/
ASSERT(db->db_objset ==
dmu_objset_pool(db->db_objset)->dp_meta_objset ||
txg != spa_syncing_txg(dmu_objset_spa(db->db_objset)));
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
ASSERT0(db->db_level);
ASSERT(MUTEX_HELD(&db->db_mtx));
/*
* If this buffer is not dirty, we're done.
*/
dbuf_dirty_record_t *dr = dbuf_find_dirty_eq(db, txg);
if (dr == NULL)
return (B_FALSE);
ASSERT(dr->dr_dbuf == db);
dnode_t *dn = dr->dr_dnode;
dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size);
ASSERT(db->db.db_size != 0);
dsl_pool_undirty_space(dmu_objset_pool(dn->dn_objset),
dr->dr_accounted, txg);
list_remove(&db->db_dirty_records, dr);
/*
* Note that there are three places in dbuf_dirty()
* where this dirty record may be put on a list.
* Make sure to do a list_remove corresponding to
* every one of those list_insert calls.
*/
if (dr->dr_parent) {
mutex_enter(&dr->dr_parent->dt.di.dr_mtx);
list_remove(&dr->dr_parent->dt.di.dr_children, dr);
mutex_exit(&dr->dr_parent->dt.di.dr_mtx);
} else if (db->db_blkid == DMU_SPILL_BLKID ||
db->db_level + 1 == dn->dn_nlevels) {
ASSERT(db->db_blkptr == NULL || db->db_parent == dn->dn_dbuf);
mutex_enter(&dn->dn_mtx);
list_remove(&dn->dn_dirty_records[txg & TXG_MASK], dr);
mutex_exit(&dn->dn_mtx);
}
if (db->db_state != DB_NOFILL) {
dbuf_unoverride(dr);
ASSERT(db->db_buf != NULL);
ASSERT(dr->dt.dl.dr_data != NULL);
if (dr->dt.dl.dr_data != db->db_buf)
arc_buf_destroy(dr->dt.dl.dr_data, db);
}
kmem_free(dr, sizeof (dbuf_dirty_record_t));
ASSERT(db->db_dirtycnt > 0);
db->db_dirtycnt -= 1;
if (zfs_refcount_remove(&db->db_holds, (void *)(uintptr_t)txg) == 0) {
ASSERT(db->db_state == DB_NOFILL || arc_released(db->db_buf));
dbuf_destroy(db);
return (B_TRUE);
}
return (B_FALSE);
}
static void
dmu_buf_will_dirty_impl(dmu_buf_t *db_fake, int flags, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
ASSERT(tx->tx_txg != 0);
ASSERT(!zfs_refcount_is_zero(&db->db_holds));
/*
* Quick check for dirtiness. For already dirty blocks, this
* reduces runtime of this function by >90%, and overall performance
* by 50% for some workloads (e.g. file deletion with indirect blocks
* cached).
*/
mutex_enter(&db->db_mtx);
if (db->db_state == DB_CACHED) {
dbuf_dirty_record_t *dr = dbuf_find_dirty_eq(db, tx->tx_txg);
/*
* It's possible that it is already dirty but not cached,
* because there are some calls to dbuf_dirty() that don't
* go through dmu_buf_will_dirty().
*/
if (dr != NULL) {
/* This dbuf is already dirty and cached. */
dbuf_redirty(dr);
mutex_exit(&db->db_mtx);
return;
}
}
mutex_exit(&db->db_mtx);
DB_DNODE_ENTER(db);
if (RW_WRITE_HELD(&DB_DNODE(db)->dn_struct_rwlock))
flags |= DB_RF_HAVESTRUCT;
DB_DNODE_EXIT(db);
(void) dbuf_read(db, NULL, flags);
(void) dbuf_dirty(db, tx);
}
void
dmu_buf_will_dirty(dmu_buf_t *db_fake, dmu_tx_t *tx)
{
dmu_buf_will_dirty_impl(db_fake,
DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH, tx);
}
boolean_t
dmu_buf_is_dirty(dmu_buf_t *db_fake, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
dbuf_dirty_record_t *dr;
mutex_enter(&db->db_mtx);
dr = dbuf_find_dirty_eq(db, tx->tx_txg);
mutex_exit(&db->db_mtx);
return (dr != NULL);
}
void
dmu_buf_will_not_fill(dmu_buf_t *db_fake, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
db->db_state = DB_NOFILL;
DTRACE_SET_STATE(db, "allocating NOFILL buffer");
dmu_buf_will_fill(db_fake, tx);
}
void
dmu_buf_will_fill(dmu_buf_t *db_fake, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
ASSERT(tx->tx_txg != 0);
ASSERT(db->db_level == 0);
ASSERT(!zfs_refcount_is_zero(&db->db_holds));
ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT ||
dmu_tx_private_ok(tx));
dbuf_noread(db);
(void) dbuf_dirty(db, tx);
}
/*
* This function is effectively the same as dmu_buf_will_dirty(), but
* indicates the caller expects raw encrypted data in the db, and provides
* the crypt params (byteorder, salt, iv, mac) which should be stored in the
* blkptr_t when this dbuf is written. This is only used for blocks of
* dnodes, during raw receive.
*/
void
dmu_buf_set_crypt_params(dmu_buf_t *db_fake, boolean_t byteorder,
const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
dbuf_dirty_record_t *dr;
/*
* dr_has_raw_params is only processed for blocks of dnodes
* (see dbuf_sync_dnode_leaf_crypt()).
*/
ASSERT3U(db->db.db_object, ==, DMU_META_DNODE_OBJECT);
ASSERT3U(db->db_level, ==, 0);
ASSERT(db->db_objset->os_raw_receive);
dmu_buf_will_dirty_impl(db_fake,
DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH | DB_RF_NO_DECRYPT, tx);
dr = dbuf_find_dirty_eq(db, tx->tx_txg);
ASSERT3P(dr, !=, NULL);
dr->dt.dl.dr_has_raw_params = B_TRUE;
dr->dt.dl.dr_byteorder = byteorder;
bcopy(salt, dr->dt.dl.dr_salt, ZIO_DATA_SALT_LEN);
bcopy(iv, dr->dt.dl.dr_iv, ZIO_DATA_IV_LEN);
bcopy(mac, dr->dt.dl.dr_mac, ZIO_DATA_MAC_LEN);
}
static void
dbuf_override_impl(dmu_buf_impl_t *db, const blkptr_t *bp, dmu_tx_t *tx)
{
struct dirty_leaf *dl;
dbuf_dirty_record_t *dr;
dr = list_head(&db->db_dirty_records);
ASSERT3U(dr->dr_txg, ==, tx->tx_txg);
dl = &dr->dt.dl;
dl->dr_overridden_by = *bp;
dl->dr_override_state = DR_OVERRIDDEN;
dl->dr_overridden_by.blk_birth = dr->dr_txg;
}
/* ARGSUSED */
void
dmu_buf_fill_done(dmu_buf_t *dbuf, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf;
dbuf_states_t old_state;
mutex_enter(&db->db_mtx);
DBUF_VERIFY(db);
old_state = db->db_state;
db->db_state = DB_CACHED;
if (old_state == DB_FILL) {
if (db->db_level == 0 && db->db_freed_in_flight) {
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
/* we were freed while filling */
/* XXX dbuf_undirty? */
bzero(db->db.db_data, db->db.db_size);
db->db_freed_in_flight = FALSE;
DTRACE_SET_STATE(db,
"fill done handling freed in flight");
} else {
DTRACE_SET_STATE(db, "fill done");
}
cv_broadcast(&db->db_changed);
}
mutex_exit(&db->db_mtx);
}
void
dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
bp_embedded_type_t etype, enum zio_compress comp,
int uncompressed_size, int compressed_size, int byteorder,
dmu_tx_t *tx)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf;
struct dirty_leaf *dl;
dmu_object_type_t type;
dbuf_dirty_record_t *dr;
if (etype == BP_EMBEDDED_TYPE_DATA) {
ASSERT(spa_feature_is_active(dmu_objset_spa(db->db_objset),
SPA_FEATURE_EMBEDDED_DATA));
}
DB_DNODE_ENTER(db);
type = DB_DNODE(db)->dn_type;
DB_DNODE_EXIT(db);
ASSERT0(db->db_level);
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
dmu_buf_will_not_fill(dbuf, tx);
dr = list_head(&db->db_dirty_records);
ASSERT3U(dr->dr_txg, ==, tx->tx_txg);
dl = &dr->dt.dl;
encode_embedded_bp_compressed(&dl->dr_overridden_by,
data, comp, uncompressed_size, compressed_size);
BPE_SET_ETYPE(&dl->dr_overridden_by, etype);
BP_SET_TYPE(&dl->dr_overridden_by, type);
BP_SET_LEVEL(&dl->dr_overridden_by, 0);
BP_SET_BYTEORDER(&dl->dr_overridden_by, byteorder);
dl->dr_override_state = DR_OVERRIDDEN;
dl->dr_overridden_by.blk_birth = dr->dr_txg;
}
void
dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf;
dmu_object_type_t type;
ASSERT(dsl_dataset_feature_is_active(db->db_objset->os_dsl_dataset,
SPA_FEATURE_REDACTED_DATASETS));
DB_DNODE_ENTER(db);
type = DB_DNODE(db)->dn_type;
DB_DNODE_EXIT(db);
ASSERT0(db->db_level);
dmu_buf_will_not_fill(dbuf, tx);
blkptr_t bp = { { { {0} } } };
BP_SET_TYPE(&bp, type);
BP_SET_LEVEL(&bp, 0);
BP_SET_BIRTH(&bp, tx->tx_txg, 0);
BP_SET_REDACTED(&bp);
BPE_SET_LSIZE(&bp, dbuf->db_size);
dbuf_override_impl(db, &bp, tx);
}
/*
* Directly assign a provided arc buf to a given dbuf if it's not referenced
* by anybody except our caller. Otherwise copy arcbuf's contents to dbuf.
*/
void
dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx)
{
ASSERT(!zfs_refcount_is_zero(&db->db_holds));
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
ASSERT(db->db_level == 0);
ASSERT3U(dbuf_is_metadata(db), ==, arc_is_metadata(buf));
ASSERT(buf != NULL);
ASSERT3U(arc_buf_lsize(buf), ==, db->db.db_size);
ASSERT(tx->tx_txg != 0);
arc_return_buf(buf, db);
ASSERT(arc_released(buf));
mutex_enter(&db->db_mtx);
while (db->db_state == DB_READ || db->db_state == DB_FILL)
cv_wait(&db->db_changed, &db->db_mtx);
ASSERT(db->db_state == DB_CACHED || db->db_state == DB_UNCACHED);
if (db->db_state == DB_CACHED &&
zfs_refcount_count(&db->db_holds) - 1 > db->db_dirtycnt) {
/*
* In practice, we will never have a case where we have an
* encrypted arc buffer while additional holds exist on the
* dbuf. We don't handle this here so we simply assert that
* fact instead.
*/
ASSERT(!arc_is_encrypted(buf));
mutex_exit(&db->db_mtx);
(void) dbuf_dirty(db, tx);
bcopy(buf->b_data, db->db.db_data, db->db.db_size);
arc_buf_destroy(buf, db);
return;
}
if (db->db_state == DB_CACHED) {
dbuf_dirty_record_t *dr = list_head(&db->db_dirty_records);
ASSERT(db->db_buf != NULL);
if (dr != NULL && dr->dr_txg == tx->tx_txg) {
ASSERT(dr->dt.dl.dr_data == db->db_buf);
if (!arc_released(db->db_buf)) {
ASSERT(dr->dt.dl.dr_override_state ==
DR_OVERRIDDEN);
arc_release(db->db_buf, db);
}
dr->dt.dl.dr_data = buf;
arc_buf_destroy(db->db_buf, db);
} else if (dr == NULL || dr->dt.dl.dr_data != db->db_buf) {
arc_release(db->db_buf, db);
arc_buf_destroy(db->db_buf, db);
}
db->db_buf = NULL;
}
ASSERT(db->db_buf == NULL);
dbuf_set_data(db, buf);
db->db_state = DB_FILL;
DTRACE_SET_STATE(db, "filling assigned arcbuf");
mutex_exit(&db->db_mtx);
(void) dbuf_dirty(db, tx);
dmu_buf_fill_done(&db->db, tx);
}
void
dbuf_destroy(dmu_buf_impl_t *db)
{
dnode_t *dn;
dmu_buf_impl_t *parent = db->db_parent;
dmu_buf_impl_t *dndb;
ASSERT(MUTEX_HELD(&db->db_mtx));
ASSERT(zfs_refcount_is_zero(&db->db_holds));
if (db->db_buf != NULL) {
arc_buf_destroy(db->db_buf, db);
db->db_buf = NULL;
}
if (db->db_blkid == DMU_BONUS_BLKID) {
int slots = DB_DNODE(db)->dn_num_slots;
int bonuslen = DN_SLOTS_TO_BONUSLEN(slots);
if (db->db.db_data != NULL) {
kmem_free(db->db.db_data, bonuslen);
arc_space_return(bonuslen, ARC_SPACE_BONUS);
db->db_state = DB_UNCACHED;
DTRACE_SET_STATE(db, "buffer cleared");
}
}
dbuf_clear_data(db);
if (multilist_link_active(&db->db_cache_link)) {
ASSERT(db->db_caching_status == DB_DBUF_CACHE ||
db->db_caching_status == DB_DBUF_METADATA_CACHE);
multilist_remove(&dbuf_caches[db->db_caching_status].cache, db);
(void) zfs_refcount_remove_many(
&dbuf_caches[db->db_caching_status].size,
db->db.db_size, db);
if (db->db_caching_status == DB_DBUF_METADATA_CACHE) {
DBUF_STAT_BUMPDOWN(metadata_cache_count);
} else {
DBUF_STAT_BUMPDOWN(cache_levels[db->db_level]);
DBUF_STAT_BUMPDOWN(cache_count);
DBUF_STAT_DECR(cache_levels_bytes[db->db_level],
db->db.db_size);
}
db->db_caching_status = DB_NO_CACHE;
}
ASSERT(db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL);
ASSERT(db->db_data_pending == NULL);
ASSERT(list_is_empty(&db->db_dirty_records));
db->db_state = DB_EVICTING;
DTRACE_SET_STATE(db, "buffer eviction started");
db->db_blkptr = NULL;
/*
* Now that db_state is DB_EVICTING, nobody else can find this via
* the hash table. We can now drop db_mtx, which allows us to
* acquire the dn_dbufs_mtx.
*/
mutex_exit(&db->db_mtx);
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
dndb = dn->dn_dbuf;
if (db->db_blkid != DMU_BONUS_BLKID) {
boolean_t needlock = !MUTEX_HELD(&dn->dn_dbufs_mtx);
if (needlock)
mutex_enter_nested(&dn->dn_dbufs_mtx,
NESTED_SINGLE);
avl_remove(&dn->dn_dbufs, db);
membar_producer();
DB_DNODE_EXIT(db);
if (needlock)
mutex_exit(&dn->dn_dbufs_mtx);
/*
* Decrementing the dbuf count means that the hold corresponding
* to the removed dbuf is no longer discounted in dnode_move(),
* so the dnode cannot be moved until after we release the hold.
* The membar_producer() ensures visibility of the decremented
* value in dnode_move(), since DB_DNODE_EXIT doesn't actually
* release any lock.
*/
mutex_enter(&dn->dn_mtx);
dnode_rele_and_unlock(dn, db, B_TRUE);
db->db_dnode_handle = NULL;
dbuf_hash_remove(db);
} else {
DB_DNODE_EXIT(db);
}
ASSERT(zfs_refcount_is_zero(&db->db_holds));
db->db_parent = NULL;
ASSERT(db->db_buf == NULL);
ASSERT(db->db.db_data == NULL);
ASSERT(db->db_hash_next == NULL);
ASSERT(db->db_blkptr == NULL);
ASSERT(db->db_data_pending == NULL);
ASSERT3U(db->db_caching_status, ==, DB_NO_CACHE);
ASSERT(!multilist_link_active(&db->db_cache_link));
kmem_cache_free(dbuf_kmem_cache, db);
arc_space_return(sizeof (dmu_buf_impl_t), ARC_SPACE_DBUF);
/*
* If this dbuf is referenced from an indirect dbuf,
* decrement the ref count on the indirect dbuf.
*/
if (parent && parent != dndb) {
mutex_enter(&parent->db_mtx);
dbuf_rele_and_unlock(parent, db, B_TRUE);
}
}
/*
* Note: While bpp will always be updated if the function returns success,
* parentp will not be updated if the dnode does not have dn_dbuf filled in;
* this happens when the dnode is the meta-dnode, or {user|group|project}used
* object.
*/
__attribute__((always_inline))
static inline int
dbuf_findbp(dnode_t *dn, int level, uint64_t blkid, int fail_sparse,
dmu_buf_impl_t **parentp, blkptr_t **bpp)
{
*parentp = NULL;
*bpp = NULL;
ASSERT(blkid != DMU_BONUS_BLKID);
if (blkid == DMU_SPILL_BLKID) {
mutex_enter(&dn->dn_mtx);
if (dn->dn_have_spill &&
(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR))
*bpp = DN_SPILL_BLKPTR(dn->dn_phys);
else
*bpp = NULL;
dbuf_add_ref(dn->dn_dbuf, NULL);
*parentp = dn->dn_dbuf;
mutex_exit(&dn->dn_mtx);
return (0);
}
int nlevels =
(dn->dn_phys->dn_nlevels == 0) ? 1 : dn->dn_phys->dn_nlevels;
int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
ASSERT3U(level * epbs, <, 64);
ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock));
/*
* This assertion shouldn't trip as long as the max indirect block size
* is less than 1M. The reason for this is that up to that point,
* the number of levels required to address an entire object with blocks
* of size SPA_MINBLOCKSIZE satisfies nlevels * epbs + 1 <= 64. In
* other words, if N * epbs + 1 > 64, then if (N-1) * epbs + 1 > 55
* (i.e. we can address the entire object), objects will all use at most
* N-1 levels and the assertion won't overflow. However, once epbs is
* 13, 4 * 13 + 1 = 53, but 5 * 13 + 1 = 66. Then, 4 levels will not be
* enough to address an entire object, so objects will have 5 levels,
* but then this assertion will overflow.
*
* All this is to say that if we ever increase DN_MAX_INDBLKSHIFT, we
* need to redo this logic to handle overflows.
*/
ASSERT(level >= nlevels ||
((nlevels - level - 1) * epbs) +
highbit64(dn->dn_phys->dn_nblkptr) <= 64);
if (level >= nlevels ||
blkid >= ((uint64_t)dn->dn_phys->dn_nblkptr <<
((nlevels - level - 1) * epbs)) ||
(fail_sparse &&
blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs)))) {
/* the buffer has no parent yet */
return (SET_ERROR(ENOENT));
} else if (level < nlevels-1) {
/* this block is referenced from an indirect block */
int err;
err = dbuf_hold_impl(dn, level + 1,
blkid >> epbs, fail_sparse, FALSE, NULL, parentp);
if (err)
return (err);
err = dbuf_read(*parentp, NULL,
(DB_RF_HAVESTRUCT | DB_RF_NOPREFETCH | DB_RF_CANFAIL));
if (err) {
dbuf_rele(*parentp, NULL);
*parentp = NULL;
return (err);
}
rw_enter(&(*parentp)->db_rwlock, RW_READER);
*bpp = ((blkptr_t *)(*parentp)->db.db_data) +
(blkid & ((1ULL << epbs) - 1));
if (blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs)))
ASSERT(BP_IS_HOLE(*bpp));
rw_exit(&(*parentp)->db_rwlock);
return (0);
} else {
/* the block is referenced from the dnode */
ASSERT3U(level, ==, nlevels-1);
ASSERT(dn->dn_phys->dn_nblkptr == 0 ||
blkid < dn->dn_phys->dn_nblkptr);
if (dn->dn_dbuf) {
dbuf_add_ref(dn->dn_dbuf, NULL);
*parentp = dn->dn_dbuf;
}
*bpp = &dn->dn_phys->dn_blkptr[blkid];
return (0);
}
}
static dmu_buf_impl_t *
dbuf_create(dnode_t *dn, uint8_t level, uint64_t blkid,
dmu_buf_impl_t *parent, blkptr_t *blkptr)
{
objset_t *os = dn->dn_objset;
dmu_buf_impl_t *db, *odb;
ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock));
ASSERT(dn->dn_type != DMU_OT_NONE);
db = kmem_cache_alloc(dbuf_kmem_cache, KM_SLEEP);
list_create(&db->db_dirty_records, sizeof (dbuf_dirty_record_t),
offsetof(dbuf_dirty_record_t, dr_dbuf_node));
db->db_objset = os;
db->db.db_object = dn->dn_object;
db->db_level = level;
db->db_blkid = blkid;
db->db_dirtycnt = 0;
db->db_dnode_handle = dn->dn_handle;
db->db_parent = parent;
db->db_blkptr = blkptr;
db->db_user = NULL;
db->db_user_immediate_evict = FALSE;
db->db_freed_in_flight = FALSE;
db->db_pending_evict = FALSE;
if (blkid == DMU_BONUS_BLKID) {
ASSERT3P(parent, ==, dn->dn_dbuf);
db->db.db_size = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots) -
(dn->dn_nblkptr-1) * sizeof (blkptr_t);
ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen);
db->db.db_offset = DMU_BONUS_BLKID;
db->db_state = DB_UNCACHED;
DTRACE_SET_STATE(db, "bonus buffer created");
db->db_caching_status = DB_NO_CACHE;
/* the bonus dbuf is not placed in the hash table */
arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_DBUF);
return (db);
} else if (blkid == DMU_SPILL_BLKID) {
db->db.db_size = (blkptr != NULL) ?
BP_GET_LSIZE(blkptr) : SPA_MINBLOCKSIZE;
db->db.db_offset = 0;
} else {
int blocksize =
db->db_level ? 1 << dn->dn_indblkshift : dn->dn_datablksz;
db->db.db_size = blocksize;
db->db.db_offset = db->db_blkid * blocksize;
}
/*
* Hold the dn_dbufs_mtx while we get the new dbuf
* in the hash table *and* added to the dbufs list.
* This prevents a possible deadlock with someone
* trying to look up this dbuf before it's added to the
* dn_dbufs list.
*/
mutex_enter(&dn->dn_dbufs_mtx);
db->db_state = DB_EVICTING; /* not worth logging this state change */
if ((odb = dbuf_hash_insert(db)) != NULL) {
/* someone else inserted it first */
kmem_cache_free(dbuf_kmem_cache, db);
mutex_exit(&dn->dn_dbufs_mtx);
DBUF_STAT_BUMP(hash_insert_race);
return (odb);
}
avl_add(&dn->dn_dbufs, db);
db->db_state = DB_UNCACHED;
DTRACE_SET_STATE(db, "regular buffer created");
db->db_caching_status = DB_NO_CACHE;
mutex_exit(&dn->dn_dbufs_mtx);
arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_DBUF);
if (parent && parent != dn->dn_dbuf)
dbuf_add_ref(parent, db);
ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT ||
zfs_refcount_count(&dn->dn_holds) > 0);
(void) zfs_refcount_add(&dn->dn_holds, db);
dprintf_dbuf(db, "db=%p\n", db);
return (db);
}
/*
* This function returns a block pointer and information about the object,
* given a dnode and a block. This is a publicly accessible version of
* dbuf_findbp that only returns some information, rather than the
* dbuf. Note that the dnode passed in must be held, and the dn_struct_rwlock
* should be locked as (at least) a reader.
*/
int
dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid,
blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift)
{
dmu_buf_impl_t *dbp = NULL;
blkptr_t *bp2;
int err = 0;
ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock));
err = dbuf_findbp(dn, level, blkid, B_FALSE, &dbp, &bp2);
if (err == 0) {
*bp = *bp2;
if (dbp != NULL)
dbuf_rele(dbp, NULL);
if (datablkszsec != NULL)
*datablkszsec = dn->dn_phys->dn_datablkszsec;
if (indblkshift != NULL)
*indblkshift = dn->dn_phys->dn_indblkshift;
}
return (err);
}
typedef struct dbuf_prefetch_arg {
spa_t *dpa_spa; /* The spa to issue the prefetch in. */
zbookmark_phys_t dpa_zb; /* The target block to prefetch. */
int dpa_epbs; /* Entries (blkptr_t's) Per Block Shift. */
int dpa_curlevel; /* The current level that we're reading */
dnode_t *dpa_dnode; /* The dnode associated with the prefetch */
zio_priority_t dpa_prio; /* The priority I/Os should be issued at. */
zio_t *dpa_zio; /* The parent zio_t for all prefetches. */
arc_flags_t dpa_aflags; /* Flags to pass to the final prefetch. */
dbuf_prefetch_fn dpa_cb; /* prefetch completion callback */
void *dpa_arg; /* prefetch completion arg */
} dbuf_prefetch_arg_t;
static void
dbuf_prefetch_fini(dbuf_prefetch_arg_t *dpa, boolean_t io_done)
{
if (dpa->dpa_cb != NULL)
dpa->dpa_cb(dpa->dpa_arg, io_done);
kmem_free(dpa, sizeof (*dpa));
}
static void
dbuf_issue_final_prefetch_done(zio_t *zio, const zbookmark_phys_t *zb,
const blkptr_t *iobp, arc_buf_t *abuf, void *private)
{
dbuf_prefetch_arg_t *dpa = private;
dbuf_prefetch_fini(dpa, B_TRUE);
if (abuf != NULL)
arc_buf_destroy(abuf, private);
}
/*
* Actually issue the prefetch read for the block given.
*/
static void
dbuf_issue_final_prefetch(dbuf_prefetch_arg_t *dpa, blkptr_t *bp)
{
ASSERT(!BP_IS_REDACTED(bp) ||
dsl_dataset_feature_is_active(
dpa->dpa_dnode->dn_objset->os_dsl_dataset,
SPA_FEATURE_REDACTED_DATASETS));
if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) || BP_IS_REDACTED(bp))
return (dbuf_prefetch_fini(dpa, B_FALSE));
int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE;
arc_flags_t aflags =
dpa->dpa_aflags | ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH |
ARC_FLAG_NO_BUF;
/* dnodes are always read as raw and then converted later */
if (BP_GET_TYPE(bp) == DMU_OT_DNODE && BP_IS_PROTECTED(bp) &&
dpa->dpa_curlevel == 0)
zio_flags |= ZIO_FLAG_RAW;
ASSERT3U(dpa->dpa_curlevel, ==, BP_GET_LEVEL(bp));
ASSERT3U(dpa->dpa_curlevel, ==, dpa->dpa_zb.zb_level);
ASSERT(dpa->dpa_zio != NULL);
(void) arc_read(dpa->dpa_zio, dpa->dpa_spa, bp,
dbuf_issue_final_prefetch_done, dpa,
dpa->dpa_prio, zio_flags, &aflags, &dpa->dpa_zb);
}
/*
* Called when an indirect block above our prefetch target is read in. This
* will either read in the next indirect block down the tree or issue the actual
* prefetch if the next block down is our target.
*/
static void
dbuf_prefetch_indirect_done(zio_t *zio, const zbookmark_phys_t *zb,
const blkptr_t *iobp, arc_buf_t *abuf, void *private)
{
dbuf_prefetch_arg_t *dpa = private;
ASSERT3S(dpa->dpa_zb.zb_level, <, dpa->dpa_curlevel);
ASSERT3S(dpa->dpa_curlevel, >, 0);
if (abuf == NULL) {
ASSERT(zio == NULL || zio->io_error != 0);
return (dbuf_prefetch_fini(dpa, B_TRUE));
}
ASSERT(zio == NULL || zio->io_error == 0);
/*
* The dpa_dnode is only valid if we are called with a NULL
* zio. This indicates that the arc_read() returned without
* first calling zio_read() to issue a physical read. Once
* a physical read is made the dpa_dnode must be invalidated
* as the locks guarding it may have been dropped. If the
* dpa_dnode is still valid, then we want to add it to the dbuf
* cache. To do so, we must hold the dbuf associated with the block
* we just prefetched, read its contents so that we associate it
* with an arc_buf_t, and then release it.
*/
if (zio != NULL) {
ASSERT3S(BP_GET_LEVEL(zio->io_bp), ==, dpa->dpa_curlevel);
if (zio->io_flags & ZIO_FLAG_RAW_COMPRESS) {
ASSERT3U(BP_GET_PSIZE(zio->io_bp), ==, zio->io_size);
} else {
ASSERT3U(BP_GET_LSIZE(zio->io_bp), ==, zio->io_size);
}
ASSERT3P(zio->io_spa, ==, dpa->dpa_spa);
dpa->dpa_dnode = NULL;
} else if (dpa->dpa_dnode != NULL) {
uint64_t curblkid = dpa->dpa_zb.zb_blkid >>
(dpa->dpa_epbs * (dpa->dpa_curlevel -
dpa->dpa_zb.zb_level));
dmu_buf_impl_t *db = dbuf_hold_level(dpa->dpa_dnode,
dpa->dpa_curlevel, curblkid, FTAG);
if (db == NULL) {
arc_buf_destroy(abuf, private);
return (dbuf_prefetch_fini(dpa, B_TRUE));
}
(void) dbuf_read(db, NULL,
DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH | DB_RF_HAVESTRUCT);
dbuf_rele(db, FTAG);
}
dpa->dpa_curlevel--;
uint64_t nextblkid = dpa->dpa_zb.zb_blkid >>
(dpa->dpa_epbs * (dpa->dpa_curlevel - dpa->dpa_zb.zb_level));
blkptr_t *bp = ((blkptr_t *)abuf->b_data) +
P2PHASE(nextblkid, 1ULL << dpa->dpa_epbs);
ASSERT(!BP_IS_REDACTED(bp) ||
dsl_dataset_feature_is_active(
dpa->dpa_dnode->dn_objset->os_dsl_dataset,
SPA_FEATURE_REDACTED_DATASETS));
if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) {
dbuf_prefetch_fini(dpa, B_TRUE);
} else if (dpa->dpa_curlevel == dpa->dpa_zb.zb_level) {
ASSERT3U(nextblkid, ==, dpa->dpa_zb.zb_blkid);
dbuf_issue_final_prefetch(dpa, bp);
} else {
arc_flags_t iter_aflags = ARC_FLAG_NOWAIT;
zbookmark_phys_t zb;
/* flag if L2ARC eligible, l2arc_noprefetch then decides */
if (dpa->dpa_aflags & ARC_FLAG_L2CACHE)
iter_aflags |= ARC_FLAG_L2CACHE;
ASSERT3U(dpa->dpa_curlevel, ==, BP_GET_LEVEL(bp));
SET_BOOKMARK(&zb, dpa->dpa_zb.zb_objset,
dpa->dpa_zb.zb_object, dpa->dpa_curlevel, nextblkid);
(void) arc_read(dpa->dpa_zio, dpa->dpa_spa,
bp, dbuf_prefetch_indirect_done, dpa, dpa->dpa_prio,
ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
&iter_aflags, &zb);
}
arc_buf_destroy(abuf, private);
}
/*
* Issue prefetch reads for the given block on the given level. If the indirect
* blocks above that block are not in memory, we will read them in
* asynchronously. As a result, this call never blocks waiting for a read to
* complete. Note that the prefetch might fail if the dataset is encrypted and
* the encryption key is unmapped before the IO completes.
*/
int
dbuf_prefetch_impl(dnode_t *dn, int64_t level, uint64_t blkid,
zio_priority_t prio, arc_flags_t aflags, dbuf_prefetch_fn cb,
void *arg)
{
blkptr_t bp;
int epbs, nlevels, curlevel;
uint64_t curblkid;
ASSERT(blkid != DMU_BONUS_BLKID);
ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock));
if (blkid > dn->dn_maxblkid)
goto no_issue;
if (level == 0 && dnode_block_freed(dn, blkid))
goto no_issue;
/*
* This dnode hasn't been written to disk yet, so there's nothing to
* prefetch.
*/
nlevels = dn->dn_phys->dn_nlevels;
if (level >= nlevels || dn->dn_phys->dn_nblkptr == 0)
goto no_issue;
epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT;
if (dn->dn_phys->dn_maxblkid < blkid << (epbs * level))
goto no_issue;
dmu_buf_impl_t *db = dbuf_find(dn->dn_objset, dn->dn_object,
level, blkid);
if (db != NULL) {
mutex_exit(&db->db_mtx);
/*
* This dbuf already exists. It is either CACHED, or
* (we assume) about to be read or filled.
*/
goto no_issue;
}
/*
* Find the closest ancestor (indirect block) of the target block
* that is present in the cache. In this indirect block, we will
* find the bp that is at curlevel, curblkid.
*/
curlevel = level;
curblkid = blkid;
while (curlevel < nlevels - 1) {
int parent_level = curlevel + 1;
uint64_t parent_blkid = curblkid >> epbs;
dmu_buf_impl_t *db;
if (dbuf_hold_impl(dn, parent_level, parent_blkid,
FALSE, TRUE, FTAG, &db) == 0) {
blkptr_t *bpp = db->db_buf->b_data;
bp = bpp[P2PHASE(curblkid, 1 << epbs)];
dbuf_rele(db, FTAG);
break;
}
curlevel = parent_level;
curblkid = parent_blkid;
}
if (curlevel == nlevels - 1) {
/* No cached indirect blocks found. */
ASSERT3U(curblkid, <, dn->dn_phys->dn_nblkptr);
bp = dn->dn_phys->dn_blkptr[curblkid];
}
ASSERT(!BP_IS_REDACTED(&bp) ||
dsl_dataset_feature_is_active(dn->dn_objset->os_dsl_dataset,
SPA_FEATURE_REDACTED_DATASETS));
if (BP_IS_HOLE(&bp) || BP_IS_REDACTED(&bp))
goto no_issue;
ASSERT3U(curlevel, ==, BP_GET_LEVEL(&bp));
zio_t *pio = zio_root(dmu_objset_spa(dn->dn_objset), NULL, NULL,
ZIO_FLAG_CANFAIL);
dbuf_prefetch_arg_t *dpa = kmem_zalloc(sizeof (*dpa), KM_SLEEP);
dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset;
SET_BOOKMARK(&dpa->dpa_zb, ds != NULL ? ds->ds_object : DMU_META_OBJSET,
dn->dn_object, level, blkid);
dpa->dpa_curlevel = curlevel;
dpa->dpa_prio = prio;
dpa->dpa_aflags = aflags;
dpa->dpa_spa = dn->dn_objset->os_spa;
dpa->dpa_dnode = dn;
dpa->dpa_epbs = epbs;
dpa->dpa_zio = pio;
dpa->dpa_cb = cb;
dpa->dpa_arg = arg;
/* flag if L2ARC eligible, l2arc_noprefetch then decides */
if (DNODE_LEVEL_IS_L2CACHEABLE(dn, level))
dpa->dpa_aflags |= ARC_FLAG_L2CACHE;
/*
* If we have the indirect just above us, no need to do the asynchronous
* prefetch chain; we'll just run the last step ourselves. If we're at
* a higher level, though, we want to issue the prefetches for all the
* indirect blocks asynchronously, so we can go on with whatever we were
* doing.
*/
if (curlevel == level) {
ASSERT3U(curblkid, ==, blkid);
dbuf_issue_final_prefetch(dpa, &bp);
} else {
arc_flags_t iter_aflags = ARC_FLAG_NOWAIT;
zbookmark_phys_t zb;
/* flag if L2ARC eligible, l2arc_noprefetch then decides */
if (DNODE_LEVEL_IS_L2CACHEABLE(dn, level))
iter_aflags |= ARC_FLAG_L2CACHE;
SET_BOOKMARK(&zb, ds != NULL ? ds->ds_object : DMU_META_OBJSET,
dn->dn_object, curlevel, curblkid);
(void) arc_read(dpa->dpa_zio, dpa->dpa_spa,
&bp, dbuf_prefetch_indirect_done, dpa, prio,
ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
&iter_aflags, &zb);
}
/*
* We use pio here instead of dpa_zio since it's possible that
* dpa may have already been freed.
*/
zio_nowait(pio);
return (1);
no_issue:
if (cb != NULL)
cb(arg, B_FALSE);
return (0);
}
int
dbuf_prefetch(dnode_t *dn, int64_t level, uint64_t blkid, zio_priority_t prio,
arc_flags_t aflags)
{
return (dbuf_prefetch_impl(dn, level, blkid, prio, aflags, NULL, NULL));
}
/*
* Helper function for dbuf_hold_impl() to copy a buffer. Handles
* the case of encrypted, compressed and uncompressed buffers by
* allocating the new buffer, respectively, with arc_alloc_raw_buf(),
* arc_alloc_compressed_buf() or arc_alloc_buf().*
*
* NOTE: Declared noinline to avoid stack bloat in dbuf_hold_impl().
*/
noinline static void
dbuf_hold_copy(dnode_t *dn, dmu_buf_impl_t *db)
{
dbuf_dirty_record_t *dr = db->db_data_pending;
arc_buf_t *data = dr->dt.dl.dr_data;
enum zio_compress compress_type = arc_get_compression(data);
uint8_t complevel = arc_get_complevel(data);
if (arc_is_encrypted(data)) {
boolean_t byteorder;
uint8_t salt[ZIO_DATA_SALT_LEN];
uint8_t iv[ZIO_DATA_IV_LEN];
uint8_t mac[ZIO_DATA_MAC_LEN];
arc_get_raw_params(data, &byteorder, salt, iv, mac);
dbuf_set_data(db, arc_alloc_raw_buf(dn->dn_objset->os_spa, db,
dmu_objset_id(dn->dn_objset), byteorder, salt, iv, mac,
dn->dn_type, arc_buf_size(data), arc_buf_lsize(data),
compress_type, complevel));
} else if (compress_type != ZIO_COMPRESS_OFF) {
dbuf_set_data(db, arc_alloc_compressed_buf(
dn->dn_objset->os_spa, db, arc_buf_size(data),
arc_buf_lsize(data), compress_type, complevel));
} else {
dbuf_set_data(db, arc_alloc_buf(dn->dn_objset->os_spa, db,
DBUF_GET_BUFC_TYPE(db), db->db.db_size));
}
rw_enter(&db->db_rwlock, RW_WRITER);
bcopy(data->b_data, db->db.db_data, arc_buf_size(data));
rw_exit(&db->db_rwlock);
}
/*
* Returns with db_holds incremented, and db_mtx not held.
* Note: dn_struct_rwlock must be held.
*/
int
dbuf_hold_impl(dnode_t *dn, uint8_t level, uint64_t blkid,
boolean_t fail_sparse, boolean_t fail_uncached,
void *tag, dmu_buf_impl_t **dbp)
{
dmu_buf_impl_t *db, *parent = NULL;
/* If the pool has been created, verify the tx_sync_lock is not held */
spa_t *spa = dn->dn_objset->os_spa;
dsl_pool_t *dp = spa->spa_dsl_pool;
if (dp != NULL) {
ASSERT(!MUTEX_HELD(&dp->dp_tx.tx_sync_lock));
}
ASSERT(blkid != DMU_BONUS_BLKID);
ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock));
ASSERT3U(dn->dn_nlevels, >, level);
*dbp = NULL;
/* dbuf_find() returns with db_mtx held */
db = dbuf_find(dn->dn_objset, dn->dn_object, level, blkid);
if (db == NULL) {
blkptr_t *bp = NULL;
int err;
if (fail_uncached)
return (SET_ERROR(ENOENT));
ASSERT3P(parent, ==, NULL);
err = dbuf_findbp(dn, level, blkid, fail_sparse, &parent, &bp);
if (fail_sparse) {
if (err == 0 && bp && BP_IS_HOLE(bp))
err = SET_ERROR(ENOENT);
if (err) {
if (parent)
dbuf_rele(parent, NULL);
return (err);
}
}
if (err && err != ENOENT)
return (err);
db = dbuf_create(dn, level, blkid, parent, bp);
}
if (fail_uncached && db->db_state != DB_CACHED) {
mutex_exit(&db->db_mtx);
return (SET_ERROR(ENOENT));
}
if (db->db_buf != NULL) {
arc_buf_access(db->db_buf);
ASSERT3P(db->db.db_data, ==, db->db_buf->b_data);
}
ASSERT(db->db_buf == NULL || arc_referenced(db->db_buf));
/*
* If this buffer is currently syncing out, and we are
* still referencing it from db_data, we need to make a copy
* of it in case we decide we want to dirty it again in this txg.
*/
if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID &&
dn->dn_object != DMU_META_DNODE_OBJECT &&
db->db_state == DB_CACHED && db->db_data_pending) {
dbuf_dirty_record_t *dr = db->db_data_pending;
if (dr->dt.dl.dr_data == db->db_buf)
dbuf_hold_copy(dn, db);
}
if (multilist_link_active(&db->db_cache_link)) {
ASSERT(zfs_refcount_is_zero(&db->db_holds));
ASSERT(db->db_caching_status == DB_DBUF_CACHE ||
db->db_caching_status == DB_DBUF_METADATA_CACHE);
multilist_remove(&dbuf_caches[db->db_caching_status].cache, db);
(void) zfs_refcount_remove_many(
&dbuf_caches[db->db_caching_status].size,
db->db.db_size, db);
if (db->db_caching_status == DB_DBUF_METADATA_CACHE) {
DBUF_STAT_BUMPDOWN(metadata_cache_count);
} else {
DBUF_STAT_BUMPDOWN(cache_levels[db->db_level]);
DBUF_STAT_BUMPDOWN(cache_count);
DBUF_STAT_DECR(cache_levels_bytes[db->db_level],
db->db.db_size);
}
db->db_caching_status = DB_NO_CACHE;
}
(void) zfs_refcount_add(&db->db_holds, tag);
DBUF_VERIFY(db);
mutex_exit(&db->db_mtx);
/* NOTE: we can't rele the parent until after we drop the db_mtx */
if (parent)
dbuf_rele(parent, NULL);
ASSERT3P(DB_DNODE(db), ==, dn);
ASSERT3U(db->db_blkid, ==, blkid);
ASSERT3U(db->db_level, ==, level);
*dbp = db;
return (0);
}
dmu_buf_impl_t *
dbuf_hold(dnode_t *dn, uint64_t blkid, void *tag)
{
return (dbuf_hold_level(dn, 0, blkid, tag));
}
dmu_buf_impl_t *
dbuf_hold_level(dnode_t *dn, int level, uint64_t blkid, void *tag)
{
dmu_buf_impl_t *db;
int err = dbuf_hold_impl(dn, level, blkid, FALSE, FALSE, tag, &db);
return (err ? NULL : db);
}
void
dbuf_create_bonus(dnode_t *dn)
{
ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock));
ASSERT(dn->dn_bonus == NULL);
dn->dn_bonus = dbuf_create(dn, 0, DMU_BONUS_BLKID, dn->dn_dbuf, NULL);
}
int
dbuf_spill_set_blksz(dmu_buf_t *db_fake, uint64_t blksz, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
if (db->db_blkid != DMU_SPILL_BLKID)
return (SET_ERROR(ENOTSUP));
if (blksz == 0)
blksz = SPA_MINBLOCKSIZE;
ASSERT3U(blksz, <=, spa_maxblocksize(dmu_objset_spa(db->db_objset)));
blksz = P2ROUNDUP(blksz, SPA_MINBLOCKSIZE);
dbuf_new_size(db, blksz, tx);
return (0);
}
void
dbuf_rm_spill(dnode_t *dn, dmu_tx_t *tx)
{
dbuf_free_range(dn, DMU_SPILL_BLKID, DMU_SPILL_BLKID, tx);
}
#pragma weak dmu_buf_add_ref = dbuf_add_ref
void
dbuf_add_ref(dmu_buf_impl_t *db, void *tag)
{
int64_t holds = zfs_refcount_add(&db->db_holds, tag);
VERIFY3S(holds, >, 1);
}
#pragma weak dmu_buf_try_add_ref = dbuf_try_add_ref
boolean_t
dbuf_try_add_ref(dmu_buf_t *db_fake, objset_t *os, uint64_t obj, uint64_t blkid,
void *tag)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
dmu_buf_impl_t *found_db;
boolean_t result = B_FALSE;
if (blkid == DMU_BONUS_BLKID)
found_db = dbuf_find_bonus(os, obj);
else
found_db = dbuf_find(os, obj, 0, blkid);
if (found_db != NULL) {
if (db == found_db && dbuf_refcount(db) > db->db_dirtycnt) {
(void) zfs_refcount_add(&db->db_holds, tag);
result = B_TRUE;
}
mutex_exit(&found_db->db_mtx);
}
return (result);
}
/*
* If you call dbuf_rele() you had better not be referencing the dnode handle
* unless you have some other direct or indirect hold on the dnode. (An indirect
* hold is a hold on one of the dnode's dbufs, including the bonus buffer.)
* Without that, the dbuf_rele() could lead to a dnode_rele() followed by the
* dnode's parent dbuf evicting its dnode handles.
*/
void
dbuf_rele(dmu_buf_impl_t *db, void *tag)
{
mutex_enter(&db->db_mtx);
dbuf_rele_and_unlock(db, tag, B_FALSE);
}
void
dmu_buf_rele(dmu_buf_t *db, void *tag)
{
dbuf_rele((dmu_buf_impl_t *)db, tag);
}
/*
* dbuf_rele() for an already-locked dbuf. This is necessary to allow
* db_dirtycnt and db_holds to be updated atomically. The 'evicting'
* argument should be set if we are already in the dbuf-evicting code
* path, in which case we don't want to recursively evict. This allows us to
* avoid deeply nested stacks that would have a call flow similar to this:
*
* dbuf_rele()-->dbuf_rele_and_unlock()-->dbuf_evict_notify()
* ^ |
* | |
* +-----dbuf_destroy()<--dbuf_evict_one()<--------+
*
*/
void
dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag, boolean_t evicting)
{
int64_t holds;
uint64_t size;
ASSERT(MUTEX_HELD(&db->db_mtx));
DBUF_VERIFY(db);
/*
* Remove the reference to the dbuf before removing its hold on the
* dnode so we can guarantee in dnode_move() that a referenced bonus
* buffer has a corresponding dnode hold.
*/
holds = zfs_refcount_remove(&db->db_holds, tag);
ASSERT(holds >= 0);
/*
* We can't freeze indirects if there is a possibility that they
* may be modified in the current syncing context.
*/
if (db->db_buf != NULL &&
holds == (db->db_level == 0 ? db->db_dirtycnt : 0)) {
arc_buf_freeze(db->db_buf);
}
if (holds == db->db_dirtycnt &&
db->db_level == 0 && db->db_user_immediate_evict)
dbuf_evict_user(db);
if (holds == 0) {
if (db->db_blkid == DMU_BONUS_BLKID) {
dnode_t *dn;
boolean_t evict_dbuf = db->db_pending_evict;
/*
* If the dnode moves here, we cannot cross this
* barrier until the move completes.
*/
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
atomic_dec_32(&dn->dn_dbufs_count);
/*
* Decrementing the dbuf count means that the bonus
* buffer's dnode hold is no longer discounted in
* dnode_move(). The dnode cannot move until after
* the dnode_rele() below.
*/
DB_DNODE_EXIT(db);
/*
* Do not reference db after its lock is dropped.
* Another thread may evict it.
*/
mutex_exit(&db->db_mtx);
if (evict_dbuf)
dnode_evict_bonus(dn);
dnode_rele(dn, db);
} else if (db->db_buf == NULL) {
/*
* This is a special case: we never associated this
* dbuf with any data allocated from the ARC.
*/
ASSERT(db->db_state == DB_UNCACHED ||
db->db_state == DB_NOFILL);
dbuf_destroy(db);
} else if (arc_released(db->db_buf)) {
/*
* This dbuf has anonymous data associated with it.
*/
dbuf_destroy(db);
} else {
boolean_t do_arc_evict = B_FALSE;
blkptr_t bp;
spa_t *spa = dmu_objset_spa(db->db_objset);
if (!DBUF_IS_CACHEABLE(db) &&
db->db_blkptr != NULL &&
!BP_IS_HOLE(db->db_blkptr) &&
!BP_IS_EMBEDDED(db->db_blkptr)) {
do_arc_evict = B_TRUE;
bp = *db->db_blkptr;
}
if (!DBUF_IS_CACHEABLE(db) ||
db->db_pending_evict) {
dbuf_destroy(db);
} else if (!multilist_link_active(&db->db_cache_link)) {
ASSERT3U(db->db_caching_status, ==,
DB_NO_CACHE);
dbuf_cached_state_t dcs =
dbuf_include_in_metadata_cache(db) ?
DB_DBUF_METADATA_CACHE : DB_DBUF_CACHE;
db->db_caching_status = dcs;
multilist_insert(&dbuf_caches[dcs].cache, db);
uint64_t db_size = db->db.db_size;
size = zfs_refcount_add_many(
&dbuf_caches[dcs].size, db_size, db);
uint8_t db_level = db->db_level;
mutex_exit(&db->db_mtx);
if (dcs == DB_DBUF_METADATA_CACHE) {
DBUF_STAT_BUMP(metadata_cache_count);
DBUF_STAT_MAX(
metadata_cache_size_bytes_max,
size);
} else {
DBUF_STAT_BUMP(cache_count);
DBUF_STAT_MAX(cache_size_bytes_max,
size);
DBUF_STAT_BUMP(cache_levels[db_level]);
DBUF_STAT_INCR(
cache_levels_bytes[db_level],
db_size);
}
if (dcs == DB_DBUF_CACHE && !evicting)
dbuf_evict_notify(size);
}
if (do_arc_evict)
arc_freed(spa, &bp);
}
} else {
mutex_exit(&db->db_mtx);
}
}
#pragma weak dmu_buf_refcount = dbuf_refcount
uint64_t
dbuf_refcount(dmu_buf_impl_t *db)
{
return (zfs_refcount_count(&db->db_holds));
}
uint64_t
dmu_buf_user_refcount(dmu_buf_t *db_fake)
{
uint64_t holds;
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
mutex_enter(&db->db_mtx);
ASSERT3U(zfs_refcount_count(&db->db_holds), >=, db->db_dirtycnt);
holds = zfs_refcount_count(&db->db_holds) - db->db_dirtycnt;
mutex_exit(&db->db_mtx);
return (holds);
}
void *
dmu_buf_replace_user(dmu_buf_t *db_fake, dmu_buf_user_t *old_user,
dmu_buf_user_t *new_user)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
mutex_enter(&db->db_mtx);
dbuf_verify_user(db, DBVU_NOT_EVICTING);
if (db->db_user == old_user)
db->db_user = new_user;
else
old_user = db->db_user;
dbuf_verify_user(db, DBVU_NOT_EVICTING);
mutex_exit(&db->db_mtx);
return (old_user);
}
void *
dmu_buf_set_user(dmu_buf_t *db_fake, dmu_buf_user_t *user)
{
return (dmu_buf_replace_user(db_fake, NULL, user));
}
void *
dmu_buf_set_user_ie(dmu_buf_t *db_fake, dmu_buf_user_t *user)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
db->db_user_immediate_evict = TRUE;
return (dmu_buf_set_user(db_fake, user));
}
void *
dmu_buf_remove_user(dmu_buf_t *db_fake, dmu_buf_user_t *user)
{
return (dmu_buf_replace_user(db_fake, user, NULL));
}
void *
dmu_buf_get_user(dmu_buf_t *db_fake)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
dbuf_verify_user(db, DBVU_NOT_EVICTING);
return (db->db_user);
}
void
dmu_buf_user_evict_wait()
{
taskq_wait(dbu_evict_taskq);
}
blkptr_t *
dmu_buf_get_blkptr(dmu_buf_t *db)
{
dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
return (dbi->db_blkptr);
}
objset_t *
dmu_buf_get_objset(dmu_buf_t *db)
{
dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
return (dbi->db_objset);
}
dnode_t *
dmu_buf_dnode_enter(dmu_buf_t *db)
{
dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
DB_DNODE_ENTER(dbi);
return (DB_DNODE(dbi));
}
void
dmu_buf_dnode_exit(dmu_buf_t *db)
{
dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
DB_DNODE_EXIT(dbi);
}
static void
dbuf_check_blkptr(dnode_t *dn, dmu_buf_impl_t *db)
{
/* ASSERT(dmu_tx_is_syncing(tx) */
ASSERT(MUTEX_HELD(&db->db_mtx));
if (db->db_blkptr != NULL)
return;
if (db->db_blkid == DMU_SPILL_BLKID) {
db->db_blkptr = DN_SPILL_BLKPTR(dn->dn_phys);
BP_ZERO(db->db_blkptr);
return;
}
if (db->db_level == dn->dn_phys->dn_nlevels-1) {
/*
* This buffer was allocated at a time when there was
* no available blkptrs from the dnode, or it was
* inappropriate to hook it in (i.e., nlevels mismatch).
*/
ASSERT(db->db_blkid < dn->dn_phys->dn_nblkptr);
ASSERT(db->db_parent == NULL);
db->db_parent = dn->dn_dbuf;
db->db_blkptr = &dn->dn_phys->dn_blkptr[db->db_blkid];
DBUF_VERIFY(db);
} else {
dmu_buf_impl_t *parent = db->db_parent;
int epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT;
ASSERT(dn->dn_phys->dn_nlevels > 1);
if (parent == NULL) {
mutex_exit(&db->db_mtx);
rw_enter(&dn->dn_struct_rwlock, RW_READER);
parent = dbuf_hold_level(dn, db->db_level + 1,
db->db_blkid >> epbs, db);
rw_exit(&dn->dn_struct_rwlock);
mutex_enter(&db->db_mtx);
db->db_parent = parent;
}
db->db_blkptr = (blkptr_t *)parent->db.db_data +
(db->db_blkid & ((1ULL << epbs) - 1));
DBUF_VERIFY(db);
}
}
static void
dbuf_sync_bonus(dbuf_dirty_record_t *dr, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = dr->dr_dbuf;
void *data = dr->dt.dl.dr_data;
ASSERT0(db->db_level);
ASSERT(MUTEX_HELD(&db->db_mtx));
ASSERT(db->db_blkid == DMU_BONUS_BLKID);
ASSERT(data != NULL);
dnode_t *dn = dr->dr_dnode;
ASSERT3U(DN_MAX_BONUS_LEN(dn->dn_phys), <=,
DN_SLOTS_TO_BONUSLEN(dn->dn_phys->dn_extra_slots + 1));
bcopy(data, DN_BONUS(dn->dn_phys), DN_MAX_BONUS_LEN(dn->dn_phys));
dbuf_sync_leaf_verify_bonus_dnode(dr);
dbuf_undirty_bonus(dr);
dbuf_rele_and_unlock(db, (void *)(uintptr_t)tx->tx_txg, B_FALSE);
}
/*
* When syncing out a blocks of dnodes, adjust the block to deal with
* encryption. Normally, we make sure the block is decrypted before writing
* it. If we have crypt params, then we are writing a raw (encrypted) block,
* from a raw receive. In this case, set the ARC buf's crypt params so
* that the BP will be filled with the correct byteorder, salt, iv, and mac.
*/
static void
dbuf_prepare_encrypted_dnode_leaf(dbuf_dirty_record_t *dr)
{
int err;
dmu_buf_impl_t *db = dr->dr_dbuf;
ASSERT(MUTEX_HELD(&db->db_mtx));
ASSERT3U(db->db.db_object, ==, DMU_META_DNODE_OBJECT);
ASSERT3U(db->db_level, ==, 0);
if (!db->db_objset->os_raw_receive && arc_is_encrypted(db->db_buf)) {
zbookmark_phys_t zb;
/*
* Unfortunately, there is currently no mechanism for
* syncing context to handle decryption errors. An error
* here is only possible if an attacker maliciously
* changed a dnode block and updated the associated
* checksums going up the block tree.
*/
SET_BOOKMARK(&zb, dmu_objset_id(db->db_objset),
db->db.db_object, db->db_level, db->db_blkid);
err = arc_untransform(db->db_buf, db->db_objset->os_spa,
&zb, B_TRUE);
if (err)
panic("Invalid dnode block MAC");
} else if (dr->dt.dl.dr_has_raw_params) {
(void) arc_release(dr->dt.dl.dr_data, db);
arc_convert_to_raw(dr->dt.dl.dr_data,
dmu_objset_id(db->db_objset),
dr->dt.dl.dr_byteorder, DMU_OT_DNODE,
dr->dt.dl.dr_salt, dr->dt.dl.dr_iv, dr->dt.dl.dr_mac);
}
}
/*
* dbuf_sync_indirect() is called recursively from dbuf_sync_list() so it
* is critical the we not allow the compiler to inline this function in to
* dbuf_sync_list() thereby drastically bloating the stack usage.
*/
noinline static void
dbuf_sync_indirect(dbuf_dirty_record_t *dr, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = dr->dr_dbuf;
dnode_t *dn = dr->dr_dnode;
ASSERT(dmu_tx_is_syncing(tx));
dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr);
mutex_enter(&db->db_mtx);
ASSERT(db->db_level > 0);
DBUF_VERIFY(db);
/* Read the block if it hasn't been read yet. */
if (db->db_buf == NULL) {
mutex_exit(&db->db_mtx);
(void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED);
mutex_enter(&db->db_mtx);
}
ASSERT3U(db->db_state, ==, DB_CACHED);
ASSERT(db->db_buf != NULL);
/* Indirect block size must match what the dnode thinks it is. */
ASSERT3U(db->db.db_size, ==, 1<dn_phys->dn_indblkshift);
dbuf_check_blkptr(dn, db);
/* Provide the pending dirty record to child dbufs */
db->db_data_pending = dr;
mutex_exit(&db->db_mtx);
dbuf_write(dr, db->db_buf, tx);
zio_t *zio = dr->dr_zio;
mutex_enter(&dr->dt.di.dr_mtx);
dbuf_sync_list(&dr->dt.di.dr_children, db->db_level - 1, tx);
ASSERT(list_head(&dr->dt.di.dr_children) == NULL);
mutex_exit(&dr->dt.di.dr_mtx);
zio_nowait(zio);
}
/*
* Verify that the size of the data in our bonus buffer does not exceed
* its recorded size.
*
* The purpose of this verification is to catch any cases in development
* where the size of a phys structure (i.e space_map_phys_t) grows and,
* due to incorrect feature management, older pools expect to read more
* data even though they didn't actually write it to begin with.
*
* For a example, this would catch an error in the feature logic where we
* open an older pool and we expect to write the space map histogram of
* a space map with size SPACE_MAP_SIZE_V0.
*/
static void
dbuf_sync_leaf_verify_bonus_dnode(dbuf_dirty_record_t *dr)
{
#ifdef ZFS_DEBUG
dnode_t *dn = dr->dr_dnode;
/*
* Encrypted bonus buffers can have data past their bonuslen.
* Skip the verification of these blocks.
*/
if (DMU_OT_IS_ENCRYPTED(dn->dn_bonustype))
return;
uint16_t bonuslen = dn->dn_phys->dn_bonuslen;
uint16_t maxbonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots);
ASSERT3U(bonuslen, <=, maxbonuslen);
arc_buf_t *datap = dr->dt.dl.dr_data;
char *datap_end = ((char *)datap) + bonuslen;
char *datap_max = ((char *)datap) + maxbonuslen;
/* ensure that everything is zero after our data */
for (; datap_end < datap_max; datap_end++)
ASSERT(*datap_end == 0);
#endif
}
static blkptr_t *
dbuf_lightweight_bp(dbuf_dirty_record_t *dr)
{
/* This must be a lightweight dirty record. */
ASSERT3P(dr->dr_dbuf, ==, NULL);
dnode_t *dn = dr->dr_dnode;
if (dn->dn_phys->dn_nlevels == 1) {
VERIFY3U(dr->dt.dll.dr_blkid, <, dn->dn_phys->dn_nblkptr);
return (&dn->dn_phys->dn_blkptr[dr->dt.dll.dr_blkid]);
} else {
dmu_buf_impl_t *parent_db = dr->dr_parent->dr_dbuf;
int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
VERIFY3U(parent_db->db_level, ==, 1);
VERIFY3P(parent_db->db_dnode_handle->dnh_dnode, ==, dn);
VERIFY3U(dr->dt.dll.dr_blkid >> epbs, ==, parent_db->db_blkid);
blkptr_t *bp = parent_db->db.db_data;
return (&bp[dr->dt.dll.dr_blkid & ((1 << epbs) - 1)]);
}
}
static void
dbuf_lightweight_ready(zio_t *zio)
{
dbuf_dirty_record_t *dr = zio->io_private;
blkptr_t *bp = zio->io_bp;
if (zio->io_error != 0)
return;
dnode_t *dn = dr->dr_dnode;
blkptr_t *bp_orig = dbuf_lightweight_bp(dr);
spa_t *spa = dmu_objset_spa(dn->dn_objset);
int64_t delta = bp_get_dsize_sync(spa, bp) -
bp_get_dsize_sync(spa, bp_orig);
dnode_diduse_space(dn, delta);
uint64_t blkid = dr->dt.dll.dr_blkid;
mutex_enter(&dn->dn_mtx);
if (blkid > dn->dn_phys->dn_maxblkid) {
ASSERT0(dn->dn_objset->os_raw_receive);
dn->dn_phys->dn_maxblkid = blkid;
}
mutex_exit(&dn->dn_mtx);
if (!BP_IS_EMBEDDED(bp)) {
uint64_t fill = BP_IS_HOLE(bp) ? 0 : 1;
BP_SET_FILL(bp, fill);
}
dmu_buf_impl_t *parent_db;
EQUIV(dr->dr_parent == NULL, dn->dn_phys->dn_nlevels == 1);
if (dr->dr_parent == NULL) {
parent_db = dn->dn_dbuf;
} else {
parent_db = dr->dr_parent->dr_dbuf;
}
rw_enter(&parent_db->db_rwlock, RW_WRITER);
*bp_orig = *bp;
rw_exit(&parent_db->db_rwlock);
}
static void
dbuf_lightweight_physdone(zio_t *zio)
{
dbuf_dirty_record_t *dr = zio->io_private;
dsl_pool_t *dp = spa_get_dsl(zio->io_spa);
ASSERT3U(dr->dr_txg, ==, zio->io_txg);
/*
* The callback will be called io_phys_children times. Retire one
* portion of our dirty space each time we are called. Any rounding
* error will be cleaned up by dbuf_lightweight_done().
*/
int delta = dr->dr_accounted / zio->io_phys_children;
dsl_pool_undirty_space(dp, delta, zio->io_txg);
}
static void
dbuf_lightweight_done(zio_t *zio)
{
dbuf_dirty_record_t *dr = zio->io_private;
VERIFY0(zio->io_error);
objset_t *os = dr->dr_dnode->dn_objset;
dmu_tx_t *tx = os->os_synctx;
if (zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE)) {
ASSERT(BP_EQUAL(zio->io_bp, &zio->io_bp_orig));
} else {
dsl_dataset_t *ds = os->os_dsl_dataset;
(void) dsl_dataset_block_kill(ds, &zio->io_bp_orig, tx, B_TRUE);
dsl_dataset_block_born(ds, zio->io_bp, tx);
}
/*
* See comment in dbuf_write_done().
*/
if (zio->io_phys_children == 0) {
dsl_pool_undirty_space(dmu_objset_pool(os),
dr->dr_accounted, zio->io_txg);
} else {
dsl_pool_undirty_space(dmu_objset_pool(os),
dr->dr_accounted % zio->io_phys_children, zio->io_txg);
}
abd_free(dr->dt.dll.dr_abd);
kmem_free(dr, sizeof (*dr));
}
noinline static void
dbuf_sync_lightweight(dbuf_dirty_record_t *dr, dmu_tx_t *tx)
{
dnode_t *dn = dr->dr_dnode;
zio_t *pio;
if (dn->dn_phys->dn_nlevels == 1) {
pio = dn->dn_zio;
} else {
pio = dr->dr_parent->dr_zio;
}
zbookmark_phys_t zb = {
.zb_objset = dmu_objset_id(dn->dn_objset),
.zb_object = dn->dn_object,
.zb_level = 0,
.zb_blkid = dr->dt.dll.dr_blkid,
};
/*
* See comment in dbuf_write(). This is so that zio->io_bp_orig
* will have the old BP in dbuf_lightweight_done().
*/
dr->dr_bp_copy = *dbuf_lightweight_bp(dr);
dr->dr_zio = zio_write(pio, dmu_objset_spa(dn->dn_objset),
dmu_tx_get_txg(tx), &dr->dr_bp_copy, dr->dt.dll.dr_abd,
dn->dn_datablksz, abd_get_size(dr->dt.dll.dr_abd),
&dr->dt.dll.dr_props, dbuf_lightweight_ready, NULL,
dbuf_lightweight_physdone, dbuf_lightweight_done, dr,
ZIO_PRIORITY_ASYNC_WRITE,
ZIO_FLAG_MUSTSUCCEED | dr->dt.dll.dr_flags, &zb);
zio_nowait(dr->dr_zio);
}
/*
* dbuf_sync_leaf() is called recursively from dbuf_sync_list() so it is
* critical the we not allow the compiler to inline this function in to
* dbuf_sync_list() thereby drastically bloating the stack usage.
*/
noinline static void
dbuf_sync_leaf(dbuf_dirty_record_t *dr, dmu_tx_t *tx)
{
arc_buf_t **datap = &dr->dt.dl.dr_data;
dmu_buf_impl_t *db = dr->dr_dbuf;
dnode_t *dn = dr->dr_dnode;
objset_t *os;
uint64_t txg = tx->tx_txg;
ASSERT(dmu_tx_is_syncing(tx));
dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr);
mutex_enter(&db->db_mtx);
/*
* To be synced, we must be dirtied. But we
* might have been freed after the dirty.
*/
if (db->db_state == DB_UNCACHED) {
/* This buffer has been freed since it was dirtied */
ASSERT(db->db.db_data == NULL);
} else if (db->db_state == DB_FILL) {
/* This buffer was freed and is now being re-filled */
ASSERT(db->db.db_data != dr->dt.dl.dr_data);
} else {
ASSERT(db->db_state == DB_CACHED || db->db_state == DB_NOFILL);
}
DBUF_VERIFY(db);
if (db->db_blkid == DMU_SPILL_BLKID) {
mutex_enter(&dn->dn_mtx);
if (!(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) {
/*
* In the previous transaction group, the bonus buffer
* was entirely used to store the attributes for the
* dnode which overrode the dn_spill field. However,
* when adding more attributes to the file a spill
* block was required to hold the extra attributes.
*
* Make sure to clear the garbage left in the dn_spill
* field from the previous attributes in the bonus
* buffer. Otherwise, after writing out the spill
* block to the new allocated dva, it will free
* the old block pointed to by the invalid dn_spill.
*/
db->db_blkptr = NULL;
}
dn->dn_phys->dn_flags |= DNODE_FLAG_SPILL_BLKPTR;
mutex_exit(&dn->dn_mtx);
}
/*
* If this is a bonus buffer, simply copy the bonus data into the
* dnode. It will be written out when the dnode is synced (and it
* will be synced, since it must have been dirty for dbuf_sync to
* be called).
*/
if (db->db_blkid == DMU_BONUS_BLKID) {
ASSERT(dr->dr_dbuf == db);
dbuf_sync_bonus(dr, tx);
return;
}
os = dn->dn_objset;
/*
* This function may have dropped the db_mtx lock allowing a dmu_sync
* operation to sneak in. As a result, we need to ensure that we
* don't check the dr_override_state until we have returned from
* dbuf_check_blkptr.
*/
dbuf_check_blkptr(dn, db);
/*
* If this buffer is in the middle of an immediate write,
* wait for the synchronous IO to complete.
*/
while (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC) {
ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
cv_wait(&db->db_changed, &db->db_mtx);
ASSERT(dr->dt.dl.dr_override_state != DR_NOT_OVERRIDDEN);
}
/*
* If this is a dnode block, ensure it is appropriately encrypted
* or decrypted, depending on what we are writing to it this txg.
*/
if (os->os_encrypted && dn->dn_object == DMU_META_DNODE_OBJECT)
dbuf_prepare_encrypted_dnode_leaf(dr);
if (db->db_state != DB_NOFILL &&
dn->dn_object != DMU_META_DNODE_OBJECT &&
zfs_refcount_count(&db->db_holds) > 1 &&
dr->dt.dl.dr_override_state != DR_OVERRIDDEN &&
*datap == db->db_buf) {
/*
* If this buffer is currently "in use" (i.e., there
* are active holds and db_data still references it),
* then make a copy before we start the write so that
* any modifications from the open txg will not leak
* into this write.
*
* NOTE: this copy does not need to be made for
* objects only modified in the syncing context (e.g.
* DNONE_DNODE blocks).
*/
int psize = arc_buf_size(*datap);
int lsize = arc_buf_lsize(*datap);
arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db);
enum zio_compress compress_type = arc_get_compression(*datap);
uint8_t complevel = arc_get_complevel(*datap);
if (arc_is_encrypted(*datap)) {
boolean_t byteorder;
uint8_t salt[ZIO_DATA_SALT_LEN];
uint8_t iv[ZIO_DATA_IV_LEN];
uint8_t mac[ZIO_DATA_MAC_LEN];
arc_get_raw_params(*datap, &byteorder, salt, iv, mac);
*datap = arc_alloc_raw_buf(os->os_spa, db,
dmu_objset_id(os), byteorder, salt, iv, mac,
dn->dn_type, psize, lsize, compress_type,
complevel);
} else if (compress_type != ZIO_COMPRESS_OFF) {
ASSERT3U(type, ==, ARC_BUFC_DATA);
*datap = arc_alloc_compressed_buf(os->os_spa, db,
psize, lsize, compress_type, complevel);
} else {
*datap = arc_alloc_buf(os->os_spa, db, type, psize);
}
bcopy(db->db.db_data, (*datap)->b_data, psize);
}
db->db_data_pending = dr;
mutex_exit(&db->db_mtx);
dbuf_write(dr, *datap, tx);
ASSERT(!list_link_active(&dr->dr_dirty_node));
if (dn->dn_object == DMU_META_DNODE_OBJECT) {
list_insert_tail(&dn->dn_dirty_records[txg & TXG_MASK], dr);
} else {
zio_nowait(dr->dr_zio);
}
}
void
dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx)
{
dbuf_dirty_record_t *dr;
while ((dr = list_head(list))) {
if (dr->dr_zio != NULL) {
/*
* If we find an already initialized zio then we
* are processing the meta-dnode, and we have finished.
* The dbufs for all dnodes are put back on the list
* during processing, so that we can zio_wait()
* these IOs after initiating all child IOs.
*/
ASSERT3U(dr->dr_dbuf->db.db_object, ==,
DMU_META_DNODE_OBJECT);
break;
}
list_remove(list, dr);
if (dr->dr_dbuf == NULL) {
dbuf_sync_lightweight(dr, tx);
} else {
if (dr->dr_dbuf->db_blkid != DMU_BONUS_BLKID &&
dr->dr_dbuf->db_blkid != DMU_SPILL_BLKID) {
VERIFY3U(dr->dr_dbuf->db_level, ==, level);
}
if (dr->dr_dbuf->db_level > 0)
dbuf_sync_indirect(dr, tx);
else
dbuf_sync_leaf(dr, tx);
}
}
}
/* ARGSUSED */
static void
dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb)
{
dmu_buf_impl_t *db = vdb;
dnode_t *dn;
blkptr_t *bp = zio->io_bp;
blkptr_t *bp_orig = &zio->io_bp_orig;
spa_t *spa = zio->io_spa;
int64_t delta;
uint64_t fill = 0;
int i;
ASSERT3P(db->db_blkptr, !=, NULL);
ASSERT3P(&db->db_data_pending->dr_bp_copy, ==, bp);
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
delta = bp_get_dsize_sync(spa, bp) - bp_get_dsize_sync(spa, bp_orig);
dnode_diduse_space(dn, delta - zio->io_prev_space_delta);
zio->io_prev_space_delta = delta;
if (bp->blk_birth != 0) {
ASSERT((db->db_blkid != DMU_SPILL_BLKID &&
BP_GET_TYPE(bp) == dn->dn_type) ||
(db->db_blkid == DMU_SPILL_BLKID &&
BP_GET_TYPE(bp) == dn->dn_bonustype) ||
BP_IS_EMBEDDED(bp));
ASSERT(BP_GET_LEVEL(bp) == db->db_level);
}
mutex_enter(&db->db_mtx);
#ifdef ZFS_DEBUG
if (db->db_blkid == DMU_SPILL_BLKID) {
ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR);
ASSERT(!(BP_IS_HOLE(bp)) &&
db->db_blkptr == DN_SPILL_BLKPTR(dn->dn_phys));
}
#endif
if (db->db_level == 0) {
mutex_enter(&dn->dn_mtx);
if (db->db_blkid > dn->dn_phys->dn_maxblkid &&
db->db_blkid != DMU_SPILL_BLKID) {
ASSERT0(db->db_objset->os_raw_receive);
dn->dn_phys->dn_maxblkid = db->db_blkid;
}
mutex_exit(&dn->dn_mtx);
if (dn->dn_type == DMU_OT_DNODE) {
i = 0;
while (i < db->db.db_size) {
dnode_phys_t *dnp =
(void *)(((char *)db->db.db_data) + i);
i += DNODE_MIN_SIZE;
if (dnp->dn_type != DMU_OT_NONE) {
fill++;
i += dnp->dn_extra_slots *
DNODE_MIN_SIZE;
}
}
} else {
if (BP_IS_HOLE(bp)) {
fill = 0;
} else {
fill = 1;
}
}
} else {
blkptr_t *ibp = db->db.db_data;
ASSERT3U(db->db.db_size, ==, 1<dn_phys->dn_indblkshift);
for (i = db->db.db_size >> SPA_BLKPTRSHIFT; i > 0; i--, ibp++) {
if (BP_IS_HOLE(ibp))
continue;
fill += BP_GET_FILL(ibp);
}
}
DB_DNODE_EXIT(db);
if (!BP_IS_EMBEDDED(bp))
BP_SET_FILL(bp, fill);
mutex_exit(&db->db_mtx);
db_lock_type_t dblt = dmu_buf_lock_parent(db, RW_WRITER, FTAG);
*db->db_blkptr = *bp;
dmu_buf_unlock_parent(db, dblt, FTAG);
}
/* ARGSUSED */
/*
* This function gets called just prior to running through the compression
* stage of the zio pipeline. If we're an indirect block comprised of only
* holes, then we want this indirect to be compressed away to a hole. In
* order to do that we must zero out any information about the holes that
* this indirect points to prior to before we try to compress it.
*/
static void
dbuf_write_children_ready(zio_t *zio, arc_buf_t *buf, void *vdb)
{
dmu_buf_impl_t *db = vdb;
dnode_t *dn;
blkptr_t *bp;
unsigned int epbs, i;
ASSERT3U(db->db_level, >, 0);
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT;
ASSERT3U(epbs, <, 31);
/* Determine if all our children are holes */
for (i = 0, bp = db->db.db_data; i < 1ULL << epbs; i++, bp++) {
if (!BP_IS_HOLE(bp))
break;
}
/*
* If all the children are holes, then zero them all out so that
* we may get compressed away.
*/
if (i == 1ULL << epbs) {
/*
* We only found holes. Grab the rwlock to prevent
* anybody from reading the blocks we're about to
* zero out.
*/
rw_enter(&db->db_rwlock, RW_WRITER);
bzero(db->db.db_data, db->db.db_size);
rw_exit(&db->db_rwlock);
}
DB_DNODE_EXIT(db);
}
/*
* The SPA will call this callback several times for each zio - once
* for every physical child i/o (zio->io_phys_children times). This
* allows the DMU to monitor the progress of each logical i/o. For example,
* there may be 2 copies of an indirect block, or many fragments of a RAID-Z
* block. There may be a long delay before all copies/fragments are completed,
* so this callback allows us to retire dirty space gradually, as the physical
* i/os complete.
*/
/* ARGSUSED */
static void
dbuf_write_physdone(zio_t *zio, arc_buf_t *buf, void *arg)
{
dmu_buf_impl_t *db = arg;
objset_t *os = db->db_objset;
dsl_pool_t *dp = dmu_objset_pool(os);
dbuf_dirty_record_t *dr;
int delta = 0;
dr = db->db_data_pending;
ASSERT3U(dr->dr_txg, ==, zio->io_txg);
/*
* The callback will be called io_phys_children times. Retire one
* portion of our dirty space each time we are called. Any rounding
* error will be cleaned up by dbuf_write_done().
*/
delta = dr->dr_accounted / zio->io_phys_children;
dsl_pool_undirty_space(dp, delta, zio->io_txg);
}
/* ARGSUSED */
static void
dbuf_write_done(zio_t *zio, arc_buf_t *buf, void *vdb)
{
dmu_buf_impl_t *db = vdb;
blkptr_t *bp_orig = &zio->io_bp_orig;
blkptr_t *bp = db->db_blkptr;
objset_t *os = db->db_objset;
dmu_tx_t *tx = os->os_synctx;
ASSERT0(zio->io_error);
ASSERT(db->db_blkptr == bp);
/*
* For nopwrites and rewrites we ensure that the bp matches our
* original and bypass all the accounting.
*/
if (zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE)) {
ASSERT(BP_EQUAL(bp, bp_orig));
} else {
dsl_dataset_t *ds = os->os_dsl_dataset;
(void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
dsl_dataset_block_born(ds, bp, tx);
}
mutex_enter(&db->db_mtx);
DBUF_VERIFY(db);
dbuf_dirty_record_t *dr = db->db_data_pending;
dnode_t *dn = dr->dr_dnode;
ASSERT(!list_link_active(&dr->dr_dirty_node));
ASSERT(dr->dr_dbuf == db);
ASSERT(list_next(&db->db_dirty_records, dr) == NULL);
list_remove(&db->db_dirty_records, dr);
#ifdef ZFS_DEBUG
if (db->db_blkid == DMU_SPILL_BLKID) {
ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR);
ASSERT(!(BP_IS_HOLE(db->db_blkptr)) &&
db->db_blkptr == DN_SPILL_BLKPTR(dn->dn_phys));
}
#endif
if (db->db_level == 0) {
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
if (db->db_state != DB_NOFILL) {
if (dr->dt.dl.dr_data != db->db_buf)
arc_buf_destroy(dr->dt.dl.dr_data, db);
}
} else {
ASSERT(list_head(&dr->dt.di.dr_children) == NULL);
ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift);
if (!BP_IS_HOLE(db->db_blkptr)) {
int epbs __maybe_unused = dn->dn_phys->dn_indblkshift -
SPA_BLKPTRSHIFT;
ASSERT3U(db->db_blkid, <=,
dn->dn_phys->dn_maxblkid >> (db->db_level * epbs));
ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==,
db->db.db_size);
}
mutex_destroy(&dr->dt.di.dr_mtx);
list_destroy(&dr->dt.di.dr_children);
}
cv_broadcast(&db->db_changed);
ASSERT(db->db_dirtycnt > 0);
db->db_dirtycnt -= 1;
db->db_data_pending = NULL;
dbuf_rele_and_unlock(db, (void *)(uintptr_t)tx->tx_txg, B_FALSE);
/*
* If we didn't do a physical write in this ZIO and we
* still ended up here, it means that the space of the
* dbuf that we just released (and undirtied) above hasn't
* been marked as undirtied in the pool's accounting.
*
* Thus, we undirty that space in the pool's view of the
* world here. For physical writes this type of update
* happens in dbuf_write_physdone().
*
* If we did a physical write, cleanup any rounding errors
* that came up due to writing multiple copies of a block
* on disk [see dbuf_write_physdone()].
*/
if (zio->io_phys_children == 0) {
dsl_pool_undirty_space(dmu_objset_pool(os),
dr->dr_accounted, zio->io_txg);
} else {
dsl_pool_undirty_space(dmu_objset_pool(os),
dr->dr_accounted % zio->io_phys_children, zio->io_txg);
}
kmem_free(dr, sizeof (dbuf_dirty_record_t));
}
static void
dbuf_write_nofill_ready(zio_t *zio)
{
dbuf_write_ready(zio, NULL, zio->io_private);
}
static void
dbuf_write_nofill_done(zio_t *zio)
{
dbuf_write_done(zio, NULL, zio->io_private);
}
static void
dbuf_write_override_ready(zio_t *zio)
{
dbuf_dirty_record_t *dr = zio->io_private;
dmu_buf_impl_t *db = dr->dr_dbuf;
dbuf_write_ready(zio, NULL, db);
}
static void
dbuf_write_override_done(zio_t *zio)
{
dbuf_dirty_record_t *dr = zio->io_private;
dmu_buf_impl_t *db = dr->dr_dbuf;
blkptr_t *obp = &dr->dt.dl.dr_overridden_by;
mutex_enter(&db->db_mtx);
if (!BP_EQUAL(zio->io_bp, obp)) {
if (!BP_IS_HOLE(obp))
dsl_free(spa_get_dsl(zio->io_spa), zio->io_txg, obp);
arc_release(dr->dt.dl.dr_data, db);
}
mutex_exit(&db->db_mtx);
dbuf_write_done(zio, NULL, db);
if (zio->io_abd != NULL)
abd_free(zio->io_abd);
}
typedef struct dbuf_remap_impl_callback_arg {
objset_t *drica_os;
uint64_t drica_blk_birth;
dmu_tx_t *drica_tx;
} dbuf_remap_impl_callback_arg_t;
static void
dbuf_remap_impl_callback(uint64_t vdev, uint64_t offset, uint64_t size,
void *arg)
{
dbuf_remap_impl_callback_arg_t *drica = arg;
objset_t *os = drica->drica_os;
spa_t *spa = dmu_objset_spa(os);
dmu_tx_t *tx = drica->drica_tx;
ASSERT(dsl_pool_sync_context(spa_get_dsl(spa)));
if (os == spa_meta_objset(spa)) {
spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx);
} else {
dsl_dataset_block_remapped(dmu_objset_ds(os), vdev, offset,
size, drica->drica_blk_birth, tx);
}
}
static void
dbuf_remap_impl(dnode_t *dn, blkptr_t *bp, krwlock_t *rw, dmu_tx_t *tx)
{
blkptr_t bp_copy = *bp;
spa_t *spa = dmu_objset_spa(dn->dn_objset);
dbuf_remap_impl_callback_arg_t drica;
ASSERT(dsl_pool_sync_context(spa_get_dsl(spa)));
drica.drica_os = dn->dn_objset;
drica.drica_blk_birth = bp->blk_birth;
drica.drica_tx = tx;
if (spa_remap_blkptr(spa, &bp_copy, dbuf_remap_impl_callback,
&drica)) {
/*
* If the blkptr being remapped is tracked by a livelist,
* then we need to make sure the livelist reflects the update.
* First, cancel out the old blkptr by appending a 'FREE'
* entry. Next, add an 'ALLOC' to track the new version. This
* way we avoid trying to free an inaccurate blkptr at delete.
* Note that embedded blkptrs are not tracked in livelists.
*/
if (dn->dn_objset != spa_meta_objset(spa)) {
dsl_dataset_t *ds = dmu_objset_ds(dn->dn_objset);
if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
bp->blk_birth > ds->ds_dir->dd_origin_txg) {
ASSERT(!BP_IS_EMBEDDED(bp));
ASSERT(dsl_dir_is_clone(ds->ds_dir));
ASSERT(spa_feature_is_enabled(spa,
SPA_FEATURE_LIVELIST));
bplist_append(&ds->ds_dir->dd_pending_frees,
bp);
bplist_append(&ds->ds_dir->dd_pending_allocs,
&bp_copy);
}
}
/*
* The db_rwlock prevents dbuf_read_impl() from
* dereferencing the BP while we are changing it. To
* avoid lock contention, only grab it when we are actually
* changing the BP.
*/
if (rw != NULL)
rw_enter(rw, RW_WRITER);
*bp = bp_copy;
if (rw != NULL)
rw_exit(rw);
}
}
/*
* Remap any existing BP's to concrete vdevs, if possible.
*/
static void
dbuf_remap(dnode_t *dn, dmu_buf_impl_t *db, dmu_tx_t *tx)
{
spa_t *spa = dmu_objset_spa(db->db_objset);
ASSERT(dsl_pool_sync_context(spa_get_dsl(spa)));
if (!spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL))
return;
if (db->db_level > 0) {
blkptr_t *bp = db->db.db_data;
for (int i = 0; i < db->db.db_size >> SPA_BLKPTRSHIFT; i++) {
dbuf_remap_impl(dn, &bp[i], &db->db_rwlock, tx);
}
} else if (db->db.db_object == DMU_META_DNODE_OBJECT) {
dnode_phys_t *dnp = db->db.db_data;
ASSERT3U(db->db_dnode_handle->dnh_dnode->dn_type, ==,
DMU_OT_DNODE);
for (int i = 0; i < db->db.db_size >> DNODE_SHIFT;
i += dnp[i].dn_extra_slots + 1) {
for (int j = 0; j < dnp[i].dn_nblkptr; j++) {
krwlock_t *lock = (dn->dn_dbuf == NULL ? NULL :
&dn->dn_dbuf->db_rwlock);
dbuf_remap_impl(dn, &dnp[i].dn_blkptr[j], lock,
tx);
}
}
}
}
/* Issue I/O to commit a dirty buffer to disk. */
static void
dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = dr->dr_dbuf;
dnode_t *dn = dr->dr_dnode;
objset_t *os;
dmu_buf_impl_t *parent = db->db_parent;
uint64_t txg = tx->tx_txg;
zbookmark_phys_t zb;
zio_prop_t zp;
zio_t *pio; /* parent I/O */
int wp_flag = 0;
ASSERT(dmu_tx_is_syncing(tx));
os = dn->dn_objset;
if (db->db_state != DB_NOFILL) {
if (db->db_level > 0 || dn->dn_type == DMU_OT_DNODE) {
/*
* Private object buffers are released here rather
* than in dbuf_dirty() since they are only modified
* in the syncing context and we don't want the
* overhead of making multiple copies of the data.
*/
if (BP_IS_HOLE(db->db_blkptr)) {
arc_buf_thaw(data);
} else {
dbuf_release_bp(db);
}
dbuf_remap(dn, db, tx);
}
}
if (parent != dn->dn_dbuf) {
/* Our parent is an indirect block. */
/* We have a dirty parent that has been scheduled for write. */
ASSERT(parent && parent->db_data_pending);
/* Our parent's buffer is one level closer to the dnode. */
ASSERT(db->db_level == parent->db_level-1);
/*
* We're about to modify our parent's db_data by modifying
* our block pointer, so the parent must be released.
*/
ASSERT(arc_released(parent->db_buf));
pio = parent->db_data_pending->dr_zio;
} else {
/* Our parent is the dnode itself. */
ASSERT((db->db_level == dn->dn_phys->dn_nlevels-1 &&
db->db_blkid != DMU_SPILL_BLKID) ||
(db->db_blkid == DMU_SPILL_BLKID && db->db_level == 0));
if (db->db_blkid != DMU_SPILL_BLKID)
ASSERT3P(db->db_blkptr, ==,
&dn->dn_phys->dn_blkptr[db->db_blkid]);
pio = dn->dn_zio;
}
ASSERT(db->db_level == 0 || data == db->db_buf);
ASSERT3U(db->db_blkptr->blk_birth, <=, txg);
ASSERT(pio);
SET_BOOKMARK(&zb, os->os_dsl_dataset ?
os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
db->db.db_object, db->db_level, db->db_blkid);
if (db->db_blkid == DMU_SPILL_BLKID)
wp_flag = WP_SPILL;
wp_flag |= (db->db_state == DB_NOFILL) ? WP_NOFILL : 0;
dmu_write_policy(os, dn, db->db_level, wp_flag, &zp);
/*
* We copy the blkptr now (rather than when we instantiate the dirty
* record), because its value can change between open context and
* syncing context. We do not need to hold dn_struct_rwlock to read
* db_blkptr because we are in syncing context.
*/
dr->dr_bp_copy = *db->db_blkptr;
if (db->db_level == 0 &&
dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
/*
* The BP for this block has been provided by open context
* (by dmu_sync() or dmu_buf_write_embedded()).
*/
abd_t *contents = (data != NULL) ?
abd_get_from_buf(data->b_data, arc_buf_size(data)) : NULL;
dr->dr_zio = zio_write(pio, os->os_spa, txg, &dr->dr_bp_copy,
contents, db->db.db_size, db->db.db_size, &zp,
dbuf_write_override_ready, NULL, NULL,
dbuf_write_override_done,
dr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
mutex_enter(&db->db_mtx);
dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
zio_write_override(dr->dr_zio, &dr->dt.dl.dr_overridden_by,
dr->dt.dl.dr_copies, dr->dt.dl.dr_nopwrite);
mutex_exit(&db->db_mtx);
} else if (db->db_state == DB_NOFILL) {
ASSERT(zp.zp_checksum == ZIO_CHECKSUM_OFF ||
zp.zp_checksum == ZIO_CHECKSUM_NOPARITY);
dr->dr_zio = zio_write(pio, os->os_spa, txg,
&dr->dr_bp_copy, NULL, db->db.db_size, db->db.db_size, &zp,
dbuf_write_nofill_ready, NULL, NULL,
dbuf_write_nofill_done, db,
ZIO_PRIORITY_ASYNC_WRITE,
ZIO_FLAG_MUSTSUCCEED | ZIO_FLAG_NODATA, &zb);
} else {
ASSERT(arc_released(data));
/*
* For indirect blocks, we want to setup the children
* ready callback so that we can properly handle an indirect
* block that only contains holes.
*/
arc_write_done_func_t *children_ready_cb = NULL;
if (db->db_level != 0)
children_ready_cb = dbuf_write_children_ready;
dr->dr_zio = arc_write(pio, os->os_spa, txg,
&dr->dr_bp_copy, data, DBUF_IS_L2CACHEABLE(db),
&zp, dbuf_write_ready,
children_ready_cb, dbuf_write_physdone,
dbuf_write_done, db, ZIO_PRIORITY_ASYNC_WRITE,
ZIO_FLAG_MUSTSUCCEED, &zb);
}
}
EXPORT_SYMBOL(dbuf_find);
EXPORT_SYMBOL(dbuf_is_metadata);
EXPORT_SYMBOL(dbuf_destroy);
EXPORT_SYMBOL(dbuf_loan_arcbuf);
EXPORT_SYMBOL(dbuf_whichblock);
EXPORT_SYMBOL(dbuf_read);
EXPORT_SYMBOL(dbuf_unoverride);
EXPORT_SYMBOL(dbuf_free_range);
EXPORT_SYMBOL(dbuf_new_size);
EXPORT_SYMBOL(dbuf_release_bp);
EXPORT_SYMBOL(dbuf_dirty);
EXPORT_SYMBOL(dmu_buf_set_crypt_params);
EXPORT_SYMBOL(dmu_buf_will_dirty);
EXPORT_SYMBOL(dmu_buf_is_dirty);
EXPORT_SYMBOL(dmu_buf_will_not_fill);
EXPORT_SYMBOL(dmu_buf_will_fill);
EXPORT_SYMBOL(dmu_buf_fill_done);
EXPORT_SYMBOL(dmu_buf_rele);
EXPORT_SYMBOL(dbuf_assign_arcbuf);
EXPORT_SYMBOL(dbuf_prefetch);
EXPORT_SYMBOL(dbuf_hold_impl);
EXPORT_SYMBOL(dbuf_hold);
EXPORT_SYMBOL(dbuf_hold_level);
EXPORT_SYMBOL(dbuf_create_bonus);
EXPORT_SYMBOL(dbuf_spill_set_blksz);
EXPORT_SYMBOL(dbuf_rm_spill);
EXPORT_SYMBOL(dbuf_add_ref);
EXPORT_SYMBOL(dbuf_rele);
EXPORT_SYMBOL(dbuf_rele_and_unlock);
EXPORT_SYMBOL(dbuf_refcount);
EXPORT_SYMBOL(dbuf_sync_list);
EXPORT_SYMBOL(dmu_buf_set_user);
EXPORT_SYMBOL(dmu_buf_set_user_ie);
EXPORT_SYMBOL(dmu_buf_get_user);
EXPORT_SYMBOL(dmu_buf_get_blkptr);
/* BEGIN CSTYLED */
ZFS_MODULE_PARAM(zfs_dbuf_cache, dbuf_cache_, max_bytes, ULONG, ZMOD_RW,
"Maximum size in bytes of the dbuf cache.");
ZFS_MODULE_PARAM(zfs_dbuf_cache, dbuf_cache_, hiwater_pct, UINT, ZMOD_RW,
"Percentage over dbuf_cache_max_bytes when dbufs must be evicted "
"directly.");
ZFS_MODULE_PARAM(zfs_dbuf_cache, dbuf_cache_, lowater_pct, UINT, ZMOD_RW,
"Percentage below dbuf_cache_max_bytes when the evict thread stops "
"evicting dbufs.");
ZFS_MODULE_PARAM(zfs_dbuf, dbuf_, metadata_cache_max_bytes, ULONG, ZMOD_RW,
"Maximum size in bytes of the dbuf metadata cache.");
ZFS_MODULE_PARAM(zfs_dbuf, dbuf_, cache_shift, INT, ZMOD_RW,
"Set the size of the dbuf cache to a log2 fraction of arc size.");
ZFS_MODULE_PARAM(zfs_dbuf, dbuf_, metadata_cache_shift, INT, ZMOD_RW,
"Set the size of the dbuf metadata cache to a log2 fraction of arc "
"size.");
/* END CSTYLED */
diff --git a/sys/contrib/openzfs/module/zfs/ddt.c b/sys/contrib/openzfs/module/zfs/ddt.c
index b94a9f54ece3..7b0b1d896761 100644
--- a/sys/contrib/openzfs/module/zfs/ddt.c
+++ b/sys/contrib/openzfs/module/zfs/ddt.c
@@ -1,1187 +1,1187 @@
/*
* 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) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012, 2016 by Delphix. All rights reserved.
*/
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
static kmem_cache_t *ddt_cache;
static kmem_cache_t *ddt_entry_cache;
/*
* Enable/disable prefetching of dedup-ed blocks which are going to be freed.
*/
int zfs_dedup_prefetch = 0;
static const ddt_ops_t *ddt_ops[DDT_TYPES] = {
&ddt_zap_ops,
};
static const char *ddt_class_name[DDT_CLASSES] = {
"ditto",
"duplicate",
"unique",
};
static void
ddt_object_create(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
dmu_tx_t *tx)
{
spa_t *spa = ddt->ddt_spa;
objset_t *os = ddt->ddt_os;
uint64_t *objectp = &ddt->ddt_object[type][class];
boolean_t prehash = zio_checksum_table[ddt->ddt_checksum].ci_flags &
ZCHECKSUM_FLAG_DEDUP;
char name[DDT_NAMELEN];
ddt_object_name(ddt, type, class, name);
ASSERT(*objectp == 0);
VERIFY(ddt_ops[type]->ddt_op_create(os, objectp, tx, prehash) == 0);
ASSERT(*objectp != 0);
VERIFY(zap_add(os, DMU_POOL_DIRECTORY_OBJECT, name,
sizeof (uint64_t), 1, objectp, tx) == 0);
VERIFY(zap_add(os, spa->spa_ddt_stat_object, name,
sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t),
&ddt->ddt_histogram[type][class], tx) == 0);
}
static void
ddt_object_destroy(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
dmu_tx_t *tx)
{
spa_t *spa = ddt->ddt_spa;
objset_t *os = ddt->ddt_os;
uint64_t *objectp = &ddt->ddt_object[type][class];
uint64_t count;
char name[DDT_NAMELEN];
ddt_object_name(ddt, type, class, name);
ASSERT(*objectp != 0);
ASSERT(ddt_histogram_empty(&ddt->ddt_histogram[type][class]));
VERIFY(ddt_object_count(ddt, type, class, &count) == 0 && count == 0);
VERIFY(zap_remove(os, DMU_POOL_DIRECTORY_OBJECT, name, tx) == 0);
VERIFY(zap_remove(os, spa->spa_ddt_stat_object, name, tx) == 0);
VERIFY(ddt_ops[type]->ddt_op_destroy(os, *objectp, tx) == 0);
bzero(&ddt->ddt_object_stats[type][class], sizeof (ddt_object_t));
*objectp = 0;
}
static int
ddt_object_load(ddt_t *ddt, enum ddt_type type, enum ddt_class class)
{
ddt_object_t *ddo = &ddt->ddt_object_stats[type][class];
dmu_object_info_t doi;
uint64_t count;
char name[DDT_NAMELEN];
int error;
ddt_object_name(ddt, type, class, name);
error = zap_lookup(ddt->ddt_os, DMU_POOL_DIRECTORY_OBJECT, name,
sizeof (uint64_t), 1, &ddt->ddt_object[type][class]);
if (error != 0)
return (error);
error = zap_lookup(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name,
sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t),
&ddt->ddt_histogram[type][class]);
if (error != 0)
return (error);
/*
* Seed the cached statistics.
*/
error = ddt_object_info(ddt, type, class, &doi);
if (error)
return (error);
error = ddt_object_count(ddt, type, class, &count);
if (error)
return (error);
ddo->ddo_count = count;
ddo->ddo_dspace = doi.doi_physical_blocks_512 << 9;
ddo->ddo_mspace = doi.doi_fill_count * doi.doi_data_block_size;
return (0);
}
static void
ddt_object_sync(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
dmu_tx_t *tx)
{
ddt_object_t *ddo = &ddt->ddt_object_stats[type][class];
dmu_object_info_t doi;
uint64_t count;
char name[DDT_NAMELEN];
ddt_object_name(ddt, type, class, name);
VERIFY(zap_update(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name,
sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t),
&ddt->ddt_histogram[type][class], tx) == 0);
/*
* Cache DDT statistics; this is the only time they'll change.
*/
VERIFY(ddt_object_info(ddt, type, class, &doi) == 0);
VERIFY(ddt_object_count(ddt, type, class, &count) == 0);
ddo->ddo_count = count;
ddo->ddo_dspace = doi.doi_physical_blocks_512 << 9;
ddo->ddo_mspace = doi.doi_fill_count * doi.doi_data_block_size;
}
static int
ddt_object_lookup(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
ddt_entry_t *dde)
{
if (!ddt_object_exists(ddt, type, class))
return (SET_ERROR(ENOENT));
return (ddt_ops[type]->ddt_op_lookup(ddt->ddt_os,
ddt->ddt_object[type][class], dde));
}
static void
ddt_object_prefetch(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
ddt_entry_t *dde)
{
if (!ddt_object_exists(ddt, type, class))
return;
ddt_ops[type]->ddt_op_prefetch(ddt->ddt_os,
ddt->ddt_object[type][class], dde);
}
int
ddt_object_update(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
ddt_entry_t *dde, dmu_tx_t *tx)
{
ASSERT(ddt_object_exists(ddt, type, class));
return (ddt_ops[type]->ddt_op_update(ddt->ddt_os,
ddt->ddt_object[type][class], dde, tx));
}
static int
ddt_object_remove(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
ddt_entry_t *dde, dmu_tx_t *tx)
{
ASSERT(ddt_object_exists(ddt, type, class));
return (ddt_ops[type]->ddt_op_remove(ddt->ddt_os,
ddt->ddt_object[type][class], dde, tx));
}
int
ddt_object_walk(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
uint64_t *walk, ddt_entry_t *dde)
{
ASSERT(ddt_object_exists(ddt, type, class));
return (ddt_ops[type]->ddt_op_walk(ddt->ddt_os,
ddt->ddt_object[type][class], dde, walk));
}
int
ddt_object_count(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
uint64_t *count)
{
ASSERT(ddt_object_exists(ddt, type, class));
return (ddt_ops[type]->ddt_op_count(ddt->ddt_os,
ddt->ddt_object[type][class], count));
}
int
ddt_object_info(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
dmu_object_info_t *doi)
{
if (!ddt_object_exists(ddt, type, class))
return (SET_ERROR(ENOENT));
return (dmu_object_info(ddt->ddt_os, ddt->ddt_object[type][class],
doi));
}
boolean_t
ddt_object_exists(ddt_t *ddt, enum ddt_type type, enum ddt_class class)
{
return (!!ddt->ddt_object[type][class]);
}
void
ddt_object_name(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
char *name)
{
(void) snprintf(name, DDT_NAMELEN, DMU_POOL_DDT,
zio_checksum_table[ddt->ddt_checksum].ci_name,
ddt_ops[type]->ddt_op_name, ddt_class_name[class]);
}
void
ddt_bp_fill(const ddt_phys_t *ddp, blkptr_t *bp, uint64_t txg)
{
ASSERT(txg != 0);
for (int d = 0; d < SPA_DVAS_PER_BP; d++)
bp->blk_dva[d] = ddp->ddp_dva[d];
BP_SET_BIRTH(bp, txg, ddp->ddp_phys_birth);
}
/*
* The bp created via this function may be used for repairs and scrub, but it
* will be missing the salt / IV required to do a full decrypting read.
*/
void
ddt_bp_create(enum zio_checksum checksum,
const ddt_key_t *ddk, const ddt_phys_t *ddp, blkptr_t *bp)
{
BP_ZERO(bp);
if (ddp != NULL)
ddt_bp_fill(ddp, bp, ddp->ddp_phys_birth);
bp->blk_cksum = ddk->ddk_cksum;
BP_SET_LSIZE(bp, DDK_GET_LSIZE(ddk));
BP_SET_PSIZE(bp, DDK_GET_PSIZE(ddk));
BP_SET_COMPRESS(bp, DDK_GET_COMPRESS(ddk));
BP_SET_CRYPT(bp, DDK_GET_CRYPT(ddk));
BP_SET_FILL(bp, 1);
BP_SET_CHECKSUM(bp, checksum);
BP_SET_TYPE(bp, DMU_OT_DEDUP);
BP_SET_LEVEL(bp, 0);
BP_SET_DEDUP(bp, 1);
BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
}
void
ddt_key_fill(ddt_key_t *ddk, const blkptr_t *bp)
{
ddk->ddk_cksum = bp->blk_cksum;
ddk->ddk_prop = 0;
ASSERT(BP_IS_ENCRYPTED(bp) || !BP_USES_CRYPT(bp));
DDK_SET_LSIZE(ddk, BP_GET_LSIZE(bp));
DDK_SET_PSIZE(ddk, BP_GET_PSIZE(bp));
DDK_SET_COMPRESS(ddk, BP_GET_COMPRESS(bp));
DDK_SET_CRYPT(ddk, BP_USES_CRYPT(bp));
}
void
ddt_phys_fill(ddt_phys_t *ddp, const blkptr_t *bp)
{
ASSERT(ddp->ddp_phys_birth == 0);
for (int d = 0; d < SPA_DVAS_PER_BP; d++)
ddp->ddp_dva[d] = bp->blk_dva[d];
ddp->ddp_phys_birth = BP_PHYSICAL_BIRTH(bp);
}
void
ddt_phys_clear(ddt_phys_t *ddp)
{
bzero(ddp, sizeof (*ddp));
}
void
ddt_phys_addref(ddt_phys_t *ddp)
{
ddp->ddp_refcnt++;
}
void
ddt_phys_decref(ddt_phys_t *ddp)
{
if (ddp) {
ASSERT(ddp->ddp_refcnt > 0);
ddp->ddp_refcnt--;
}
}
void
ddt_phys_free(ddt_t *ddt, ddt_key_t *ddk, ddt_phys_t *ddp, uint64_t txg)
{
blkptr_t blk;
ddt_bp_create(ddt->ddt_checksum, ddk, ddp, &blk);
/*
* We clear the dedup bit so that zio_free() will actually free the
* space, rather than just decrementing the refcount in the DDT.
*/
BP_SET_DEDUP(&blk, 0);
ddt_phys_clear(ddp);
zio_free(ddt->ddt_spa, txg, &blk);
}
ddt_phys_t *
ddt_phys_select(const ddt_entry_t *dde, const blkptr_t *bp)
{
ddt_phys_t *ddp = (ddt_phys_t *)dde->dde_phys;
for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
if (DVA_EQUAL(BP_IDENTITY(bp), &ddp->ddp_dva[0]) &&
BP_PHYSICAL_BIRTH(bp) == ddp->ddp_phys_birth)
return (ddp);
}
return (NULL);
}
uint64_t
ddt_phys_total_refcnt(const ddt_entry_t *dde)
{
uint64_t refcnt = 0;
for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++)
refcnt += dde->dde_phys[p].ddp_refcnt;
return (refcnt);
}
static void
ddt_stat_generate(ddt_t *ddt, ddt_entry_t *dde, ddt_stat_t *dds)
{
spa_t *spa = ddt->ddt_spa;
ddt_phys_t *ddp = dde->dde_phys;
ddt_key_t *ddk = &dde->dde_key;
uint64_t lsize = DDK_GET_LSIZE(ddk);
uint64_t psize = DDK_GET_PSIZE(ddk);
bzero(dds, sizeof (*dds));
for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
uint64_t dsize = 0;
uint64_t refcnt = ddp->ddp_refcnt;
if (ddp->ddp_phys_birth == 0)
continue;
for (int d = 0; d < DDE_GET_NDVAS(dde); d++)
dsize += dva_get_dsize_sync(spa, &ddp->ddp_dva[d]);
dds->dds_blocks += 1;
dds->dds_lsize += lsize;
dds->dds_psize += psize;
dds->dds_dsize += dsize;
dds->dds_ref_blocks += refcnt;
dds->dds_ref_lsize += lsize * refcnt;
dds->dds_ref_psize += psize * refcnt;
dds->dds_ref_dsize += dsize * refcnt;
}
}
void
ddt_stat_add(ddt_stat_t *dst, const ddt_stat_t *src, uint64_t neg)
{
const uint64_t *s = (const uint64_t *)src;
uint64_t *d = (uint64_t *)dst;
uint64_t *d_end = (uint64_t *)(dst + 1);
ASSERT(neg == 0 || neg == -1ULL); /* add or subtract */
- while (d < d_end)
- *d++ += (*s++ ^ neg) - neg;
+ for (int i = 0; i < d_end - d; i++)
+ d[i] += (s[i] ^ neg) - neg;
}
static void
ddt_stat_update(ddt_t *ddt, ddt_entry_t *dde, uint64_t neg)
{
ddt_stat_t dds;
ddt_histogram_t *ddh;
int bucket;
ddt_stat_generate(ddt, dde, &dds);
bucket = highbit64(dds.dds_ref_blocks) - 1;
ASSERT(bucket >= 0);
ddh = &ddt->ddt_histogram[dde->dde_type][dde->dde_class];
ddt_stat_add(&ddh->ddh_stat[bucket], &dds, neg);
}
void
ddt_histogram_add(ddt_histogram_t *dst, const ddt_histogram_t *src)
{
for (int h = 0; h < 64; h++)
ddt_stat_add(&dst->ddh_stat[h], &src->ddh_stat[h], 0);
}
void
ddt_histogram_stat(ddt_stat_t *dds, const ddt_histogram_t *ddh)
{
bzero(dds, sizeof (*dds));
for (int h = 0; h < 64; h++)
ddt_stat_add(dds, &ddh->ddh_stat[h], 0);
}
boolean_t
ddt_histogram_empty(const ddt_histogram_t *ddh)
{
const uint64_t *s = (const uint64_t *)ddh;
const uint64_t *s_end = (const uint64_t *)(ddh + 1);
while (s < s_end)
if (*s++ != 0)
return (B_FALSE);
return (B_TRUE);
}
void
ddt_get_dedup_object_stats(spa_t *spa, ddt_object_t *ddo_total)
{
/* Sum the statistics we cached in ddt_object_sync(). */
for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
ddt_t *ddt = spa->spa_ddt[c];
for (enum ddt_type type = 0; type < DDT_TYPES; type++) {
for (enum ddt_class class = 0; class < DDT_CLASSES;
class++) {
ddt_object_t *ddo =
&ddt->ddt_object_stats[type][class];
ddo_total->ddo_count += ddo->ddo_count;
ddo_total->ddo_dspace += ddo->ddo_dspace;
ddo_total->ddo_mspace += ddo->ddo_mspace;
}
}
}
/* ... and compute the averages. */
if (ddo_total->ddo_count != 0) {
ddo_total->ddo_dspace /= ddo_total->ddo_count;
ddo_total->ddo_mspace /= ddo_total->ddo_count;
}
}
void
ddt_get_dedup_histogram(spa_t *spa, ddt_histogram_t *ddh)
{
for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
ddt_t *ddt = spa->spa_ddt[c];
for (enum ddt_type type = 0; type < DDT_TYPES; type++) {
for (enum ddt_class class = 0; class < DDT_CLASSES;
class++) {
ddt_histogram_add(ddh,
&ddt->ddt_histogram_cache[type][class]);
}
}
}
}
void
ddt_get_dedup_stats(spa_t *spa, ddt_stat_t *dds_total)
{
ddt_histogram_t *ddh_total;
ddh_total = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
ddt_get_dedup_histogram(spa, ddh_total);
ddt_histogram_stat(dds_total, ddh_total);
kmem_free(ddh_total, sizeof (ddt_histogram_t));
}
uint64_t
ddt_get_dedup_dspace(spa_t *spa)
{
ddt_stat_t dds_total;
if (spa->spa_dedup_dspace != ~0ULL)
return (spa->spa_dedup_dspace);
bzero(&dds_total, sizeof (ddt_stat_t));
/* Calculate and cache the stats */
ddt_get_dedup_stats(spa, &dds_total);
spa->spa_dedup_dspace = dds_total.dds_ref_dsize - dds_total.dds_dsize;
return (spa->spa_dedup_dspace);
}
uint64_t
ddt_get_pool_dedup_ratio(spa_t *spa)
{
ddt_stat_t dds_total = { 0 };
ddt_get_dedup_stats(spa, &dds_total);
if (dds_total.dds_dsize == 0)
return (100);
return (dds_total.dds_ref_dsize * 100 / dds_total.dds_dsize);
}
size_t
ddt_compress(void *src, uchar_t *dst, size_t s_len, size_t d_len)
{
uchar_t *version = dst++;
int cpfunc = ZIO_COMPRESS_ZLE;
zio_compress_info_t *ci = &zio_compress_table[cpfunc];
size_t c_len;
ASSERT(d_len >= s_len + 1); /* no compression plus version byte */
c_len = ci->ci_compress(src, dst, s_len, d_len - 1, ci->ci_level);
if (c_len == s_len) {
cpfunc = ZIO_COMPRESS_OFF;
bcopy(src, dst, s_len);
}
*version = cpfunc;
/* CONSTCOND */
if (ZFS_HOST_BYTEORDER)
*version |= DDT_COMPRESS_BYTEORDER_MASK;
return (c_len + 1);
}
void
ddt_decompress(uchar_t *src, void *dst, size_t s_len, size_t d_len)
{
uchar_t version = *src++;
int cpfunc = version & DDT_COMPRESS_FUNCTION_MASK;
zio_compress_info_t *ci = &zio_compress_table[cpfunc];
if (ci->ci_decompress != NULL)
(void) ci->ci_decompress(src, dst, s_len, d_len, ci->ci_level);
else
bcopy(src, dst, d_len);
if (((version & DDT_COMPRESS_BYTEORDER_MASK) != 0) !=
(ZFS_HOST_BYTEORDER != 0))
byteswap_uint64_array(dst, d_len);
}
ddt_t *
ddt_select(spa_t *spa, const blkptr_t *bp)
{
return (spa->spa_ddt[BP_GET_CHECKSUM(bp)]);
}
void
ddt_enter(ddt_t *ddt)
{
mutex_enter(&ddt->ddt_lock);
}
void
ddt_exit(ddt_t *ddt)
{
mutex_exit(&ddt->ddt_lock);
}
void
ddt_init(void)
{
ddt_cache = kmem_cache_create("ddt_cache",
sizeof (ddt_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
ddt_entry_cache = kmem_cache_create("ddt_entry_cache",
sizeof (ddt_entry_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
}
void
ddt_fini(void)
{
kmem_cache_destroy(ddt_entry_cache);
kmem_cache_destroy(ddt_cache);
}
static ddt_entry_t *
ddt_alloc(const ddt_key_t *ddk)
{
ddt_entry_t *dde;
dde = kmem_cache_alloc(ddt_entry_cache, KM_SLEEP);
bzero(dde, sizeof (ddt_entry_t));
cv_init(&dde->dde_cv, NULL, CV_DEFAULT, NULL);
dde->dde_key = *ddk;
return (dde);
}
static void
ddt_free(ddt_entry_t *dde)
{
ASSERT(!dde->dde_loading);
for (int p = 0; p < DDT_PHYS_TYPES; p++)
ASSERT(dde->dde_lead_zio[p] == NULL);
if (dde->dde_repair_abd != NULL)
abd_free(dde->dde_repair_abd);
cv_destroy(&dde->dde_cv);
kmem_cache_free(ddt_entry_cache, dde);
}
void
ddt_remove(ddt_t *ddt, ddt_entry_t *dde)
{
ASSERT(MUTEX_HELD(&ddt->ddt_lock));
avl_remove(&ddt->ddt_tree, dde);
ddt_free(dde);
}
ddt_entry_t *
ddt_lookup(ddt_t *ddt, const blkptr_t *bp, boolean_t add)
{
ddt_entry_t *dde, dde_search;
enum ddt_type type;
enum ddt_class class;
avl_index_t where;
int error;
ASSERT(MUTEX_HELD(&ddt->ddt_lock));
ddt_key_fill(&dde_search.dde_key, bp);
dde = avl_find(&ddt->ddt_tree, &dde_search, &where);
if (dde == NULL) {
if (!add)
return (NULL);
dde = ddt_alloc(&dde_search.dde_key);
avl_insert(&ddt->ddt_tree, dde, where);
}
while (dde->dde_loading)
cv_wait(&dde->dde_cv, &ddt->ddt_lock);
if (dde->dde_loaded)
return (dde);
dde->dde_loading = B_TRUE;
ddt_exit(ddt);
error = ENOENT;
for (type = 0; type < DDT_TYPES; type++) {
for (class = 0; class < DDT_CLASSES; class++) {
error = ddt_object_lookup(ddt, type, class, dde);
if (error != ENOENT) {
ASSERT0(error);
break;
}
}
if (error != ENOENT)
break;
}
ddt_enter(ddt);
ASSERT(dde->dde_loaded == B_FALSE);
ASSERT(dde->dde_loading == B_TRUE);
dde->dde_type = type; /* will be DDT_TYPES if no entry found */
dde->dde_class = class; /* will be DDT_CLASSES if no entry found */
dde->dde_loaded = B_TRUE;
dde->dde_loading = B_FALSE;
if (error == 0)
ddt_stat_update(ddt, dde, -1ULL);
cv_broadcast(&dde->dde_cv);
return (dde);
}
void
ddt_prefetch(spa_t *spa, const blkptr_t *bp)
{
ddt_t *ddt;
ddt_entry_t dde;
if (!zfs_dedup_prefetch || bp == NULL || !BP_GET_DEDUP(bp))
return;
/*
* We only remove the DDT once all tables are empty and only
* prefetch dedup blocks when there are entries in the DDT.
* Thus no locking is required as the DDT can't disappear on us.
*/
ddt = ddt_select(spa, bp);
ddt_key_fill(&dde.dde_key, bp);
for (enum ddt_type type = 0; type < DDT_TYPES; type++) {
for (enum ddt_class class = 0; class < DDT_CLASSES; class++) {
ddt_object_prefetch(ddt, type, class, &dde);
}
}
}
/*
* Opaque struct used for ddt_key comparison
*/
#define DDT_KEY_CMP_LEN (sizeof (ddt_key_t) / sizeof (uint16_t))
typedef struct ddt_key_cmp {
uint16_t u16[DDT_KEY_CMP_LEN];
} ddt_key_cmp_t;
int
ddt_entry_compare(const void *x1, const void *x2)
{
const ddt_entry_t *dde1 = x1;
const ddt_entry_t *dde2 = x2;
const ddt_key_cmp_t *k1 = (const ddt_key_cmp_t *)&dde1->dde_key;
const ddt_key_cmp_t *k2 = (const ddt_key_cmp_t *)&dde2->dde_key;
int32_t cmp = 0;
for (int i = 0; i < DDT_KEY_CMP_LEN; i++) {
cmp = (int32_t)k1->u16[i] - (int32_t)k2->u16[i];
if (likely(cmp))
break;
}
return (TREE_ISIGN(cmp));
}
static ddt_t *
ddt_table_alloc(spa_t *spa, enum zio_checksum c)
{
ddt_t *ddt;
ddt = kmem_cache_alloc(ddt_cache, KM_SLEEP);
bzero(ddt, sizeof (ddt_t));
mutex_init(&ddt->ddt_lock, NULL, MUTEX_DEFAULT, NULL);
avl_create(&ddt->ddt_tree, ddt_entry_compare,
sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node));
avl_create(&ddt->ddt_repair_tree, ddt_entry_compare,
sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node));
ddt->ddt_checksum = c;
ddt->ddt_spa = spa;
ddt->ddt_os = spa->spa_meta_objset;
return (ddt);
}
static void
ddt_table_free(ddt_t *ddt)
{
ASSERT(avl_numnodes(&ddt->ddt_tree) == 0);
ASSERT(avl_numnodes(&ddt->ddt_repair_tree) == 0);
avl_destroy(&ddt->ddt_tree);
avl_destroy(&ddt->ddt_repair_tree);
mutex_destroy(&ddt->ddt_lock);
kmem_cache_free(ddt_cache, ddt);
}
void
ddt_create(spa_t *spa)
{
spa->spa_dedup_checksum = ZIO_DEDUPCHECKSUM;
for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++)
spa->spa_ddt[c] = ddt_table_alloc(spa, c);
}
int
ddt_load(spa_t *spa)
{
int error;
ddt_create(spa);
error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_DDT_STATS, sizeof (uint64_t), 1,
&spa->spa_ddt_stat_object);
if (error)
return (error == ENOENT ? 0 : error);
for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
ddt_t *ddt = spa->spa_ddt[c];
for (enum ddt_type type = 0; type < DDT_TYPES; type++) {
for (enum ddt_class class = 0; class < DDT_CLASSES;
class++) {
error = ddt_object_load(ddt, type, class);
if (error != 0 && error != ENOENT)
return (error);
}
}
/*
* Seed the cached histograms.
*/
bcopy(ddt->ddt_histogram, &ddt->ddt_histogram_cache,
sizeof (ddt->ddt_histogram));
spa->spa_dedup_dspace = ~0ULL;
}
return (0);
}
void
ddt_unload(spa_t *spa)
{
for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
if (spa->spa_ddt[c]) {
ddt_table_free(spa->spa_ddt[c]);
spa->spa_ddt[c] = NULL;
}
}
}
boolean_t
ddt_class_contains(spa_t *spa, enum ddt_class max_class, const blkptr_t *bp)
{
ddt_t *ddt;
ddt_entry_t *dde;
if (!BP_GET_DEDUP(bp))
return (B_FALSE);
if (max_class == DDT_CLASS_UNIQUE)
return (B_TRUE);
ddt = spa->spa_ddt[BP_GET_CHECKSUM(bp)];
dde = kmem_cache_alloc(ddt_entry_cache, KM_SLEEP);
ddt_key_fill(&(dde->dde_key), bp);
for (enum ddt_type type = 0; type < DDT_TYPES; type++) {
for (enum ddt_class class = 0; class <= max_class; class++) {
if (ddt_object_lookup(ddt, type, class, dde) == 0) {
kmem_cache_free(ddt_entry_cache, dde);
return (B_TRUE);
}
}
}
kmem_cache_free(ddt_entry_cache, dde);
return (B_FALSE);
}
ddt_entry_t *
ddt_repair_start(ddt_t *ddt, const blkptr_t *bp)
{
ddt_key_t ddk;
ddt_entry_t *dde;
ddt_key_fill(&ddk, bp);
dde = ddt_alloc(&ddk);
for (enum ddt_type type = 0; type < DDT_TYPES; type++) {
for (enum ddt_class class = 0; class < DDT_CLASSES; class++) {
/*
* We can only do repair if there are multiple copies
* of the block. For anything in the UNIQUE class,
* there's definitely only one copy, so don't even try.
*/
if (class != DDT_CLASS_UNIQUE &&
ddt_object_lookup(ddt, type, class, dde) == 0)
return (dde);
}
}
bzero(dde->dde_phys, sizeof (dde->dde_phys));
return (dde);
}
void
ddt_repair_done(ddt_t *ddt, ddt_entry_t *dde)
{
avl_index_t where;
ddt_enter(ddt);
if (dde->dde_repair_abd != NULL && spa_writeable(ddt->ddt_spa) &&
avl_find(&ddt->ddt_repair_tree, dde, &where) == NULL)
avl_insert(&ddt->ddt_repair_tree, dde, where);
else
ddt_free(dde);
ddt_exit(ddt);
}
static void
ddt_repair_entry_done(zio_t *zio)
{
ddt_entry_t *rdde = zio->io_private;
ddt_free(rdde);
}
static void
ddt_repair_entry(ddt_t *ddt, ddt_entry_t *dde, ddt_entry_t *rdde, zio_t *rio)
{
ddt_phys_t *ddp = dde->dde_phys;
ddt_phys_t *rddp = rdde->dde_phys;
ddt_key_t *ddk = &dde->dde_key;
ddt_key_t *rddk = &rdde->dde_key;
zio_t *zio;
blkptr_t blk;
zio = zio_null(rio, rio->io_spa, NULL,
ddt_repair_entry_done, rdde, rio->io_flags);
for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++, rddp++) {
if (ddp->ddp_phys_birth == 0 ||
ddp->ddp_phys_birth != rddp->ddp_phys_birth ||
bcmp(ddp->ddp_dva, rddp->ddp_dva, sizeof (ddp->ddp_dva)))
continue;
ddt_bp_create(ddt->ddt_checksum, ddk, ddp, &blk);
zio_nowait(zio_rewrite(zio, zio->io_spa, 0, &blk,
rdde->dde_repair_abd, DDK_GET_PSIZE(rddk), NULL, NULL,
ZIO_PRIORITY_SYNC_WRITE, ZIO_DDT_CHILD_FLAGS(zio), NULL));
}
zio_nowait(zio);
}
static void
ddt_repair_table(ddt_t *ddt, zio_t *rio)
{
spa_t *spa = ddt->ddt_spa;
ddt_entry_t *dde, *rdde_next, *rdde;
avl_tree_t *t = &ddt->ddt_repair_tree;
blkptr_t blk;
if (spa_sync_pass(spa) > 1)
return;
ddt_enter(ddt);
for (rdde = avl_first(t); rdde != NULL; rdde = rdde_next) {
rdde_next = AVL_NEXT(t, rdde);
avl_remove(&ddt->ddt_repair_tree, rdde);
ddt_exit(ddt);
ddt_bp_create(ddt->ddt_checksum, &rdde->dde_key, NULL, &blk);
dde = ddt_repair_start(ddt, &blk);
ddt_repair_entry(ddt, dde, rdde, rio);
ddt_repair_done(ddt, dde);
ddt_enter(ddt);
}
ddt_exit(ddt);
}
static void
ddt_sync_entry(ddt_t *ddt, ddt_entry_t *dde, dmu_tx_t *tx, uint64_t txg)
{
dsl_pool_t *dp = ddt->ddt_spa->spa_dsl_pool;
ddt_phys_t *ddp = dde->dde_phys;
ddt_key_t *ddk = &dde->dde_key;
enum ddt_type otype = dde->dde_type;
enum ddt_type ntype = DDT_TYPE_CURRENT;
enum ddt_class oclass = dde->dde_class;
enum ddt_class nclass;
uint64_t total_refcnt = 0;
ASSERT(dde->dde_loaded);
ASSERT(!dde->dde_loading);
for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
ASSERT(dde->dde_lead_zio[p] == NULL);
if (ddp->ddp_phys_birth == 0) {
ASSERT(ddp->ddp_refcnt == 0);
continue;
}
if (p == DDT_PHYS_DITTO) {
/*
* Note, we no longer create DDT-DITTO blocks, but we
* don't want to leak any written by older software.
*/
ddt_phys_free(ddt, ddk, ddp, txg);
continue;
}
if (ddp->ddp_refcnt == 0)
ddt_phys_free(ddt, ddk, ddp, txg);
total_refcnt += ddp->ddp_refcnt;
}
/* We do not create new DDT-DITTO blocks. */
ASSERT0(dde->dde_phys[DDT_PHYS_DITTO].ddp_phys_birth);
if (total_refcnt > 1)
nclass = DDT_CLASS_DUPLICATE;
else
nclass = DDT_CLASS_UNIQUE;
if (otype != DDT_TYPES &&
(otype != ntype || oclass != nclass || total_refcnt == 0)) {
VERIFY(ddt_object_remove(ddt, otype, oclass, dde, tx) == 0);
ASSERT(ddt_object_lookup(ddt, otype, oclass, dde) == ENOENT);
}
if (total_refcnt != 0) {
dde->dde_type = ntype;
dde->dde_class = nclass;
ddt_stat_update(ddt, dde, 0);
if (!ddt_object_exists(ddt, ntype, nclass))
ddt_object_create(ddt, ntype, nclass, tx);
VERIFY(ddt_object_update(ddt, ntype, nclass, dde, tx) == 0);
/*
* If the class changes, the order that we scan this bp
* changes. If it decreases, we could miss it, so
* scan it right now. (This covers both class changing
* while we are doing ddt_walk(), and when we are
* traversing.)
*/
if (nclass < oclass) {
dsl_scan_ddt_entry(dp->dp_scan,
ddt->ddt_checksum, dde, tx);
}
}
}
static void
ddt_sync_table(ddt_t *ddt, dmu_tx_t *tx, uint64_t txg)
{
spa_t *spa = ddt->ddt_spa;
ddt_entry_t *dde;
void *cookie = NULL;
if (avl_numnodes(&ddt->ddt_tree) == 0)
return;
ASSERT(spa->spa_uberblock.ub_version >= SPA_VERSION_DEDUP);
if (spa->spa_ddt_stat_object == 0) {
spa->spa_ddt_stat_object = zap_create_link(ddt->ddt_os,
DMU_OT_DDT_STATS, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_DDT_STATS, tx);
}
while ((dde = avl_destroy_nodes(&ddt->ddt_tree, &cookie)) != NULL) {
ddt_sync_entry(ddt, dde, tx, txg);
ddt_free(dde);
}
for (enum ddt_type type = 0; type < DDT_TYPES; type++) {
uint64_t add, count = 0;
for (enum ddt_class class = 0; class < DDT_CLASSES; class++) {
if (ddt_object_exists(ddt, type, class)) {
ddt_object_sync(ddt, type, class, tx);
VERIFY(ddt_object_count(ddt, type, class,
&add) == 0);
count += add;
}
}
for (enum ddt_class class = 0; class < DDT_CLASSES; class++) {
if (count == 0 && ddt_object_exists(ddt, type, class))
ddt_object_destroy(ddt, type, class, tx);
}
}
bcopy(ddt->ddt_histogram, &ddt->ddt_histogram_cache,
sizeof (ddt->ddt_histogram));
spa->spa_dedup_dspace = ~0ULL;
}
void
ddt_sync(spa_t *spa, uint64_t txg)
{
dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
dmu_tx_t *tx;
zio_t *rio;
ASSERT(spa_syncing_txg(spa) == txg);
tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
rio = zio_root(spa, NULL, NULL,
ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SELF_HEAL);
/*
* This function may cause an immediate scan of ddt blocks (see
* the comment above dsl_scan_ddt() for details). We set the
* scan's root zio here so that we can wait for any scan IOs in
* addition to the regular ddt IOs.
*/
ASSERT3P(scn->scn_zio_root, ==, NULL);
scn->scn_zio_root = rio;
for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
ddt_t *ddt = spa->spa_ddt[c];
if (ddt == NULL)
continue;
ddt_sync_table(ddt, tx, txg);
ddt_repair_table(ddt, rio);
}
(void) zio_wait(rio);
scn->scn_zio_root = NULL;
dmu_tx_commit(tx);
}
int
ddt_walk(spa_t *spa, ddt_bookmark_t *ddb, ddt_entry_t *dde)
{
do {
do {
do {
ddt_t *ddt = spa->spa_ddt[ddb->ddb_checksum];
int error = ENOENT;
if (ddt_object_exists(ddt, ddb->ddb_type,
ddb->ddb_class)) {
error = ddt_object_walk(ddt,
ddb->ddb_type, ddb->ddb_class,
&ddb->ddb_cursor, dde);
}
dde->dde_type = ddb->ddb_type;
dde->dde_class = ddb->ddb_class;
if (error == 0)
return (0);
if (error != ENOENT)
return (error);
ddb->ddb_cursor = 0;
} while (++ddb->ddb_checksum < ZIO_CHECKSUM_FUNCTIONS);
ddb->ddb_checksum = 0;
} while (++ddb->ddb_type < DDT_TYPES);
ddb->ddb_type = 0;
} while (++ddb->ddb_class < DDT_CLASSES);
return (SET_ERROR(ENOENT));
}
/* BEGIN CSTYLED */
ZFS_MODULE_PARAM(zfs_dedup, zfs_dedup_, prefetch, INT, ZMOD_RW,
"Enable prefetching dedup-ed blks");
/* END CSTYLED */
diff --git a/sys/contrib/openzfs/module/zfs/dmu_objset.c b/sys/contrib/openzfs/module/zfs/dmu_objset.c
index 22deee7f3dc9..af107fb8ad63 100644
--- a/sys/contrib/openzfs/module/zfs/dmu_objset.c
+++ b/sys/contrib/openzfs/module/zfs/dmu_objset.c
@@ -1,3042 +1,3050 @@
/*
* 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, 2020 by Delphix. All rights reserved.
* Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
* Copyright (c) 2013, Joyent, Inc. All rights reserved.
* Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
* Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
* Copyright (c) 2016 Actifio, Inc. All rights reserved.
* Copyright 2017 Nexenta Systems, Inc.
* Copyright (c) 2017 Open-E, Inc. All Rights Reserved.
* Copyright (c) 2018, loli10K . All rights reserved.
* Copyright (c) 2019, Klara Inc.
* Copyright (c) 2019, Allan Jude
*/
/* Portions Copyright 2010 Robert Milkowski */
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include "zfs_namecheck.h"
/*
* Needed to close a window in dnode_move() that allows the objset to be freed
* before it can be safely accessed.
*/
krwlock_t os_lock;
/*
* Tunable to overwrite the maximum number of threads for the parallelization
* of dmu_objset_find_dp, needed to speed up the import of pools with many
* datasets.
* Default is 4 times the number of leaf vdevs.
*/
int dmu_find_threads = 0;
/*
* Backfill lower metadnode objects after this many have been freed.
* Backfilling negatively impacts object creation rates, so only do it
* if there are enough holes to fill.
*/
int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT;
static char *upgrade_tag = "upgrade_tag";
static void dmu_objset_find_dp_cb(void *arg);
static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb);
static void dmu_objset_upgrade_stop(objset_t *os);
void
dmu_objset_init(void)
{
rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
}
void
dmu_objset_fini(void)
{
rw_destroy(&os_lock);
}
spa_t *
dmu_objset_spa(objset_t *os)
{
return (os->os_spa);
}
zilog_t *
dmu_objset_zil(objset_t *os)
{
return (os->os_zil);
}
dsl_pool_t *
dmu_objset_pool(objset_t *os)
{
dsl_dataset_t *ds;
if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
return (ds->ds_dir->dd_pool);
else
return (spa_get_dsl(os->os_spa));
}
dsl_dataset_t *
dmu_objset_ds(objset_t *os)
{
return (os->os_dsl_dataset);
}
dmu_objset_type_t
dmu_objset_type(objset_t *os)
{
return (os->os_phys->os_type);
}
void
dmu_objset_name(objset_t *os, char *buf)
{
dsl_dataset_name(os->os_dsl_dataset, buf);
}
uint64_t
dmu_objset_id(objset_t *os)
{
dsl_dataset_t *ds = os->os_dsl_dataset;
return (ds ? ds->ds_object : 0);
}
uint64_t
dmu_objset_dnodesize(objset_t *os)
{
return (os->os_dnodesize);
}
zfs_sync_type_t
dmu_objset_syncprop(objset_t *os)
{
return (os->os_sync);
}
zfs_logbias_op_t
dmu_objset_logbias(objset_t *os)
{
return (os->os_logbias);
}
static void
checksum_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
/*
* Inheritance should have been done by now.
*/
ASSERT(newval != ZIO_CHECKSUM_INHERIT);
os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
}
static void
compression_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
/*
* Inheritance and range checking should have been done by now.
*/
ASSERT(newval != ZIO_COMPRESS_INHERIT);
os->os_compress = zio_compress_select(os->os_spa,
ZIO_COMPRESS_ALGO(newval), ZIO_COMPRESS_ON);
os->os_complevel = zio_complevel_select(os->os_spa, os->os_compress,
ZIO_COMPRESS_LEVEL(newval), ZIO_COMPLEVEL_DEFAULT);
}
static void
copies_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
/*
* Inheritance and range checking should have been done by now.
*/
ASSERT(newval > 0);
ASSERT(newval <= spa_max_replication(os->os_spa));
os->os_copies = newval;
}
static void
dedup_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
spa_t *spa = os->os_spa;
enum zio_checksum checksum;
/*
* Inheritance should have been done by now.
*/
ASSERT(newval != ZIO_CHECKSUM_INHERIT);
checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
}
static void
primary_cache_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
/*
* Inheritance and range checking should have been done by now.
*/
ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
newval == ZFS_CACHE_METADATA);
os->os_primary_cache = newval;
}
static void
secondary_cache_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
/*
* Inheritance and range checking should have been done by now.
*/
ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
newval == ZFS_CACHE_METADATA);
os->os_secondary_cache = newval;
}
static void
sync_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
/*
* Inheritance and range checking should have been done by now.
*/
ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
newval == ZFS_SYNC_DISABLED);
os->os_sync = newval;
if (os->os_zil)
zil_set_sync(os->os_zil, newval);
}
static void
redundant_metadata_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
/*
* Inheritance and range checking should have been done by now.
*/
ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
newval == ZFS_REDUNDANT_METADATA_MOST);
os->os_redundant_metadata = newval;
}
static void
dnodesize_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
switch (newval) {
case ZFS_DNSIZE_LEGACY:
os->os_dnodesize = DNODE_MIN_SIZE;
break;
case ZFS_DNSIZE_AUTO:
/*
* Choose a dnode size that will work well for most
* workloads if the user specified "auto". Future code
* improvements could dynamically select a dnode size
* based on observed workload patterns.
*/
os->os_dnodesize = DNODE_MIN_SIZE * 2;
break;
case ZFS_DNSIZE_1K:
case ZFS_DNSIZE_2K:
case ZFS_DNSIZE_4K:
case ZFS_DNSIZE_8K:
case ZFS_DNSIZE_16K:
os->os_dnodesize = newval;
break;
}
}
static void
smallblk_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
/*
* Inheritance and range checking should have been done by now.
*/
ASSERT(newval <= SPA_MAXBLOCKSIZE);
ASSERT(ISP2(newval));
os->os_zpl_special_smallblock = newval;
}
static void
logbias_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
newval == ZFS_LOGBIAS_THROUGHPUT);
os->os_logbias = newval;
if (os->os_zil)
zil_set_logbias(os->os_zil, newval);
}
static void
recordsize_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
os->os_recordsize = newval;
}
void
dmu_objset_byteswap(void *buf, size_t size)
{
objset_phys_t *osp = buf;
ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 ||
size == sizeof (objset_phys_t));
dnode_byteswap(&osp->os_meta_dnode);
byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
osp->os_type = BSWAP_64(osp->os_type);
osp->os_flags = BSWAP_64(osp->os_flags);
if (size >= OBJSET_PHYS_SIZE_V2) {
dnode_byteswap(&osp->os_userused_dnode);
dnode_byteswap(&osp->os_groupused_dnode);
if (size >= sizeof (objset_phys_t))
dnode_byteswap(&osp->os_projectused_dnode);
}
}
/*
* The hash is a CRC-based hash of the objset_t pointer and the object number.
*/
static uint64_t
dnode_hash(const objset_t *os, uint64_t obj)
{
uintptr_t osv = (uintptr_t)os;
uint64_t crc = -1ULL;
ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
/*
* The low 6 bits of the pointer don't have much entropy, because
* the objset_t is larger than 2^6 bytes long.
*/
crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF];
crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF];
crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF];
crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF];
crc ^= (osv>>14) ^ (obj>>24);
return (crc);
}
static unsigned int
dnode_multilist_index_func(multilist_t *ml, void *obj)
{
dnode_t *dn = obj;
- return (dnode_hash(dn->dn_objset, dn->dn_object) %
+
+ /*
+ * The low order bits of the hash value are thought to be
+ * distributed evenly. Otherwise, in the case that the multilist
+ * has a power of two number of sublists, each sublists' usage
+ * would not be evenly distributed. In this context full 64bit
+ * division would be a waste of time, so limit it to 32 bits.
+ */
+ return ((unsigned int)dnode_hash(dn->dn_objset, dn->dn_object) %
multilist_get_num_sublists(ml));
}
/*
* Instantiates the objset_t in-memory structure corresponding to the
* objset_phys_t that's pointed to by the specified blkptr_t.
*/
int
dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
objset_t **osp)
{
objset_t *os;
int i, err;
ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
ASSERT(!BP_IS_REDACTED(bp));
/*
* We need the pool config lock to get properties.
*/
ASSERT(ds == NULL || dsl_pool_config_held(ds->ds_dir->dd_pool));
/*
* The $ORIGIN dataset (if it exists) doesn't have an associated
* objset, so there's no reason to open it. The $ORIGIN dataset
* will not exist on pools older than SPA_VERSION_ORIGIN.
*/
if (ds != NULL && spa_get_dsl(spa) != NULL &&
spa_get_dsl(spa)->dp_origin_snap != NULL) {
ASSERT3P(ds->ds_dir, !=,
spa_get_dsl(spa)->dp_origin_snap->ds_dir);
}
os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
os->os_dsl_dataset = ds;
os->os_spa = spa;
os->os_rootbp = bp;
if (!BP_IS_HOLE(os->os_rootbp)) {
arc_flags_t aflags = ARC_FLAG_WAIT;
zbookmark_phys_t zb;
int size;
enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
if (DMU_OS_IS_L2CACHEABLE(os))
aflags |= ARC_FLAG_L2CACHE;
if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) {
ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF);
ASSERT(BP_IS_AUTHENTICATED(bp));
zio_flags |= ZIO_FLAG_RAW;
}
dprintf_bp(os->os_rootbp, "reading %s", "");
err = arc_read(NULL, spa, os->os_rootbp,
arc_getbuf_func, &os->os_phys_buf,
ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
if (err != 0) {
kmem_free(os, sizeof (objset_t));
/* convert checksum errors into IO errors */
if (err == ECKSUM)
err = SET_ERROR(EIO);
return (err);
}
if (spa_version(spa) < SPA_VERSION_USERSPACE)
size = OBJSET_PHYS_SIZE_V1;
else if (!spa_feature_is_enabled(spa,
SPA_FEATURE_PROJECT_QUOTA))
size = OBJSET_PHYS_SIZE_V2;
else
size = sizeof (objset_phys_t);
/* Increase the blocksize if we are permitted. */
if (arc_buf_size(os->os_phys_buf) < size) {
arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf,
ARC_BUFC_METADATA, size);
bzero(buf->b_data, size);
bcopy(os->os_phys_buf->b_data, buf->b_data,
arc_buf_size(os->os_phys_buf));
arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
os->os_phys_buf = buf;
}
os->os_phys = os->os_phys_buf->b_data;
os->os_flags = os->os_phys->os_flags;
} else {
int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1;
os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf,
ARC_BUFC_METADATA, size);
os->os_phys = os->os_phys_buf->b_data;
bzero(os->os_phys, size);
}
/*
* These properties will be filled in by the logic in zfs_get_zplprop()
* when they are queried for the first time.
*/
os->os_version = OBJSET_PROP_UNINITIALIZED;
os->os_normalization = OBJSET_PROP_UNINITIALIZED;
os->os_utf8only = OBJSET_PROP_UNINITIALIZED;
os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED;
/*
* Note: the changed_cb will be called once before the register
* func returns, thus changing the checksum/compression from the
* default (fletcher2/off). Snapshots don't need to know about
* checksum/compression/copies.
*/
if (ds != NULL) {
os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0);
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
primary_cache_changed_cb, os);
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
secondary_cache_changed_cb, os);
}
if (!ds->ds_is_snapshot) {
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_CHECKSUM),
checksum_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_COMPRESSION),
compression_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_COPIES),
copies_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_DEDUP),
dedup_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_LOGBIAS),
logbias_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_SYNC),
sync_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(
ZFS_PROP_REDUNDANT_METADATA),
redundant_metadata_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
recordsize_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_DNODESIZE),
dnodesize_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(
ZFS_PROP_SPECIAL_SMALL_BLOCKS),
smallblk_changed_cb, os);
}
}
if (err != 0) {
arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
kmem_free(os, sizeof (objset_t));
return (err);
}
} else {
/* It's the meta-objset. */
os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
os->os_compress = ZIO_COMPRESS_ON;
os->os_complevel = ZIO_COMPLEVEL_DEFAULT;
os->os_encrypted = B_FALSE;
os->os_copies = spa_max_replication(spa);
os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
os->os_dedup_verify = B_FALSE;
os->os_logbias = ZFS_LOGBIAS_LATENCY;
os->os_sync = ZFS_SYNC_STANDARD;
os->os_primary_cache = ZFS_CACHE_ALL;
os->os_secondary_cache = ZFS_CACHE_ALL;
os->os_dnodesize = DNODE_MIN_SIZE;
}
if (ds == NULL || !ds->ds_is_snapshot)
os->os_zil_header = os->os_phys->os_zil_header;
os->os_zil = zil_alloc(os, &os->os_zil_header);
for (i = 0; i < TXG_SIZE; i++) {
multilist_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[i]),
dnode_multilist_index_func);
}
list_create(&os->os_dnodes, sizeof (dnode_t),
offsetof(dnode_t, dn_link));
list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
offsetof(dmu_buf_impl_t, db_link));
list_link_init(&os->os_evicting_node);
mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
os->os_obj_next_percpu_len = boot_ncpus;
os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len *
sizeof (os->os_obj_next_percpu[0]), KM_SLEEP);
dnode_special_open(os, &os->os_phys->os_meta_dnode,
DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) {
dnode_special_open(os, &os->os_phys->os_userused_dnode,
DMU_USERUSED_OBJECT, &os->os_userused_dnode);
dnode_special_open(os, &os->os_phys->os_groupused_dnode,
DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf))
dnode_special_open(os,
&os->os_phys->os_projectused_dnode,
DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode);
}
mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL);
*osp = os;
return (0);
}
int
dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
{
int err = 0;
/*
* We need the pool_config lock to manipulate the dsl_dataset_t.
* Even if the dataset is long-held, we need the pool_config lock
* to open the objset, as it needs to get properties.
*/
ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
mutex_enter(&ds->ds_opening_lock);
if (ds->ds_objset == NULL) {
objset_t *os;
rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
ds, dsl_dataset_get_blkptr(ds), &os);
rrw_exit(&ds->ds_bp_rwlock, FTAG);
if (err == 0) {
mutex_enter(&ds->ds_lock);
ASSERT(ds->ds_objset == NULL);
ds->ds_objset = os;
mutex_exit(&ds->ds_lock);
}
}
*osp = ds->ds_objset;
mutex_exit(&ds->ds_opening_lock);
return (err);
}
/*
* Holds the pool while the objset is held. Therefore only one objset
* can be held at a time.
*/
int
dmu_objset_hold_flags(const char *name, boolean_t decrypt, void *tag,
objset_t **osp)
{
dsl_pool_t *dp;
dsl_dataset_t *ds;
int err;
ds_hold_flags_t flags;
flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
err = dsl_pool_hold(name, tag, &dp);
if (err != 0)
return (err);
err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds);
if (err != 0) {
dsl_pool_rele(dp, tag);
return (err);
}
err = dmu_objset_from_ds(ds, osp);
if (err != 0) {
dsl_dataset_rele(ds, tag);
dsl_pool_rele(dp, tag);
}
return (err);
}
int
dmu_objset_hold(const char *name, void *tag, objset_t **osp)
{
return (dmu_objset_hold_flags(name, B_FALSE, tag, osp));
}
static int
dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp)
{
int err;
err = dmu_objset_from_ds(ds, osp);
if (err != 0) {
return (err);
} else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
return (SET_ERROR(EINVAL));
} else if (!readonly && dsl_dataset_is_snapshot(ds)) {
return (SET_ERROR(EROFS));
} else if (!readonly && decrypt &&
dsl_dir_incompatible_encryption_version(ds->ds_dir)) {
return (SET_ERROR(EROFS));
}
/* if we are decrypting, we can now check MACs in os->os_phys_buf */
if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) {
zbookmark_phys_t zb;
SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT,
ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa,
&zb, B_FALSE);
if (err != 0)
return (err);
ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf));
}
return (0);
}
/*
* dsl_pool must not be held when this is called.
* Upon successful return, there will be a longhold on the dataset,
* and the dsl_pool will not be held.
*/
int
dmu_objset_own(const char *name, dmu_objset_type_t type,
boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp)
{
dsl_pool_t *dp;
dsl_dataset_t *ds;
int err;
ds_hold_flags_t flags;
flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
err = dsl_pool_hold(name, FTAG, &dp);
if (err != 0)
return (err);
err = dsl_dataset_own(dp, name, flags, tag, &ds);
if (err != 0) {
dsl_pool_rele(dp, FTAG);
return (err);
}
err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp);
if (err != 0) {
dsl_dataset_disown(ds, flags, tag);
dsl_pool_rele(dp, FTAG);
return (err);
}
/*
* User accounting requires the dataset to be decrypted and rw.
* We also don't begin user accounting during claiming to help
* speed up pool import times and to keep this txg reserved
* completely for recovery work.
*/
if (!readonly && !dp->dp_spa->spa_claiming &&
(ds->ds_dir->dd_crypto_obj == 0 || decrypt)) {
if (dmu_objset_userobjspace_upgradable(*osp) ||
dmu_objset_projectquota_upgradable(*osp)) {
dmu_objset_id_quota_upgrade(*osp);
} else if (dmu_objset_userused_enabled(*osp)) {
dmu_objset_userspace_upgrade(*osp);
}
}
dsl_pool_rele(dp, FTAG);
return (0);
}
int
dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp)
{
dsl_dataset_t *ds;
int err;
ds_hold_flags_t flags;
flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds);
if (err != 0)
return (err);
err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp);
if (err != 0) {
dsl_dataset_disown(ds, flags, tag);
return (err);
}
return (0);
}
void
dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, void *tag)
{
ds_hold_flags_t flags;
dsl_pool_t *dp = dmu_objset_pool(os);
flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag);
dsl_pool_rele(dp, tag);
}
void
dmu_objset_rele(objset_t *os, void *tag)
{
dmu_objset_rele_flags(os, B_FALSE, tag);
}
/*
* When we are called, os MUST refer to an objset associated with a dataset
* that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
* == tag. We will then release and reacquire ownership of the dataset while
* holding the pool config_rwlock to avoid intervening namespace or ownership
* changes may occur.
*
* This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
* release the hold on its dataset and acquire a new one on the dataset of the
* same name so that it can be partially torn down and reconstructed.
*/
void
dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds,
boolean_t decrypt, void *tag)
{
dsl_pool_t *dp;
char name[ZFS_MAX_DATASET_NAME_LEN];
ds_hold_flags_t flags;
flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
VERIFY3P(ds, !=, NULL);
VERIFY3P(ds->ds_owner, ==, tag);
VERIFY(dsl_dataset_long_held(ds));
dsl_dataset_name(ds, name);
dp = ds->ds_dir->dd_pool;
dsl_pool_config_enter(dp, FTAG);
dsl_dataset_disown(ds, flags, tag);
VERIFY0(dsl_dataset_own(dp, name, flags, tag, newds));
dsl_pool_config_exit(dp, FTAG);
}
void
dmu_objset_disown(objset_t *os, boolean_t decrypt, void *tag)
{
ds_hold_flags_t flags;
flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
/*
* Stop upgrading thread
*/
dmu_objset_upgrade_stop(os);
dsl_dataset_disown(os->os_dsl_dataset, flags, tag);
}
void
dmu_objset_evict_dbufs(objset_t *os)
{
dnode_t *dn_marker;
dnode_t *dn;
dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP);
mutex_enter(&os->os_lock);
dn = list_head(&os->os_dnodes);
while (dn != NULL) {
/*
* Skip dnodes without holds. We have to do this dance
* because dnode_add_ref() only works if there is already a
* hold. If the dnode has no holds, then it has no dbufs.
*/
if (dnode_add_ref(dn, FTAG)) {
list_insert_after(&os->os_dnodes, dn, dn_marker);
mutex_exit(&os->os_lock);
dnode_evict_dbufs(dn);
dnode_rele(dn, FTAG);
mutex_enter(&os->os_lock);
dn = list_next(&os->os_dnodes, dn_marker);
list_remove(&os->os_dnodes, dn_marker);
} else {
dn = list_next(&os->os_dnodes, dn);
}
}
mutex_exit(&os->os_lock);
kmem_free(dn_marker, sizeof (dnode_t));
if (DMU_USERUSED_DNODE(os) != NULL) {
if (DMU_PROJECTUSED_DNODE(os) != NULL)
dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os));
dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
}
dnode_evict_dbufs(DMU_META_DNODE(os));
}
/*
* Objset eviction processing is split into into two pieces.
* The first marks the objset as evicting, evicts any dbufs that
* have a refcount of zero, and then queues up the objset for the
* second phase of eviction. Once os->os_dnodes has been cleared by
* dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
* The second phase closes the special dnodes, dequeues the objset from
* the list of those undergoing eviction, and finally frees the objset.
*
* NOTE: Due to asynchronous eviction processing (invocation of
* dnode_buf_pageout()), it is possible for the meta dnode for the
* objset to have no holds even though os->os_dnodes is not empty.
*/
void
dmu_objset_evict(objset_t *os)
{
dsl_dataset_t *ds = os->os_dsl_dataset;
for (int t = 0; t < TXG_SIZE; t++)
ASSERT(!dmu_objset_is_dirty(os, t));
if (ds)
dsl_prop_unregister_all(ds, os);
if (os->os_sa)
sa_tear_down(os);
dmu_objset_evict_dbufs(os);
mutex_enter(&os->os_lock);
spa_evicting_os_register(os->os_spa, os);
if (list_is_empty(&os->os_dnodes)) {
mutex_exit(&os->os_lock);
dmu_objset_evict_done(os);
} else {
mutex_exit(&os->os_lock);
}
}
void
dmu_objset_evict_done(objset_t *os)
{
ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
dnode_special_close(&os->os_meta_dnode);
if (DMU_USERUSED_DNODE(os)) {
if (DMU_PROJECTUSED_DNODE(os))
dnode_special_close(&os->os_projectused_dnode);
dnode_special_close(&os->os_userused_dnode);
dnode_special_close(&os->os_groupused_dnode);
}
zil_free(os->os_zil);
arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
/*
* This is a barrier to prevent the objset from going away in
* dnode_move() until we can safely ensure that the objset is still in
* use. We consider the objset valid before the barrier and invalid
* after the barrier.
*/
rw_enter(&os_lock, RW_READER);
rw_exit(&os_lock);
kmem_free(os->os_obj_next_percpu,
os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0]));
mutex_destroy(&os->os_lock);
mutex_destroy(&os->os_userused_lock);
mutex_destroy(&os->os_obj_lock);
mutex_destroy(&os->os_user_ptr_lock);
mutex_destroy(&os->os_upgrade_lock);
for (int i = 0; i < TXG_SIZE; i++)
multilist_destroy(&os->os_dirty_dnodes[i]);
spa_evicting_os_deregister(os->os_spa, os);
kmem_free(os, sizeof (objset_t));
}
inode_timespec_t
dmu_objset_snap_cmtime(objset_t *os)
{
return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
}
objset_t *
dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx)
{
objset_t *os;
dnode_t *mdn;
ASSERT(dmu_tx_is_syncing(tx));
if (blksz == 0)
blksz = DNODE_BLOCK_SIZE;
if (ibs == 0)
ibs = DN_MAX_INDBLKSHIFT;
if (ds != NULL)
VERIFY0(dmu_objset_from_ds(ds, &os));
else
VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
mdn = DMU_META_DNODE(os);
dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0,
DNODE_MIN_SLOTS, tx);
/*
* We don't want to have to increase the meta-dnode's nlevels
* later, because then we could do it in quiescing context while
* we are also accessing it in open context.
*
* This precaution is not necessary for the MOS (ds == NULL),
* because the MOS is only updated in syncing context.
* This is most fortunate: the MOS is the only objset that
* needs to be synced multiple times as spa_sync() iterates
* to convergence, so minimizing its dn_nlevels matters.
*/
if (ds != NULL) {
if (levels == 0) {
levels = 1;
/*
* Determine the number of levels necessary for the
* meta-dnode to contain DN_MAX_OBJECT dnodes. Note
* that in order to ensure that we do not overflow
* 64 bits, there has to be a nlevels that gives us a
* number of blocks > DN_MAX_OBJECT but < 2^64.
* Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)
* (10) must be less than (64 - log2(DN_MAX_OBJECT))
* (16).
*/
while ((uint64_t)mdn->dn_nblkptr <<
(mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) *
(mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
DN_MAX_OBJECT)
levels++;
}
mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
mdn->dn_nlevels = levels;
}
ASSERT(type != DMU_OST_NONE);
ASSERT(type != DMU_OST_ANY);
ASSERT(type < DMU_OST_NUMTYPES);
os->os_phys->os_type = type;
/*
* Enable user accounting if it is enabled and this is not an
* encrypted receive.
*/
if (dmu_objset_userused_enabled(os) &&
(!os->os_encrypted || !dmu_objset_is_receiving(os))) {
os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
if (dmu_objset_userobjused_enabled(os)) {
ds->ds_feature_activation[
SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE;
os->os_phys->os_flags |=
OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE;
}
if (dmu_objset_projectquota_enabled(os)) {
ds->ds_feature_activation[
SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE;
os->os_phys->os_flags |=
OBJSET_FLAG_PROJECTQUOTA_COMPLETE;
}
os->os_flags = os->os_phys->os_flags;
}
dsl_dataset_dirty(ds, tx);
return (os);
}
/* called from dsl for meta-objset */
objset_t *
dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
dmu_objset_type_t type, dmu_tx_t *tx)
{
return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx));
}
typedef struct dmu_objset_create_arg {
const char *doca_name;
cred_t *doca_cred;
proc_t *doca_proc;
void (*doca_userfunc)(objset_t *os, void *arg,
cred_t *cr, dmu_tx_t *tx);
void *doca_userarg;
dmu_objset_type_t doca_type;
uint64_t doca_flags;
dsl_crypto_params_t *doca_dcp;
} dmu_objset_create_arg_t;
/*ARGSUSED*/
static int
dmu_objset_create_check(void *arg, dmu_tx_t *tx)
{
dmu_objset_create_arg_t *doca = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dir_t *pdd;
dsl_dataset_t *parentds;
objset_t *parentos;
const char *tail;
int error;
if (strchr(doca->doca_name, '@') != NULL)
return (SET_ERROR(EINVAL));
if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN)
return (SET_ERROR(ENAMETOOLONG));
if (dataset_nestcheck(doca->doca_name) != 0)
return (SET_ERROR(ENAMETOOLONG));
error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
if (error != 0)
return (error);
if (tail == NULL) {
dsl_dir_rele(pdd, FTAG);
return (SET_ERROR(EEXIST));
}
error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL);
if (error != 0) {
dsl_dir_rele(pdd, FTAG);
return (error);
}
error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
doca->doca_cred, doca->doca_proc);
if (error != 0) {
dsl_dir_rele(pdd, FTAG);
return (error);
}
/* can't create below anything but filesystems (eg. no ZVOLs) */
error = dsl_dataset_hold_obj(pdd->dd_pool,
dsl_dir_phys(pdd)->dd_head_dataset_obj, FTAG, &parentds);
if (error != 0) {
dsl_dir_rele(pdd, FTAG);
return (error);
}
error = dmu_objset_from_ds(parentds, &parentos);
if (error != 0) {
dsl_dataset_rele(parentds, FTAG);
dsl_dir_rele(pdd, FTAG);
return (error);
}
if (dmu_objset_type(parentos) != DMU_OST_ZFS) {
dsl_dataset_rele(parentds, FTAG);
dsl_dir_rele(pdd, FTAG);
return (SET_ERROR(ZFS_ERR_WRONG_PARENT));
}
dsl_dataset_rele(parentds, FTAG);
dsl_dir_rele(pdd, FTAG);
return (error);
}
static void
dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
{
dmu_objset_create_arg_t *doca = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
spa_t *spa = dp->dp_spa;
dsl_dir_t *pdd;
const char *tail;
dsl_dataset_t *ds;
uint64_t obj;
blkptr_t *bp;
objset_t *os;
zio_t *rzio;
VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
doca->doca_cred, doca->doca_dcp, tx);
VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj,
DS_HOLD_FLAG_DECRYPT, FTAG, &ds));
rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
bp = dsl_dataset_get_blkptr(ds);
os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx);
rrw_exit(&ds->ds_bp_rwlock, FTAG);
if (doca->doca_userfunc != NULL) {
doca->doca_userfunc(os, doca->doca_userarg,
doca->doca_cred, tx);
}
/*
* The doca_userfunc() may write out some data that needs to be
* encrypted if the dataset is encrypted (specifically the root
* directory). This data must be written out before the encryption
* key mapping is removed by dsl_dataset_rele_flags(). Force the
* I/O to occur immediately by invoking the relevant sections of
* dsl_pool_sync().
*/
if (os->os_encrypted) {
dsl_dataset_t *tmpds = NULL;
boolean_t need_sync_done = B_FALSE;
mutex_enter(&ds->ds_lock);
ds->ds_owner = FTAG;
mutex_exit(&ds->ds_lock);
rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds,
tx->tx_txg);
if (tmpds != NULL) {
dsl_dataset_sync(ds, rzio, tx);
need_sync_done = B_TRUE;
}
VERIFY0(zio_wait(rzio));
dmu_objset_sync_done(os, tx);
taskq_wait(dp->dp_sync_taskq);
if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
ASSERT3P(ds->ds_key_mapping, !=, NULL);
key_mapping_rele(spa, ds->ds_key_mapping, ds);
}
rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds,
tx->tx_txg);
if (tmpds != NULL) {
dmu_buf_rele(ds->ds_dbuf, ds);
dsl_dataset_sync(ds, rzio, tx);
}
VERIFY0(zio_wait(rzio));
if (need_sync_done) {
ASSERT3P(ds->ds_key_mapping, !=, NULL);
key_mapping_rele(spa, ds->ds_key_mapping, ds);
dsl_dataset_sync_done(ds, tx);
}
mutex_enter(&ds->ds_lock);
ds->ds_owner = NULL;
mutex_exit(&ds->ds_lock);
}
spa_history_log_internal_ds(ds, "create", tx, " ");
dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG);
dsl_dir_rele(pdd, FTAG);
}
int
dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg)
{
dmu_objset_create_arg_t doca;
dsl_crypto_params_t tmp_dcp = { 0 };
doca.doca_name = name;
doca.doca_cred = CRED();
doca.doca_proc = curproc;
doca.doca_flags = flags;
doca.doca_userfunc = func;
doca.doca_userarg = arg;
doca.doca_type = type;
/*
* Some callers (mostly for testing) do not provide a dcp on their
* own but various code inside the sync task will require it to be
* allocated. Rather than adding NULL checks throughout this code
* or adding dummy dcp's to all of the callers we simply create a
* dummy one here and use that. This zero dcp will have the same
* effect as asking for inheritance of all encryption params.
*/
doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp;
int rv = dsl_sync_task(name,
dmu_objset_create_check, dmu_objset_create_sync, &doca,
6, ZFS_SPACE_CHECK_NORMAL);
if (rv == 0)
zvol_create_minor(name);
return (rv);
}
typedef struct dmu_objset_clone_arg {
const char *doca_clone;
const char *doca_origin;
cred_t *doca_cred;
proc_t *doca_proc;
} dmu_objset_clone_arg_t;
/*ARGSUSED*/
static int
dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
{
dmu_objset_clone_arg_t *doca = arg;
dsl_dir_t *pdd;
const char *tail;
int error;
dsl_dataset_t *origin;
dsl_pool_t *dp = dmu_tx_pool(tx);
if (strchr(doca->doca_clone, '@') != NULL)
return (SET_ERROR(EINVAL));
if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN)
return (SET_ERROR(ENAMETOOLONG));
error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
if (error != 0)
return (error);
if (tail == NULL) {
dsl_dir_rele(pdd, FTAG);
return (SET_ERROR(EEXIST));
}
error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
doca->doca_cred, doca->doca_proc);
if (error != 0) {
dsl_dir_rele(pdd, FTAG);
return (SET_ERROR(EDQUOT));
}
error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
if (error != 0) {
dsl_dir_rele(pdd, FTAG);
return (error);
}
/* You can only clone snapshots, not the head datasets. */
if (!origin->ds_is_snapshot) {
dsl_dataset_rele(origin, FTAG);
dsl_dir_rele(pdd, FTAG);
return (SET_ERROR(EINVAL));
}
dsl_dataset_rele(origin, FTAG);
dsl_dir_rele(pdd, FTAG);
return (0);
}
static void
dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
{
dmu_objset_clone_arg_t *doca = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dir_t *pdd;
const char *tail;
dsl_dataset_t *origin, *ds;
uint64_t obj;
char namebuf[ZFS_MAX_DATASET_NAME_LEN];
VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
doca->doca_cred, NULL, tx);
VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
dsl_dataset_name(origin, namebuf);
spa_history_log_internal_ds(ds, "clone", tx,
"origin=%s (%llu)", namebuf, (u_longlong_t)origin->ds_object);
dsl_dataset_rele(ds, FTAG);
dsl_dataset_rele(origin, FTAG);
dsl_dir_rele(pdd, FTAG);
}
int
dmu_objset_clone(const char *clone, const char *origin)
{
dmu_objset_clone_arg_t doca;
doca.doca_clone = clone;
doca.doca_origin = origin;
doca.doca_cred = CRED();
doca.doca_proc = curproc;
int rv = dsl_sync_task(clone,
dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
6, ZFS_SPACE_CHECK_NORMAL);
if (rv == 0)
zvol_create_minor(clone);
return (rv);
}
int
dmu_objset_snapshot_one(const char *fsname, const char *snapname)
{
int err;
char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
nvlist_t *snaps = fnvlist_alloc();
fnvlist_add_boolean(snaps, longsnap);
kmem_strfree(longsnap);
err = dsl_dataset_snapshot(snaps, NULL, NULL);
fnvlist_free(snaps);
return (err);
}
static void
dmu_objset_upgrade_task_cb(void *data)
{
objset_t *os = data;
mutex_enter(&os->os_upgrade_lock);
os->os_upgrade_status = EINTR;
if (!os->os_upgrade_exit) {
int status;
mutex_exit(&os->os_upgrade_lock);
status = os->os_upgrade_cb(os);
mutex_enter(&os->os_upgrade_lock);
os->os_upgrade_status = status;
}
os->os_upgrade_exit = B_TRUE;
os->os_upgrade_id = 0;
mutex_exit(&os->os_upgrade_lock);
dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
}
static void
dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb)
{
if (os->os_upgrade_id != 0)
return;
ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag);
mutex_enter(&os->os_upgrade_lock);
if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) {
os->os_upgrade_exit = B_FALSE;
os->os_upgrade_cb = cb;
os->os_upgrade_id = taskq_dispatch(
os->os_spa->spa_upgrade_taskq,
dmu_objset_upgrade_task_cb, os, TQ_SLEEP);
if (os->os_upgrade_id == TASKQID_INVALID) {
dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
os->os_upgrade_status = ENOMEM;
}
} else {
dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
}
mutex_exit(&os->os_upgrade_lock);
}
static void
dmu_objset_upgrade_stop(objset_t *os)
{
mutex_enter(&os->os_upgrade_lock);
os->os_upgrade_exit = B_TRUE;
if (os->os_upgrade_id != 0) {
taskqid_t id = os->os_upgrade_id;
os->os_upgrade_id = 0;
mutex_exit(&os->os_upgrade_lock);
if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id)) == 0) {
dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
}
txg_wait_synced(os->os_spa->spa_dsl_pool, 0);
} else {
mutex_exit(&os->os_upgrade_lock);
}
}
static void
dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx)
{
dnode_t *dn;
while ((dn = multilist_sublist_head(list)) != NULL) {
ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
ASSERT(dn->dn_dbuf->db_data_pending);
/*
* Initialize dn_zio outside dnode_sync() because the
* meta-dnode needs to set it outside dnode_sync().
*/
dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
ASSERT(dn->dn_zio);
ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
multilist_sublist_remove(list, dn);
/*
* See the comment above dnode_rele_task() for an explanation
* of why this dnode hold is always needed (even when not
* doing user accounting).
*/
multilist_t *newlist = &dn->dn_objset->os_synced_dnodes;
(void) dnode_add_ref(dn, newlist);
multilist_insert(newlist, dn);
dnode_sync(dn, tx);
}
}
/* ARGSUSED */
static void
dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
{
blkptr_t *bp = zio->io_bp;
objset_t *os = arg;
dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
uint64_t fill = 0;
ASSERT(!BP_IS_EMBEDDED(bp));
ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
ASSERT0(BP_GET_LEVEL(bp));
/*
* Update rootbp fill count: it should be the number of objects
* allocated in the object set (not counting the "special"
* objects that are stored in the objset_phys_t -- the meta
* dnode and user/group/project accounting objects).
*/
for (int i = 0; i < dnp->dn_nblkptr; i++)
fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
BP_SET_FILL(bp, fill);
if (os->os_dsl_dataset != NULL)
rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG);
*os->os_rootbp = *bp;
if (os->os_dsl_dataset != NULL)
rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG);
}
/* ARGSUSED */
static void
dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
{
blkptr_t *bp = zio->io_bp;
blkptr_t *bp_orig = &zio->io_bp_orig;
objset_t *os = arg;
if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
ASSERT(BP_EQUAL(bp, bp_orig));
} else {
dsl_dataset_t *ds = os->os_dsl_dataset;
dmu_tx_t *tx = os->os_synctx;
(void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
dsl_dataset_block_born(ds, bp, tx);
}
kmem_free(bp, sizeof (*bp));
}
typedef struct sync_dnodes_arg {
multilist_t *sda_list;
int sda_sublist_idx;
multilist_t *sda_newlist;
dmu_tx_t *sda_tx;
} sync_dnodes_arg_t;
static void
sync_dnodes_task(void *arg)
{
sync_dnodes_arg_t *sda = arg;
multilist_sublist_t *ms =
multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx);
dmu_objset_sync_dnodes(ms, sda->sda_tx);
multilist_sublist_unlock(ms);
kmem_free(sda, sizeof (*sda));
}
/* called from dsl */
void
dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
{
int txgoff;
zbookmark_phys_t zb;
zio_prop_t zp;
zio_t *zio;
list_t *list;
dbuf_dirty_record_t *dr;
int num_sublists;
multilist_t *ml;
blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP);
*blkptr_copy = *os->os_rootbp;
dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", (u_longlong_t)tx->tx_txg);
ASSERT(dmu_tx_is_syncing(tx));
/* XXX the write_done callback should really give us the tx... */
os->os_synctx = tx;
if (os->os_dsl_dataset == NULL) {
/*
* This is the MOS. If we have upgraded,
* spa_max_replication() could change, so reset
* os_copies here.
*/
os->os_copies = spa_max_replication(os->os_spa);
}
/*
* Create the root block IO
*/
SET_BOOKMARK(&zb, os->os_dsl_dataset ?
os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
arc_release(os->os_phys_buf, &os->os_phys_buf);
dmu_write_policy(os, NULL, 0, 0, &zp);
/*
* If we are either claiming the ZIL or doing a raw receive, write
* out the os_phys_buf raw. Neither of these actions will effect the
* MAC at this point.
*/
if (os->os_raw_receive ||
os->os_next_write_raw[tx->tx_txg & TXG_MASK]) {
ASSERT(os->os_encrypted);
arc_convert_to_raw(os->os_phys_buf,
os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER,
DMU_OT_OBJSET, NULL, NULL, NULL);
}
zio = arc_write(pio, os->os_spa, tx->tx_txg,
blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
&zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done,
os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
/*
* Sync special dnodes - the parent IO for the sync is the root block
*/
DMU_META_DNODE(os)->dn_zio = zio;
dnode_sync(DMU_META_DNODE(os), tx);
os->os_phys->os_flags = os->os_flags;
if (DMU_USERUSED_DNODE(os) &&
DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
DMU_USERUSED_DNODE(os)->dn_zio = zio;
dnode_sync(DMU_USERUSED_DNODE(os), tx);
DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
}
if (DMU_PROJECTUSED_DNODE(os) &&
DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
DMU_PROJECTUSED_DNODE(os)->dn_zio = zio;
dnode_sync(DMU_PROJECTUSED_DNODE(os), tx);
}
txgoff = tx->tx_txg & TXG_MASK;
/*
* We must create the list here because it uses the
* dn_dirty_link[] of this txg. But it may already
* exist because we call dsl_dataset_sync() twice per txg.
*/
if (os->os_synced_dnodes.ml_sublists == NULL) {
multilist_create(&os->os_synced_dnodes, sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[txgoff]),
dnode_multilist_index_func);
} else {
ASSERT3U(os->os_synced_dnodes.ml_offset, ==,
offsetof(dnode_t, dn_dirty_link[txgoff]));
}
ml = &os->os_dirty_dnodes[txgoff];
num_sublists = multilist_get_num_sublists(ml);
for (int i = 0; i < num_sublists; i++) {
if (multilist_sublist_is_empty_idx(ml, i))
continue;
sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP);
sda->sda_list = ml;
sda->sda_sublist_idx = i;
sda->sda_tx = tx;
(void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
sync_dnodes_task, sda, 0);
/* callback frees sda */
}
taskq_wait(dmu_objset_pool(os)->dp_sync_taskq);
list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
while ((dr = list_head(list)) != NULL) {
ASSERT0(dr->dr_dbuf->db_level);
list_remove(list, dr);
zio_nowait(dr->dr_zio);
}
/* Enable dnode backfill if enough objects have been freed. */
if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) {
os->os_rescan_dnodes = B_TRUE;
os->os_freed_dnodes = 0;
}
/*
* Free intent log blocks up to this tx.
*/
zil_sync(os->os_zil, tx);
os->os_phys->os_zil_header = os->os_zil_header;
zio_nowait(zio);
}
boolean_t
dmu_objset_is_dirty(objset_t *os, uint64_t txg)
{
return (!multilist_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]));
}
static file_info_cb_t *file_cbs[DMU_OST_NUMTYPES];
void
dmu_objset_register_type(dmu_objset_type_t ost, file_info_cb_t *cb)
{
file_cbs[ost] = cb;
}
int
dmu_get_file_info(objset_t *os, dmu_object_type_t bonustype, const void *data,
zfs_file_info_t *zfi)
{
file_info_cb_t *cb = file_cbs[os->os_phys->os_type];
if (cb == NULL)
return (EINVAL);
return (cb(bonustype, data, zfi));
}
boolean_t
dmu_objset_userused_enabled(objset_t *os)
{
return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
file_cbs[os->os_phys->os_type] != NULL &&
DMU_USERUSED_DNODE(os) != NULL);
}
boolean_t
dmu_objset_userobjused_enabled(objset_t *os)
{
return (dmu_objset_userused_enabled(os) &&
spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING));
}
boolean_t
dmu_objset_projectquota_enabled(objset_t *os)
{
return (file_cbs[os->os_phys->os_type] != NULL &&
DMU_PROJECTUSED_DNODE(os) != NULL &&
spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA));
}
typedef struct userquota_node {
/* must be in the first filed, see userquota_update_cache() */
char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN];
int64_t uqn_delta;
avl_node_t uqn_node;
} userquota_node_t;
typedef struct userquota_cache {
avl_tree_t uqc_user_deltas;
avl_tree_t uqc_group_deltas;
avl_tree_t uqc_project_deltas;
} userquota_cache_t;
static int
userquota_compare(const void *l, const void *r)
{
const userquota_node_t *luqn = l;
const userquota_node_t *ruqn = r;
int rv;
/*
* NB: can only access uqn_id because userquota_update_cache() doesn't
* pass in an entire userquota_node_t.
*/
rv = strcmp(luqn->uqn_id, ruqn->uqn_id);
return (TREE_ISIGN(rv));
}
static void
do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx)
{
void *cookie;
userquota_node_t *uqn;
ASSERT(dmu_tx_is_syncing(tx));
cookie = NULL;
while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas,
&cookie)) != NULL) {
/*
* os_userused_lock protects against concurrent calls to
* zap_increment_int(). It's needed because zap_increment_int()
* is not thread-safe (i.e. not atomic).
*/
mutex_enter(&os->os_userused_lock);
VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT,
uqn->uqn_id, uqn->uqn_delta, tx));
mutex_exit(&os->os_userused_lock);
kmem_free(uqn, sizeof (*uqn));
}
avl_destroy(&cache->uqc_user_deltas);
cookie = NULL;
while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas,
&cookie)) != NULL) {
mutex_enter(&os->os_userused_lock);
VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT,
uqn->uqn_id, uqn->uqn_delta, tx));
mutex_exit(&os->os_userused_lock);
kmem_free(uqn, sizeof (*uqn));
}
avl_destroy(&cache->uqc_group_deltas);
if (dmu_objset_projectquota_enabled(os)) {
cookie = NULL;
while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas,
&cookie)) != NULL) {
mutex_enter(&os->os_userused_lock);
VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT,
uqn->uqn_id, uqn->uqn_delta, tx));
mutex_exit(&os->os_userused_lock);
kmem_free(uqn, sizeof (*uqn));
}
avl_destroy(&cache->uqc_project_deltas);
}
}
static void
userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta)
{
userquota_node_t *uqn;
avl_index_t idx;
ASSERT(strlen(id) < sizeof (uqn->uqn_id));
/*
* Use id directly for searching because uqn_id is the first field of
* userquota_node_t and fields after uqn_id won't be accessed in
* avl_find().
*/
uqn = avl_find(avl, (const void *)id, &idx);
if (uqn == NULL) {
uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP);
strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id));
avl_insert(avl, uqn, idx);
}
uqn->uqn_delta += delta;
}
static void
do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used,
uint64_t flags, uint64_t user, uint64_t group, uint64_t project,
boolean_t subtract)
{
if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) {
int64_t delta = DNODE_MIN_SIZE + used;
char name[20];
if (subtract)
delta = -delta;
(void) snprintf(name, sizeof (name), "%llx", (longlong_t)user);
userquota_update_cache(&cache->uqc_user_deltas, name, delta);
(void) snprintf(name, sizeof (name), "%llx", (longlong_t)group);
userquota_update_cache(&cache->uqc_group_deltas, name, delta);
if (dmu_objset_projectquota_enabled(os)) {
(void) snprintf(name, sizeof (name), "%llx",
(longlong_t)project);
userquota_update_cache(&cache->uqc_project_deltas,
name, delta);
}
}
}
static void
do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags,
uint64_t user, uint64_t group, uint64_t project, boolean_t subtract)
{
if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) {
char name[20 + DMU_OBJACCT_PREFIX_LEN];
int delta = subtract ? -1 : 1;
(void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx",
(longlong_t)user);
userquota_update_cache(&cache->uqc_user_deltas, name, delta);
(void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx",
(longlong_t)group);
userquota_update_cache(&cache->uqc_group_deltas, name, delta);
if (dmu_objset_projectquota_enabled(os)) {
(void) snprintf(name, sizeof (name),
DMU_OBJACCT_PREFIX "%llx", (longlong_t)project);
userquota_update_cache(&cache->uqc_project_deltas,
name, delta);
}
}
}
typedef struct userquota_updates_arg {
objset_t *uua_os;
int uua_sublist_idx;
dmu_tx_t *uua_tx;
} userquota_updates_arg_t;
static void
userquota_updates_task(void *arg)
{
userquota_updates_arg_t *uua = arg;
objset_t *os = uua->uua_os;
dmu_tx_t *tx = uua->uua_tx;
dnode_t *dn;
userquota_cache_t cache = { { 0 } };
multilist_sublist_t *list =
multilist_sublist_lock(&os->os_synced_dnodes, uua->uua_sublist_idx);
ASSERT(multilist_sublist_head(list) == NULL ||
dmu_objset_userused_enabled(os));
avl_create(&cache.uqc_user_deltas, userquota_compare,
sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
avl_create(&cache.uqc_group_deltas, userquota_compare,
sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
if (dmu_objset_projectquota_enabled(os))
avl_create(&cache.uqc_project_deltas, userquota_compare,
sizeof (userquota_node_t), offsetof(userquota_node_t,
uqn_node));
while ((dn = multilist_sublist_head(list)) != NULL) {
int flags;
ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
dn->dn_phys->dn_flags &
DNODE_FLAG_USERUSED_ACCOUNTED);
flags = dn->dn_id_flags;
ASSERT(flags);
if (flags & DN_ID_OLD_EXIST) {
do_userquota_update(os, &cache, dn->dn_oldused,
dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid,
dn->dn_oldprojid, B_TRUE);
do_userobjquota_update(os, &cache, dn->dn_oldflags,
dn->dn_olduid, dn->dn_oldgid,
dn->dn_oldprojid, B_TRUE);
}
if (flags & DN_ID_NEW_EXIST) {
do_userquota_update(os, &cache,
DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags,
dn->dn_newuid, dn->dn_newgid,
dn->dn_newprojid, B_FALSE);
do_userobjquota_update(os, &cache,
dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid,
dn->dn_newprojid, B_FALSE);
}
mutex_enter(&dn->dn_mtx);
dn->dn_oldused = 0;
dn->dn_oldflags = 0;
if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
dn->dn_olduid = dn->dn_newuid;
dn->dn_oldgid = dn->dn_newgid;
dn->dn_oldprojid = dn->dn_newprojid;
dn->dn_id_flags |= DN_ID_OLD_EXIST;
if (dn->dn_bonuslen == 0)
dn->dn_id_flags |= DN_ID_CHKED_SPILL;
else
dn->dn_id_flags |= DN_ID_CHKED_BONUS;
}
dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
mutex_exit(&dn->dn_mtx);
multilist_sublist_remove(list, dn);
dnode_rele(dn, &os->os_synced_dnodes);
}
do_userquota_cacheflush(os, &cache, tx);
multilist_sublist_unlock(list);
kmem_free(uua, sizeof (*uua));
}
/*
* Release dnode holds from dmu_objset_sync_dnodes(). When the dnode is being
* synced (i.e. we have issued the zio's for blocks in the dnode), it can't be
* evicted because the block containing the dnode can't be evicted until it is
* written out. However, this hold is necessary to prevent the dnode_t from
* being moved (via dnode_move()) while it's still referenced by
* dbuf_dirty_record_t:dr_dnode. And dr_dnode is needed for
* dirty_lightweight_leaf-type dirty records.
*
* If we are doing user-object accounting, the dnode_rele() happens from
* userquota_updates_task() instead.
*/
static void
dnode_rele_task(void *arg)
{
userquota_updates_arg_t *uua = arg;
objset_t *os = uua->uua_os;
multilist_sublist_t *list =
multilist_sublist_lock(&os->os_synced_dnodes, uua->uua_sublist_idx);
dnode_t *dn;
while ((dn = multilist_sublist_head(list)) != NULL) {
multilist_sublist_remove(list, dn);
dnode_rele(dn, &os->os_synced_dnodes);
}
multilist_sublist_unlock(list);
kmem_free(uua, sizeof (*uua));
}
/*
* Return TRUE if userquota updates are needed.
*/
static boolean_t
dmu_objset_do_userquota_updates_prep(objset_t *os, dmu_tx_t *tx)
{
if (!dmu_objset_userused_enabled(os))
return (B_FALSE);
/*
* If this is a raw receive just return and handle accounting
* later when we have the keys loaded. We also don't do user
* accounting during claiming since the datasets are not owned
* for the duration of claiming and this txg should only be
* used for recovery.
*/
if (os->os_encrypted && dmu_objset_is_receiving(os))
return (B_FALSE);
if (tx->tx_txg <= os->os_spa->spa_claim_max_txg)
return (B_FALSE);
/* Allocate the user/group/project used objects if necessary. */
if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
VERIFY0(zap_create_claim(os,
DMU_USERUSED_OBJECT,
DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
VERIFY0(zap_create_claim(os,
DMU_GROUPUSED_OBJECT,
DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
}
if (dmu_objset_projectquota_enabled(os) &&
DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT,
DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
}
return (B_TRUE);
}
/*
* Dispatch taskq tasks to dp_sync_taskq to update the user accounting, and
* also release the holds on the dnodes from dmu_objset_sync_dnodes().
* The caller must taskq_wait(dp_sync_taskq).
*/
void
dmu_objset_sync_done(objset_t *os, dmu_tx_t *tx)
{
boolean_t need_userquota = dmu_objset_do_userquota_updates_prep(os, tx);
int num_sublists = multilist_get_num_sublists(&os->os_synced_dnodes);
for (int i = 0; i < num_sublists; i++) {
userquota_updates_arg_t *uua =
kmem_alloc(sizeof (*uua), KM_SLEEP);
uua->uua_os = os;
uua->uua_sublist_idx = i;
uua->uua_tx = tx;
/*
* If we don't need to update userquotas, use
* dnode_rele_task() to call dnode_rele()
*/
(void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
need_userquota ? userquota_updates_task : dnode_rele_task,
uua, 0);
/* callback frees uua */
}
}
/*
* Returns a pointer to data to find uid/gid from
*
* If a dirty record for transaction group that is syncing can't
* be found then NULL is returned. In the NULL case it is assumed
* the uid/gid aren't changing.
*/
static void *
dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
{
dbuf_dirty_record_t *dr;
void *data;
if (db->db_dirtycnt == 0)
return (db->db.db_data); /* Nothing is changing */
dr = dbuf_find_dirty_eq(db, tx->tx_txg);
if (dr == NULL) {
data = NULL;
} else {
if (dr->dr_dnode->dn_bonuslen == 0 &&
dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
data = dr->dt.dl.dr_data->b_data;
else
data = dr->dt.dl.dr_data;
}
return (data);
}
void
dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
{
objset_t *os = dn->dn_objset;
void *data = NULL;
dmu_buf_impl_t *db = NULL;
int flags = dn->dn_id_flags;
int error;
boolean_t have_spill = B_FALSE;
if (!dmu_objset_userused_enabled(dn->dn_objset))
return;
/*
* Raw receives introduce a problem with user accounting. Raw
* receives cannot update the user accounting info because the
* user ids and the sizes are encrypted. To guarantee that we
* never end up with bad user accounting, we simply disable it
* during raw receives. We also disable this for normal receives
* so that an incremental raw receive may be done on top of an
* existing non-raw receive.
*/
if (os->os_encrypted && dmu_objset_is_receiving(os))
return;
if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
DN_ID_CHKED_SPILL)))
return;
if (before && dn->dn_bonuslen != 0)
data = DN_BONUS(dn->dn_phys);
else if (!before && dn->dn_bonuslen != 0) {
if (dn->dn_bonus) {
db = dn->dn_bonus;
mutex_enter(&db->db_mtx);
data = dmu_objset_userquota_find_data(db, tx);
} else {
data = DN_BONUS(dn->dn_phys);
}
} else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
int rf = 0;
if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
rf |= DB_RF_HAVESTRUCT;
error = dmu_spill_hold_by_dnode(dn,
rf | DB_RF_MUST_SUCCEED,
FTAG, (dmu_buf_t **)&db);
ASSERT(error == 0);
mutex_enter(&db->db_mtx);
data = (before) ? db->db.db_data :
dmu_objset_userquota_find_data(db, tx);
have_spill = B_TRUE;
} else {
mutex_enter(&dn->dn_mtx);
dn->dn_id_flags |= DN_ID_CHKED_BONUS;
mutex_exit(&dn->dn_mtx);
return;
}
/*
* Must always call the callback in case the object
* type has changed and that type isn't an object type to track
*/
zfs_file_info_t zfi;
error = file_cbs[os->os_phys->os_type](dn->dn_bonustype, data, &zfi);
if (before) {
ASSERT(data);
dn->dn_olduid = zfi.zfi_user;
dn->dn_oldgid = zfi.zfi_group;
dn->dn_oldprojid = zfi.zfi_project;
} else if (data) {
dn->dn_newuid = zfi.zfi_user;
dn->dn_newgid = zfi.zfi_group;
dn->dn_newprojid = zfi.zfi_project;
}
/*
* Preserve existing uid/gid when the callback can't determine
* what the new uid/gid are and the callback returned EEXIST.
* The EEXIST error tells us to just use the existing uid/gid.
* If we don't know what the old values are then just assign
* them to 0, since that is a new file being created.
*/
if (!before && data == NULL && error == EEXIST) {
if (flags & DN_ID_OLD_EXIST) {
dn->dn_newuid = dn->dn_olduid;
dn->dn_newgid = dn->dn_oldgid;
dn->dn_newprojid = dn->dn_oldprojid;
} else {
dn->dn_newuid = 0;
dn->dn_newgid = 0;
dn->dn_newprojid = ZFS_DEFAULT_PROJID;
}
error = 0;
}
if (db)
mutex_exit(&db->db_mtx);
mutex_enter(&dn->dn_mtx);
if (error == 0 && before)
dn->dn_id_flags |= DN_ID_OLD_EXIST;
if (error == 0 && !before)
dn->dn_id_flags |= DN_ID_NEW_EXIST;
if (have_spill) {
dn->dn_id_flags |= DN_ID_CHKED_SPILL;
} else {
dn->dn_id_flags |= DN_ID_CHKED_BONUS;
}
mutex_exit(&dn->dn_mtx);
if (have_spill)
dmu_buf_rele((dmu_buf_t *)db, FTAG);
}
boolean_t
dmu_objset_userspace_present(objset_t *os)
{
return (os->os_phys->os_flags &
OBJSET_FLAG_USERACCOUNTING_COMPLETE);
}
boolean_t
dmu_objset_userobjspace_present(objset_t *os)
{
return (os->os_phys->os_flags &
OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE);
}
boolean_t
dmu_objset_projectquota_present(objset_t *os)
{
return (os->os_phys->os_flags &
OBJSET_FLAG_PROJECTQUOTA_COMPLETE);
}
static int
dmu_objset_space_upgrade(objset_t *os)
{
uint64_t obj;
int err = 0;
/*
* We simply need to mark every object dirty, so that it will be
* synced out and now accounted. If this is called
* concurrently, or if we already did some work before crashing,
* that's fine, since we track each object's accounted state
* independently.
*/
for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
dmu_tx_t *tx;
dmu_buf_t *db;
int objerr;
mutex_enter(&os->os_upgrade_lock);
if (os->os_upgrade_exit)
err = SET_ERROR(EINTR);
mutex_exit(&os->os_upgrade_lock);
if (err != 0)
return (err);
if (issig(JUSTLOOKING) && issig(FORREAL))
return (SET_ERROR(EINTR));
objerr = dmu_bonus_hold(os, obj, FTAG, &db);
if (objerr != 0)
continue;
tx = dmu_tx_create(os);
dmu_tx_hold_bonus(tx, obj);
objerr = dmu_tx_assign(tx, TXG_WAIT);
if (objerr != 0) {
dmu_buf_rele(db, FTAG);
dmu_tx_abort(tx);
continue;
}
dmu_buf_will_dirty(db, tx);
dmu_buf_rele(db, FTAG);
dmu_tx_commit(tx);
}
return (0);
}
static int
dmu_objset_userspace_upgrade_cb(objset_t *os)
{
int err = 0;
if (dmu_objset_userspace_present(os))
return (0);
if (dmu_objset_is_snapshot(os))
return (SET_ERROR(EINVAL));
if (!dmu_objset_userused_enabled(os))
return (SET_ERROR(ENOTSUP));
err = dmu_objset_space_upgrade(os);
if (err)
return (err);
os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
txg_wait_synced(dmu_objset_pool(os), 0);
return (0);
}
void
dmu_objset_userspace_upgrade(objset_t *os)
{
dmu_objset_upgrade(os, dmu_objset_userspace_upgrade_cb);
}
static int
dmu_objset_id_quota_upgrade_cb(objset_t *os)
{
int err = 0;
if (dmu_objset_userobjspace_present(os) &&
dmu_objset_projectquota_present(os))
return (0);
if (dmu_objset_is_snapshot(os))
return (SET_ERROR(EINVAL));
if (!dmu_objset_userused_enabled(os))
return (SET_ERROR(ENOTSUP));
if (!dmu_objset_projectquota_enabled(os) &&
dmu_objset_userobjspace_present(os))
return (SET_ERROR(ENOTSUP));
if (dmu_objset_userobjused_enabled(os))
dmu_objset_ds(os)->ds_feature_activation[
SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE;
if (dmu_objset_projectquota_enabled(os))
dmu_objset_ds(os)->ds_feature_activation[
SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE;
err = dmu_objset_space_upgrade(os);
if (err)
return (err);
os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
if (dmu_objset_userobjused_enabled(os))
os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE;
if (dmu_objset_projectquota_enabled(os))
os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE;
txg_wait_synced(dmu_objset_pool(os), 0);
return (0);
}
void
dmu_objset_id_quota_upgrade(objset_t *os)
{
dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb);
}
boolean_t
dmu_objset_userobjspace_upgradable(objset_t *os)
{
return (dmu_objset_type(os) == DMU_OST_ZFS &&
!dmu_objset_is_snapshot(os) &&
dmu_objset_userobjused_enabled(os) &&
!dmu_objset_userobjspace_present(os) &&
spa_writeable(dmu_objset_spa(os)));
}
boolean_t
dmu_objset_projectquota_upgradable(objset_t *os)
{
return (dmu_objset_type(os) == DMU_OST_ZFS &&
!dmu_objset_is_snapshot(os) &&
dmu_objset_projectquota_enabled(os) &&
!dmu_objset_projectquota_present(os) &&
spa_writeable(dmu_objset_spa(os)));
}
void
dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
uint64_t *usedobjsp, uint64_t *availobjsp)
{
dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
usedobjsp, availobjsp);
}
uint64_t
dmu_objset_fsid_guid(objset_t *os)
{
return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
}
void
dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
{
stat->dds_type = os->os_phys->os_type;
if (os->os_dsl_dataset)
dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
}
void
dmu_objset_stats(objset_t *os, nvlist_t *nv)
{
ASSERT(os->os_dsl_dataset ||
os->os_phys->os_type == DMU_OST_META);
if (os->os_dsl_dataset != NULL)
dsl_dataset_stats(os->os_dsl_dataset, nv);
dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
os->os_phys->os_type);
dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
dmu_objset_userspace_present(os));
}
int
dmu_objset_is_snapshot(objset_t *os)
{
if (os->os_dsl_dataset != NULL)
return (os->os_dsl_dataset->ds_is_snapshot);
else
return (B_FALSE);
}
int
dmu_snapshot_realname(objset_t *os, const char *name, char *real, int maxlen,
boolean_t *conflict)
{
dsl_dataset_t *ds = os->os_dsl_dataset;
uint64_t ignored;
if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
return (SET_ERROR(ENOENT));
return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
MT_NORMALIZE, real, maxlen, conflict));
}
int
dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
{
dsl_dataset_t *ds = os->os_dsl_dataset;
zap_cursor_t cursor;
zap_attribute_t attr;
ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
return (SET_ERROR(ENOENT));
zap_cursor_init_serialized(&cursor,
ds->ds_dir->dd_pool->dp_meta_objset,
dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
if (zap_cursor_retrieve(&cursor, &attr) != 0) {
zap_cursor_fini(&cursor);
return (SET_ERROR(ENOENT));
}
if (strlen(attr.za_name) + 1 > namelen) {
zap_cursor_fini(&cursor);
return (SET_ERROR(ENAMETOOLONG));
}
(void) strlcpy(name, attr.za_name, namelen);
if (idp)
*idp = attr.za_first_integer;
if (case_conflict)
*case_conflict = attr.za_normalization_conflict;
zap_cursor_advance(&cursor);
*offp = zap_cursor_serialize(&cursor);
zap_cursor_fini(&cursor);
return (0);
}
int
dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value)
{
return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value));
}
int
dmu_dir_list_next(objset_t *os, int namelen, char *name,
uint64_t *idp, uint64_t *offp)
{
dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
zap_cursor_t cursor;
zap_attribute_t attr;
/* there is no next dir on a snapshot! */
if (os->os_dsl_dataset->ds_object !=
dsl_dir_phys(dd)->dd_head_dataset_obj)
return (SET_ERROR(ENOENT));
zap_cursor_init_serialized(&cursor,
dd->dd_pool->dp_meta_objset,
dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
if (zap_cursor_retrieve(&cursor, &attr) != 0) {
zap_cursor_fini(&cursor);
return (SET_ERROR(ENOENT));
}
if (strlen(attr.za_name) + 1 > namelen) {
zap_cursor_fini(&cursor);
return (SET_ERROR(ENAMETOOLONG));
}
(void) strlcpy(name, attr.za_name, namelen);
if (idp)
*idp = attr.za_first_integer;
zap_cursor_advance(&cursor);
*offp = zap_cursor_serialize(&cursor);
zap_cursor_fini(&cursor);
return (0);
}
typedef struct dmu_objset_find_ctx {
taskq_t *dc_tq;
dsl_pool_t *dc_dp;
uint64_t dc_ddobj;
char *dc_ddname; /* last component of ddobj's name */
int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
void *dc_arg;
int dc_flags;
kmutex_t *dc_error_lock;
int *dc_error;
} dmu_objset_find_ctx_t;
static void
dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
{
dsl_pool_t *dp = dcp->dc_dp;
dsl_dir_t *dd;
dsl_dataset_t *ds;
zap_cursor_t zc;
zap_attribute_t *attr;
uint64_t thisobj;
int err = 0;
/* don't process if there already was an error */
if (*dcp->dc_error != 0)
goto out;
/*
* Note: passing the name (dc_ddname) here is optional, but it
* improves performance because we don't need to call
* zap_value_search() to determine the name.
*/
err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd);
if (err != 0)
goto out;
/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
if (dd->dd_myname[0] == '$') {
dsl_dir_rele(dd, FTAG);
goto out;
}
thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
/*
* Iterate over all children.
*/
if (dcp->dc_flags & DS_FIND_CHILDREN) {
for (zap_cursor_init(&zc, dp->dp_meta_objset,
dsl_dir_phys(dd)->dd_child_dir_zapobj);
zap_cursor_retrieve(&zc, attr) == 0;
(void) zap_cursor_advance(&zc)) {
ASSERT3U(attr->za_integer_length, ==,
sizeof (uint64_t));
ASSERT3U(attr->za_num_integers, ==, 1);
dmu_objset_find_ctx_t *child_dcp =
kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
*child_dcp = *dcp;
child_dcp->dc_ddobj = attr->za_first_integer;
child_dcp->dc_ddname = spa_strdup(attr->za_name);
if (dcp->dc_tq != NULL)
(void) taskq_dispatch(dcp->dc_tq,
dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
else
dmu_objset_find_dp_impl(child_dcp);
}
zap_cursor_fini(&zc);
}
/*
* Iterate over all snapshots.
*/
if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
dsl_dataset_t *ds;
err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
if (err == 0) {
uint64_t snapobj;
snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
dsl_dataset_rele(ds, FTAG);
for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
zap_cursor_retrieve(&zc, attr) == 0;
(void) zap_cursor_advance(&zc)) {
ASSERT3U(attr->za_integer_length, ==,
sizeof (uint64_t));
ASSERT3U(attr->za_num_integers, ==, 1);
err = dsl_dataset_hold_obj(dp,
attr->za_first_integer, FTAG, &ds);
if (err != 0)
break;
err = dcp->dc_func(dp, ds, dcp->dc_arg);
dsl_dataset_rele(ds, FTAG);
if (err != 0)
break;
}
zap_cursor_fini(&zc);
}
}
kmem_free(attr, sizeof (zap_attribute_t));
if (err != 0) {
dsl_dir_rele(dd, FTAG);
goto out;
}
/*
* Apply to self.
*/
err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
/*
* Note: we hold the dir while calling dsl_dataset_hold_obj() so
* that the dir will remain cached, and we won't have to re-instantiate
* it (which could be expensive due to finding its name via
* zap_value_search()).
*/
dsl_dir_rele(dd, FTAG);
if (err != 0)
goto out;
err = dcp->dc_func(dp, ds, dcp->dc_arg);
dsl_dataset_rele(ds, FTAG);
out:
if (err != 0) {
mutex_enter(dcp->dc_error_lock);
/* only keep first error */
if (*dcp->dc_error == 0)
*dcp->dc_error = err;
mutex_exit(dcp->dc_error_lock);
}
if (dcp->dc_ddname != NULL)
spa_strfree(dcp->dc_ddname);
kmem_free(dcp, sizeof (*dcp));
}
static void
dmu_objset_find_dp_cb(void *arg)
{
dmu_objset_find_ctx_t *dcp = arg;
dsl_pool_t *dp = dcp->dc_dp;
/*
* We need to get a pool_config_lock here, as there are several
* assert(pool_config_held) down the stack. Getting a lock via
* dsl_pool_config_enter is risky, as it might be stalled by a
* pending writer. This would deadlock, as the write lock can
* only be granted when our parent thread gives up the lock.
* The _prio interface gives us priority over a pending writer.
*/
dsl_pool_config_enter_prio(dp, FTAG);
dmu_objset_find_dp_impl(dcp);
dsl_pool_config_exit(dp, FTAG);
}
/*
* Find objsets under and including ddobj, call func(ds) on each.
* The order for the enumeration is completely undefined.
* func is called with dsl_pool_config held.
*/
int
dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
{
int error = 0;
taskq_t *tq = NULL;
int ntasks;
dmu_objset_find_ctx_t *dcp;
kmutex_t err_lock;
mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
dcp->dc_tq = NULL;
dcp->dc_dp = dp;
dcp->dc_ddobj = ddobj;
dcp->dc_ddname = NULL;
dcp->dc_func = func;
dcp->dc_arg = arg;
dcp->dc_flags = flags;
dcp->dc_error_lock = &err_lock;
dcp->dc_error = &error;
if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
/*
* In case a write lock is held we can't make use of
* parallelism, as down the stack of the worker threads
* the lock is asserted via dsl_pool_config_held.
* In case of a read lock this is solved by getting a read
* lock in each worker thread, which isn't possible in case
* of a writer lock. So we fall back to the synchronous path
* here.
* In the future it might be possible to get some magic into
* dsl_pool_config_held in a way that it returns true for
* the worker threads so that a single lock held from this
* thread suffices. For now, stay single threaded.
*/
dmu_objset_find_dp_impl(dcp);
mutex_destroy(&err_lock);
return (error);
}
ntasks = dmu_find_threads;
if (ntasks == 0)
ntasks = vdev_count_leaves(dp->dp_spa) * 4;
tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks,
INT_MAX, 0);
if (tq == NULL) {
kmem_free(dcp, sizeof (*dcp));
mutex_destroy(&err_lock);
return (SET_ERROR(ENOMEM));
}
dcp->dc_tq = tq;
/* dcp will be freed by task */
(void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
/*
* PORTING: this code relies on the property of taskq_wait to wait
* until no more tasks are queued and no more tasks are active. As
* we always queue new tasks from within other tasks, task_wait
* reliably waits for the full recursion to finish, even though we
* enqueue new tasks after taskq_wait has been called.
* On platforms other than illumos, taskq_wait may not have this
* property.
*/
taskq_wait(tq);
taskq_destroy(tq);
mutex_destroy(&err_lock);
return (error);
}
/*
* Find all objsets under name, and for each, call 'func(child_name, arg)'.
* The dp_config_rwlock must not be held when this is called, and it
* will not be held when the callback is called.
* Therefore this function should only be used when the pool is not changing
* (e.g. in syncing context), or the callback can deal with the possible races.
*/
static int
dmu_objset_find_impl(spa_t *spa, const char *name,
int func(const char *, void *), void *arg, int flags)
{
dsl_dir_t *dd;
dsl_pool_t *dp = spa_get_dsl(spa);
dsl_dataset_t *ds;
zap_cursor_t zc;
zap_attribute_t *attr;
char *child;
uint64_t thisobj;
int err;
dsl_pool_config_enter(dp, FTAG);
err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
if (err != 0) {
dsl_pool_config_exit(dp, FTAG);
return (err);
}
/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
if (dd->dd_myname[0] == '$') {
dsl_dir_rele(dd, FTAG);
dsl_pool_config_exit(dp, FTAG);
return (0);
}
thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
/*
* Iterate over all children.
*/
if (flags & DS_FIND_CHILDREN) {
for (zap_cursor_init(&zc, dp->dp_meta_objset,
dsl_dir_phys(dd)->dd_child_dir_zapobj);
zap_cursor_retrieve(&zc, attr) == 0;
(void) zap_cursor_advance(&zc)) {
ASSERT3U(attr->za_integer_length, ==,
sizeof (uint64_t));
ASSERT3U(attr->za_num_integers, ==, 1);
child = kmem_asprintf("%s/%s", name, attr->za_name);
dsl_pool_config_exit(dp, FTAG);
err = dmu_objset_find_impl(spa, child,
func, arg, flags);
dsl_pool_config_enter(dp, FTAG);
kmem_strfree(child);
if (err != 0)
break;
}
zap_cursor_fini(&zc);
if (err != 0) {
dsl_dir_rele(dd, FTAG);
dsl_pool_config_exit(dp, FTAG);
kmem_free(attr, sizeof (zap_attribute_t));
return (err);
}
}
/*
* Iterate over all snapshots.
*/
if (flags & DS_FIND_SNAPSHOTS) {
err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
if (err == 0) {
uint64_t snapobj;
snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
dsl_dataset_rele(ds, FTAG);
for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
zap_cursor_retrieve(&zc, attr) == 0;
(void) zap_cursor_advance(&zc)) {
ASSERT3U(attr->za_integer_length, ==,
sizeof (uint64_t));
ASSERT3U(attr->za_num_integers, ==, 1);
child = kmem_asprintf("%s@%s",
name, attr->za_name);
dsl_pool_config_exit(dp, FTAG);
err = func(child, arg);
dsl_pool_config_enter(dp, FTAG);
kmem_strfree(child);
if (err != 0)
break;
}
zap_cursor_fini(&zc);
}
}
dsl_dir_rele(dd, FTAG);
kmem_free(attr, sizeof (zap_attribute_t));
dsl_pool_config_exit(dp, FTAG);
if (err != 0)
return (err);
/* Apply to self. */
return (func(name, arg));
}
/*
* See comment above dmu_objset_find_impl().
*/
int
dmu_objset_find(const char *name, int func(const char *, void *), void *arg,
int flags)
{
spa_t *spa;
int error;
error = spa_open(name, &spa, FTAG);
if (error != 0)
return (error);
error = dmu_objset_find_impl(spa, name, func, arg, flags);
spa_close(spa, FTAG);
return (error);
}
boolean_t
dmu_objset_incompatible_encryption_version(objset_t *os)
{
return (dsl_dir_incompatible_encryption_version(
os->os_dsl_dataset->ds_dir));
}
void
dmu_objset_set_user(objset_t *os, void *user_ptr)
{
ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
os->os_user_ptr = user_ptr;
}
void *
dmu_objset_get_user(objset_t *os)
{
ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
return (os->os_user_ptr);
}
/*
* Determine name of filesystem, given name of snapshot.
* buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
*/
int
dmu_fsname(const char *snapname, char *buf)
{
char *atp = strchr(snapname, '@');
if (atp == NULL)
return (SET_ERROR(EINVAL));
if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN)
return (SET_ERROR(ENAMETOOLONG));
(void) strlcpy(buf, snapname, atp - snapname + 1);
return (0);
}
/*
* Call when we think we're going to write/free space in open context
* to track the amount of dirty data in the open txg, which is also the
* amount of memory that can not be evicted until this txg syncs.
*
* Note that there are two conditions where this can be called from
* syncing context:
*
* [1] When we just created the dataset, in which case we go on with
* updating any accounting of dirty data as usual.
* [2] When we are dirtying MOS data, in which case we only update the
* pool's accounting of dirty data.
*/
void
dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx)
{
dsl_dataset_t *ds = os->os_dsl_dataset;
int64_t aspace = spa_get_worst_case_asize(os->os_spa, space);
if (ds != NULL) {
dsl_dir_willuse_space(ds->ds_dir, aspace, tx);
}
dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx);
}
#if defined(_KERNEL)
EXPORT_SYMBOL(dmu_objset_zil);
EXPORT_SYMBOL(dmu_objset_pool);
EXPORT_SYMBOL(dmu_objset_ds);
EXPORT_SYMBOL(dmu_objset_type);
EXPORT_SYMBOL(dmu_objset_name);
EXPORT_SYMBOL(dmu_objset_hold);
EXPORT_SYMBOL(dmu_objset_hold_flags);
EXPORT_SYMBOL(dmu_objset_own);
EXPORT_SYMBOL(dmu_objset_rele);
EXPORT_SYMBOL(dmu_objset_rele_flags);
EXPORT_SYMBOL(dmu_objset_disown);
EXPORT_SYMBOL(dmu_objset_from_ds);
EXPORT_SYMBOL(dmu_objset_create);
EXPORT_SYMBOL(dmu_objset_clone);
EXPORT_SYMBOL(dmu_objset_stats);
EXPORT_SYMBOL(dmu_objset_fast_stat);
EXPORT_SYMBOL(dmu_objset_spa);
EXPORT_SYMBOL(dmu_objset_space);
EXPORT_SYMBOL(dmu_objset_fsid_guid);
EXPORT_SYMBOL(dmu_objset_find);
EXPORT_SYMBOL(dmu_objset_byteswap);
EXPORT_SYMBOL(dmu_objset_evict_dbufs);
EXPORT_SYMBOL(dmu_objset_snap_cmtime);
EXPORT_SYMBOL(dmu_objset_dnodesize);
EXPORT_SYMBOL(dmu_objset_sync);
EXPORT_SYMBOL(dmu_objset_is_dirty);
EXPORT_SYMBOL(dmu_objset_create_impl_dnstats);
EXPORT_SYMBOL(dmu_objset_create_impl);
EXPORT_SYMBOL(dmu_objset_open_impl);
EXPORT_SYMBOL(dmu_objset_evict);
EXPORT_SYMBOL(dmu_objset_register_type);
EXPORT_SYMBOL(dmu_objset_sync_done);
EXPORT_SYMBOL(dmu_objset_userquota_get_ids);
EXPORT_SYMBOL(dmu_objset_userused_enabled);
EXPORT_SYMBOL(dmu_objset_userspace_upgrade);
EXPORT_SYMBOL(dmu_objset_userspace_present);
EXPORT_SYMBOL(dmu_objset_userobjused_enabled);
EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable);
EXPORT_SYMBOL(dmu_objset_userobjspace_present);
EXPORT_SYMBOL(dmu_objset_projectquota_enabled);
EXPORT_SYMBOL(dmu_objset_projectquota_present);
EXPORT_SYMBOL(dmu_objset_projectquota_upgradable);
EXPORT_SYMBOL(dmu_objset_id_quota_upgrade);
#endif
diff --git a/sys/contrib/openzfs/module/zfs/metaslab.c b/sys/contrib/openzfs/module/zfs/metaslab.c
index 92f51806ace5..23f3e2989ae7 100644
--- a/sys/contrib/openzfs/module/zfs/metaslab.c
+++ b/sys/contrib/openzfs/module/zfs/metaslab.c
@@ -1,6257 +1,6262 @@
/*
* 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) 2011, 2019 by Delphix. All rights reserved.
* Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
* Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
* Copyright (c) 2017, Intel Corporation.
*/
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#define WITH_DF_BLOCK_ALLOCATOR
#define GANG_ALLOCATION(flags) \
((flags) & (METASLAB_GANG_CHILD | METASLAB_GANG_HEADER))
/*
* Metaslab granularity, in bytes. This is roughly similar to what would be
* referred to as the "stripe size" in traditional RAID arrays. In normal
* operation, we will try to write this amount of data to a top-level vdev
* before moving on to the next one.
*/
unsigned long metaslab_aliquot = 512 << 10;
/*
* For testing, make some blocks above a certain size be gang blocks.
*/
unsigned long metaslab_force_ganging = SPA_MAXBLOCKSIZE + 1;
/*
* In pools where the log space map feature is not enabled we touch
* multiple metaslabs (and their respective space maps) with each
* transaction group. Thus, we benefit from having a small space map
* block size since it allows us to issue more I/O operations scattered
* around the disk. So a sane default for the space map block size
* is 8~16K.
*/
int zfs_metaslab_sm_blksz_no_log = (1 << 14);
/*
* When the log space map feature is enabled, we accumulate a lot of
* changes per metaslab that are flushed once in a while so we benefit
* from a bigger block size like 128K for the metaslab space maps.
*/
int zfs_metaslab_sm_blksz_with_log = (1 << 17);
/*
* The in-core space map representation is more compact than its on-disk form.
* The zfs_condense_pct determines how much more compact the in-core
* space map representation must be before we compact it on-disk.
* Values should be greater than or equal to 100.
*/
int zfs_condense_pct = 200;
/*
* Condensing a metaslab is not guaranteed to actually reduce the amount of
* space used on disk. In particular, a space map uses data in increments of
* MAX(1 << ashift, space_map_blksz), so a metaslab might use the
* same number of blocks after condensing. Since the goal of condensing is to
* reduce the number of IOPs required to read the space map, we only want to
* condense when we can be sure we will reduce the number of blocks used by the
* space map. Unfortunately, we cannot precisely compute whether or not this is
* the case in metaslab_should_condense since we are holding ms_lock. Instead,
* we apply the following heuristic: do not condense a spacemap unless the
* uncondensed size consumes greater than zfs_metaslab_condense_block_threshold
* blocks.
*/
int zfs_metaslab_condense_block_threshold = 4;
/*
* The zfs_mg_noalloc_threshold defines which metaslab groups should
* be eligible for allocation. The value is defined as a percentage of
* free space. Metaslab groups that have more free space than
* zfs_mg_noalloc_threshold are always eligible for allocations. Once
* a metaslab group's free space is less than or equal to the
* zfs_mg_noalloc_threshold the allocator will avoid allocating to that
* group unless all groups in the pool have reached zfs_mg_noalloc_threshold.
* Once all groups in the pool reach zfs_mg_noalloc_threshold then all
* groups are allowed to accept allocations. Gang blocks are always
* eligible to allocate on any metaslab group. The default value of 0 means
* no metaslab group will be excluded based on this criterion.
*/
int zfs_mg_noalloc_threshold = 0;
/*
* Metaslab groups are considered eligible for allocations if their
* fragmentation metric (measured as a percentage) is less than or
* equal to zfs_mg_fragmentation_threshold. If a metaslab group
* exceeds this threshold then it will be skipped unless all metaslab
* groups within the metaslab class have also crossed this threshold.
*
* This tunable was introduced to avoid edge cases where we continue
* allocating from very fragmented disks in our pool while other, less
* fragmented disks, exists. On the other hand, if all disks in the
* pool are uniformly approaching the threshold, the threshold can
* be a speed bump in performance, where we keep switching the disks
* that we allocate from (e.g. we allocate some segments from disk A
* making it bypassing the threshold while freeing segments from disk
* B getting its fragmentation below the threshold).
*
* Empirically, we've seen that our vdev selection for allocations is
* good enough that fragmentation increases uniformly across all vdevs
* the majority of the time. Thus we set the threshold percentage high
* enough to avoid hitting the speed bump on pools that are being pushed
* to the edge.
*/
int zfs_mg_fragmentation_threshold = 95;
/*
* Allow metaslabs to keep their active state as long as their fragmentation
* percentage is less than or equal to zfs_metaslab_fragmentation_threshold. An
* active metaslab that exceeds this threshold will no longer keep its active
* status allowing better metaslabs to be selected.
*/
int zfs_metaslab_fragmentation_threshold = 70;
/*
* When set will load all metaslabs when pool is first opened.
*/
int metaslab_debug_load = 0;
/*
* When set will prevent metaslabs from being unloaded.
*/
int metaslab_debug_unload = 0;
/*
* Minimum size which forces the dynamic allocator to change
* it's allocation strategy. Once the space map cannot satisfy
* an allocation of this size then it switches to using more
* aggressive strategy (i.e search by size rather than offset).
*/
uint64_t metaslab_df_alloc_threshold = SPA_OLD_MAXBLOCKSIZE;
/*
* The minimum free space, in percent, which must be available
* in a space map to continue allocations in a first-fit fashion.
* Once the space map's free space drops below this level we dynamically
* switch to using best-fit allocations.
*/
int metaslab_df_free_pct = 4;
/*
* Maximum distance to search forward from the last offset. Without this
* limit, fragmented pools can see >100,000 iterations and
* metaslab_block_picker() becomes the performance limiting factor on
* high-performance storage.
*
* With the default setting of 16MB, we typically see less than 500
* iterations, even with very fragmented, ashift=9 pools. The maximum number
* of iterations possible is:
* metaslab_df_max_search / (2 * (1<60KB (but fewer segments in this
* bucket, and therefore a lower weight).
*/
int zfs_metaslab_find_max_tries = 100;
static uint64_t metaslab_weight(metaslab_t *, boolean_t);
static void metaslab_set_fragmentation(metaslab_t *, boolean_t);
static void metaslab_free_impl(vdev_t *, uint64_t, uint64_t, boolean_t);
static void metaslab_check_free_impl(vdev_t *, uint64_t, uint64_t);
static void metaslab_passivate(metaslab_t *msp, uint64_t weight);
static uint64_t metaslab_weight_from_range_tree(metaslab_t *msp);
static void metaslab_flush_update(metaslab_t *, dmu_tx_t *);
static unsigned int metaslab_idx_func(multilist_t *, void *);
static void metaslab_evict(metaslab_t *, uint64_t);
static void metaslab_rt_add(range_tree_t *rt, range_seg_t *rs, void *arg);
kmem_cache_t *metaslab_alloc_trace_cache;
typedef struct metaslab_stats {
kstat_named_t metaslabstat_trace_over_limit;
kstat_named_t metaslabstat_reload_tree;
kstat_named_t metaslabstat_too_many_tries;
kstat_named_t metaslabstat_try_hard;
} metaslab_stats_t;
static metaslab_stats_t metaslab_stats = {
{ "trace_over_limit", KSTAT_DATA_UINT64 },
{ "reload_tree", KSTAT_DATA_UINT64 },
{ "too_many_tries", KSTAT_DATA_UINT64 },
{ "try_hard", KSTAT_DATA_UINT64 },
};
#define METASLABSTAT_BUMP(stat) \
atomic_inc_64(&metaslab_stats.stat.value.ui64);
kstat_t *metaslab_ksp;
void
metaslab_stat_init(void)
{
ASSERT(metaslab_alloc_trace_cache == NULL);
metaslab_alloc_trace_cache = kmem_cache_create(
"metaslab_alloc_trace_cache", sizeof (metaslab_alloc_trace_t),
0, NULL, NULL, NULL, NULL, NULL, 0);
metaslab_ksp = kstat_create("zfs", 0, "metaslab_stats",
"misc", KSTAT_TYPE_NAMED, sizeof (metaslab_stats) /
sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL);
if (metaslab_ksp != NULL) {
metaslab_ksp->ks_data = &metaslab_stats;
kstat_install(metaslab_ksp);
}
}
void
metaslab_stat_fini(void)
{
if (metaslab_ksp != NULL) {
kstat_delete(metaslab_ksp);
metaslab_ksp = NULL;
}
kmem_cache_destroy(metaslab_alloc_trace_cache);
metaslab_alloc_trace_cache = NULL;
}
/*
* ==========================================================================
* Metaslab classes
* ==========================================================================
*/
metaslab_class_t *
metaslab_class_create(spa_t *spa, metaslab_ops_t *ops)
{
metaslab_class_t *mc;
mc = kmem_zalloc(offsetof(metaslab_class_t,
mc_allocator[spa->spa_alloc_count]), KM_SLEEP);
mc->mc_spa = spa;
mc->mc_ops = ops;
mutex_init(&mc->mc_lock, NULL, MUTEX_DEFAULT, NULL);
multilist_create(&mc->mc_metaslab_txg_list, sizeof (metaslab_t),
offsetof(metaslab_t, ms_class_txg_node), metaslab_idx_func);
for (int i = 0; i < spa->spa_alloc_count; i++) {
metaslab_class_allocator_t *mca = &mc->mc_allocator[i];
mca->mca_rotor = NULL;
zfs_refcount_create_tracked(&mca->mca_alloc_slots);
}
return (mc);
}
void
metaslab_class_destroy(metaslab_class_t *mc)
{
spa_t *spa = mc->mc_spa;
ASSERT(mc->mc_alloc == 0);
ASSERT(mc->mc_deferred == 0);
ASSERT(mc->mc_space == 0);
ASSERT(mc->mc_dspace == 0);
for (int i = 0; i < spa->spa_alloc_count; i++) {
metaslab_class_allocator_t *mca = &mc->mc_allocator[i];
ASSERT(mca->mca_rotor == NULL);
zfs_refcount_destroy(&mca->mca_alloc_slots);
}
mutex_destroy(&mc->mc_lock);
multilist_destroy(&mc->mc_metaslab_txg_list);
kmem_free(mc, offsetof(metaslab_class_t,
mc_allocator[spa->spa_alloc_count]));
}
int
metaslab_class_validate(metaslab_class_t *mc)
{
metaslab_group_t *mg;
vdev_t *vd;
/*
* Must hold one of the spa_config locks.
*/
ASSERT(spa_config_held(mc->mc_spa, SCL_ALL, RW_READER) ||
spa_config_held(mc->mc_spa, SCL_ALL, RW_WRITER));
if ((mg = mc->mc_allocator[0].mca_rotor) == NULL)
return (0);
do {
vd = mg->mg_vd;
ASSERT(vd->vdev_mg != NULL);
ASSERT3P(vd->vdev_top, ==, vd);
ASSERT3P(mg->mg_class, ==, mc);
ASSERT3P(vd->vdev_ops, !=, &vdev_hole_ops);
} while ((mg = mg->mg_next) != mc->mc_allocator[0].mca_rotor);
return (0);
}
static void
metaslab_class_space_update(metaslab_class_t *mc, int64_t alloc_delta,
int64_t defer_delta, int64_t space_delta, int64_t dspace_delta)
{
atomic_add_64(&mc->mc_alloc, alloc_delta);
atomic_add_64(&mc->mc_deferred, defer_delta);
atomic_add_64(&mc->mc_space, space_delta);
atomic_add_64(&mc->mc_dspace, dspace_delta);
}
uint64_t
metaslab_class_get_alloc(metaslab_class_t *mc)
{
return (mc->mc_alloc);
}
uint64_t
metaslab_class_get_deferred(metaslab_class_t *mc)
{
return (mc->mc_deferred);
}
uint64_t
metaslab_class_get_space(metaslab_class_t *mc)
{
return (mc->mc_space);
}
uint64_t
metaslab_class_get_dspace(metaslab_class_t *mc)
{
return (spa_deflate(mc->mc_spa) ? mc->mc_dspace : mc->mc_space);
}
void
metaslab_class_histogram_verify(metaslab_class_t *mc)
{
spa_t *spa = mc->mc_spa;
vdev_t *rvd = spa->spa_root_vdev;
uint64_t *mc_hist;
int i;
if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0)
return;
mc_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE,
KM_SLEEP);
mutex_enter(&mc->mc_lock);
for (int c = 0; c < rvd->vdev_children; c++) {
vdev_t *tvd = rvd->vdev_child[c];
metaslab_group_t *mg = vdev_get_mg(tvd, mc);
/*
* Skip any holes, uninitialized top-levels, or
* vdevs that are not in this metalab class.
*/
if (!vdev_is_concrete(tvd) || tvd->vdev_ms_shift == 0 ||
mg->mg_class != mc) {
continue;
}
IMPLY(mg == mg->mg_vd->vdev_log_mg,
mc == spa_embedded_log_class(mg->mg_vd->vdev_spa));
for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
mc_hist[i] += mg->mg_histogram[i];
}
for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
VERIFY3U(mc_hist[i], ==, mc->mc_histogram[i]);
}
mutex_exit(&mc->mc_lock);
kmem_free(mc_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE);
}
/*
* Calculate the metaslab class's fragmentation metric. The metric
* is weighted based on the space contribution of each metaslab group.
* The return value will be a number between 0 and 100 (inclusive), or
* ZFS_FRAG_INVALID if the metric has not been set. See comment above the
* zfs_frag_table for more information about the metric.
*/
uint64_t
metaslab_class_fragmentation(metaslab_class_t *mc)
{
vdev_t *rvd = mc->mc_spa->spa_root_vdev;
uint64_t fragmentation = 0;
spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER);
for (int c = 0; c < rvd->vdev_children; c++) {
vdev_t *tvd = rvd->vdev_child[c];
metaslab_group_t *mg = tvd->vdev_mg;
/*
* Skip any holes, uninitialized top-levels,
* or vdevs that are not in this metalab class.
*/
if (!vdev_is_concrete(tvd) || tvd->vdev_ms_shift == 0 ||
mg->mg_class != mc) {
continue;
}
/*
* If a metaslab group does not contain a fragmentation
* metric then just bail out.
*/
if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG);
return (ZFS_FRAG_INVALID);
}
/*
* Determine how much this metaslab_group is contributing
* to the overall pool fragmentation metric.
*/
fragmentation += mg->mg_fragmentation *
metaslab_group_get_space(mg);
}
fragmentation /= metaslab_class_get_space(mc);
ASSERT3U(fragmentation, <=, 100);
spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG);
return (fragmentation);
}
/*
* Calculate the amount of expandable space that is available in
* this metaslab class. If a device is expanded then its expandable
* space will be the amount of allocatable space that is currently not
* part of this metaslab class.
*/
uint64_t
metaslab_class_expandable_space(metaslab_class_t *mc)
{
vdev_t *rvd = mc->mc_spa->spa_root_vdev;
uint64_t space = 0;
spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER);
for (int c = 0; c < rvd->vdev_children; c++) {
vdev_t *tvd = rvd->vdev_child[c];
metaslab_group_t *mg = tvd->vdev_mg;
if (!vdev_is_concrete(tvd) || tvd->vdev_ms_shift == 0 ||
mg->mg_class != mc) {
continue;
}
/*
* Calculate if we have enough space to add additional
* metaslabs. We report the expandable space in terms
* of the metaslab size since that's the unit of expansion.
*/
space += P2ALIGN(tvd->vdev_max_asize - tvd->vdev_asize,
1ULL << tvd->vdev_ms_shift);
}
spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG);
return (space);
}
void
metaslab_class_evict_old(metaslab_class_t *mc, uint64_t txg)
{
multilist_t *ml = &mc->mc_metaslab_txg_list;
for (int i = 0; i < multilist_get_num_sublists(ml); i++) {
multilist_sublist_t *mls = multilist_sublist_lock(ml, i);
metaslab_t *msp = multilist_sublist_head(mls);
multilist_sublist_unlock(mls);
while (msp != NULL) {
mutex_enter(&msp->ms_lock);
/*
* If the metaslab has been removed from the list
* (which could happen if we were at the memory limit
* and it was evicted during this loop), then we can't
* proceed and we should restart the sublist.
*/
if (!multilist_link_active(&msp->ms_class_txg_node)) {
mutex_exit(&msp->ms_lock);
i--;
break;
}
mls = multilist_sublist_lock(ml, i);
metaslab_t *next_msp = multilist_sublist_next(mls, msp);
multilist_sublist_unlock(mls);
if (txg >
msp->ms_selected_txg + metaslab_unload_delay &&
gethrtime() > msp->ms_selected_time +
(uint64_t)MSEC2NSEC(metaslab_unload_delay_ms)) {
metaslab_evict(msp, txg);
} else {
/*
* Once we've hit a metaslab selected too
* recently to evict, we're done evicting for
* now.
*/
mutex_exit(&msp->ms_lock);
break;
}
mutex_exit(&msp->ms_lock);
msp = next_msp;
}
}
}
static int
metaslab_compare(const void *x1, const void *x2)
{
const metaslab_t *m1 = (const metaslab_t *)x1;
const metaslab_t *m2 = (const metaslab_t *)x2;
int sort1 = 0;
int sort2 = 0;
if (m1->ms_allocator != -1 && m1->ms_primary)
sort1 = 1;
else if (m1->ms_allocator != -1 && !m1->ms_primary)
sort1 = 2;
if (m2->ms_allocator != -1 && m2->ms_primary)
sort2 = 1;
else if (m2->ms_allocator != -1 && !m2->ms_primary)
sort2 = 2;
/*
* Sort inactive metaslabs first, then primaries, then secondaries. When
* selecting a metaslab to allocate from, an allocator first tries its
* primary, then secondary active metaslab. If it doesn't have active
* metaslabs, or can't allocate from them, it searches for an inactive
* metaslab to activate. If it can't find a suitable one, it will steal
* a primary or secondary metaslab from another allocator.
*/
if (sort1 < sort2)
return (-1);
if (sort1 > sort2)
return (1);
int cmp = TREE_CMP(m2->ms_weight, m1->ms_weight);
if (likely(cmp))
return (cmp);
IMPLY(TREE_CMP(m1->ms_start, m2->ms_start) == 0, m1 == m2);
return (TREE_CMP(m1->ms_start, m2->ms_start));
}
/*
* ==========================================================================
* Metaslab groups
* ==========================================================================
*/
/*
* Update the allocatable flag and the metaslab group's capacity.
* The allocatable flag is set to true if the capacity is below
* the zfs_mg_noalloc_threshold or has a fragmentation value that is
* greater than zfs_mg_fragmentation_threshold. If a metaslab group
* transitions from allocatable to non-allocatable or vice versa then the
* metaslab group's class is updated to reflect the transition.
*/
static void
metaslab_group_alloc_update(metaslab_group_t *mg)
{
vdev_t *vd = mg->mg_vd;
metaslab_class_t *mc = mg->mg_class;
vdev_stat_t *vs = &vd->vdev_stat;
boolean_t was_allocatable;
boolean_t was_initialized;
ASSERT(vd == vd->vdev_top);
ASSERT3U(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_READER), ==,
SCL_ALLOC);
mutex_enter(&mg->mg_lock);
was_allocatable = mg->mg_allocatable;
was_initialized = mg->mg_initialized;
mg->mg_free_capacity = ((vs->vs_space - vs->vs_alloc) * 100) /
(vs->vs_space + 1);
mutex_enter(&mc->mc_lock);
/*
* If the metaslab group was just added then it won't
* have any space until we finish syncing out this txg.
* At that point we will consider it initialized and available
* for allocations. We also don't consider non-activated
* metaslab groups (e.g. vdevs that are in the middle of being removed)
* to be initialized, because they can't be used for allocation.
*/
mg->mg_initialized = metaslab_group_initialized(mg);
if (!was_initialized && mg->mg_initialized) {
mc->mc_groups++;
} else if (was_initialized && !mg->mg_initialized) {
ASSERT3U(mc->mc_groups, >, 0);
mc->mc_groups--;
}
if (mg->mg_initialized)
mg->mg_no_free_space = B_FALSE;
/*
* A metaslab group is considered allocatable if it has plenty
* of free space or is not heavily fragmented. We only take
* fragmentation into account if the metaslab group has a valid
* fragmentation metric (i.e. a value between 0 and 100).
*/
mg->mg_allocatable = (mg->mg_activation_count > 0 &&
mg->mg_free_capacity > zfs_mg_noalloc_threshold &&
(mg->mg_fragmentation == ZFS_FRAG_INVALID ||
mg->mg_fragmentation <= zfs_mg_fragmentation_threshold));
/*
* The mc_alloc_groups maintains a count of the number of
* groups in this metaslab class that are still above the
* zfs_mg_noalloc_threshold. This is used by the allocating
* threads to determine if they should avoid allocations to
* a given group. The allocator will avoid allocations to a group
* if that group has reached or is below the zfs_mg_noalloc_threshold
* and there are still other groups that are above the threshold.
* When a group transitions from allocatable to non-allocatable or
* vice versa we update the metaslab class to reflect that change.
* When the mc_alloc_groups value drops to 0 that means that all
* groups have reached the zfs_mg_noalloc_threshold making all groups
* eligible for allocations. This effectively means that all devices
* are balanced again.
*/
if (was_allocatable && !mg->mg_allocatable)
mc->mc_alloc_groups--;
else if (!was_allocatable && mg->mg_allocatable)
mc->mc_alloc_groups++;
mutex_exit(&mc->mc_lock);
mutex_exit(&mg->mg_lock);
}
int
metaslab_sort_by_flushed(const void *va, const void *vb)
{
const metaslab_t *a = va;
const metaslab_t *b = vb;
int cmp = TREE_CMP(a->ms_unflushed_txg, b->ms_unflushed_txg);
if (likely(cmp))
return (cmp);
uint64_t a_vdev_id = a->ms_group->mg_vd->vdev_id;
uint64_t b_vdev_id = b->ms_group->mg_vd->vdev_id;
cmp = TREE_CMP(a_vdev_id, b_vdev_id);
if (cmp)
return (cmp);
return (TREE_CMP(a->ms_id, b->ms_id));
}
metaslab_group_t *
metaslab_group_create(metaslab_class_t *mc, vdev_t *vd, int allocators)
{
metaslab_group_t *mg;
mg = kmem_zalloc(offsetof(metaslab_group_t,
mg_allocator[allocators]), KM_SLEEP);
mutex_init(&mg->mg_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&mg->mg_ms_disabled_lock, NULL, MUTEX_DEFAULT, NULL);
cv_init(&mg->mg_ms_disabled_cv, NULL, CV_DEFAULT, NULL);
avl_create(&mg->mg_metaslab_tree, metaslab_compare,
sizeof (metaslab_t), offsetof(metaslab_t, ms_group_node));
mg->mg_vd = vd;
mg->mg_class = mc;
mg->mg_activation_count = 0;
mg->mg_initialized = B_FALSE;
mg->mg_no_free_space = B_TRUE;
mg->mg_allocators = allocators;
for (int i = 0; i < allocators; i++) {
metaslab_group_allocator_t *mga = &mg->mg_allocator[i];
zfs_refcount_create_tracked(&mga->mga_alloc_queue_depth);
}
mg->mg_taskq = taskq_create("metaslab_group_taskq", metaslab_load_pct,
maxclsyspri, 10, INT_MAX, TASKQ_THREADS_CPU_PCT | TASKQ_DYNAMIC);
return (mg);
}
void
metaslab_group_destroy(metaslab_group_t *mg)
{
ASSERT(mg->mg_prev == NULL);
ASSERT(mg->mg_next == NULL);
/*
* We may have gone below zero with the activation count
* either because we never activated in the first place or
* because we're done, and possibly removing the vdev.
*/
ASSERT(mg->mg_activation_count <= 0);
taskq_destroy(mg->mg_taskq);
avl_destroy(&mg->mg_metaslab_tree);
mutex_destroy(&mg->mg_lock);
mutex_destroy(&mg->mg_ms_disabled_lock);
cv_destroy(&mg->mg_ms_disabled_cv);
for (int i = 0; i < mg->mg_allocators; i++) {
metaslab_group_allocator_t *mga = &mg->mg_allocator[i];
zfs_refcount_destroy(&mga->mga_alloc_queue_depth);
}
kmem_free(mg, offsetof(metaslab_group_t,
mg_allocator[mg->mg_allocators]));
}
void
metaslab_group_activate(metaslab_group_t *mg)
{
metaslab_class_t *mc = mg->mg_class;
spa_t *spa = mc->mc_spa;
metaslab_group_t *mgprev, *mgnext;
ASSERT3U(spa_config_held(spa, SCL_ALLOC, RW_WRITER), !=, 0);
ASSERT(mg->mg_prev == NULL);
ASSERT(mg->mg_next == NULL);
ASSERT(mg->mg_activation_count <= 0);
if (++mg->mg_activation_count <= 0)
return;
mg->mg_aliquot = metaslab_aliquot * MAX(1, mg->mg_vd->vdev_children);
metaslab_group_alloc_update(mg);
if ((mgprev = mc->mc_allocator[0].mca_rotor) == NULL) {
mg->mg_prev = mg;
mg->mg_next = mg;
} else {
mgnext = mgprev->mg_next;
mg->mg_prev = mgprev;
mg->mg_next = mgnext;
mgprev->mg_next = mg;
mgnext->mg_prev = mg;
}
for (int i = 0; i < spa->spa_alloc_count; i++) {
mc->mc_allocator[i].mca_rotor = mg;
mg = mg->mg_next;
}
}
/*
* Passivate a metaslab group and remove it from the allocation rotor.
* Callers must hold both the SCL_ALLOC and SCL_ZIO lock prior to passivating
* a metaslab group. This function will momentarily drop spa_config_locks
* that are lower than the SCL_ALLOC lock (see comment below).
*/
void
metaslab_group_passivate(metaslab_group_t *mg)
{
metaslab_class_t *mc = mg->mg_class;
spa_t *spa = mc->mc_spa;
metaslab_group_t *mgprev, *mgnext;
int locks = spa_config_held(spa, SCL_ALL, RW_WRITER);
ASSERT3U(spa_config_held(spa, SCL_ALLOC | SCL_ZIO, RW_WRITER), ==,
(SCL_ALLOC | SCL_ZIO));
if (--mg->mg_activation_count != 0) {
for (int i = 0; i < spa->spa_alloc_count; i++)
ASSERT(mc->mc_allocator[i].mca_rotor != mg);
ASSERT(mg->mg_prev == NULL);
ASSERT(mg->mg_next == NULL);
ASSERT(mg->mg_activation_count < 0);
return;
}
/*
* The spa_config_lock is an array of rwlocks, ordered as
* follows (from highest to lowest):
* SCL_CONFIG > SCL_STATE > SCL_L2ARC > SCL_ALLOC >
* SCL_ZIO > SCL_FREE > SCL_VDEV
* (For more information about the spa_config_lock see spa_misc.c)
* The higher the lock, the broader its coverage. When we passivate
* a metaslab group, we must hold both the SCL_ALLOC and the SCL_ZIO
* config locks. However, the metaslab group's taskq might be trying
* to preload metaslabs so we must drop the SCL_ZIO lock and any
* lower locks to allow the I/O to complete. At a minimum,
* we continue to hold the SCL_ALLOC lock, which prevents any future
* allocations from taking place and any changes to the vdev tree.
*/
spa_config_exit(spa, locks & ~(SCL_ZIO - 1), spa);
taskq_wait_outstanding(mg->mg_taskq, 0);
spa_config_enter(spa, locks & ~(SCL_ZIO - 1), spa, RW_WRITER);
metaslab_group_alloc_update(mg);
for (int i = 0; i < mg->mg_allocators; i++) {
metaslab_group_allocator_t *mga = &mg->mg_allocator[i];
metaslab_t *msp = mga->mga_primary;
if (msp != NULL) {
mutex_enter(&msp->ms_lock);
metaslab_passivate(msp,
metaslab_weight_from_range_tree(msp));
mutex_exit(&msp->ms_lock);
}
msp = mga->mga_secondary;
if (msp != NULL) {
mutex_enter(&msp->ms_lock);
metaslab_passivate(msp,
metaslab_weight_from_range_tree(msp));
mutex_exit(&msp->ms_lock);
}
}
mgprev = mg->mg_prev;
mgnext = mg->mg_next;
if (mg == mgnext) {
mgnext = NULL;
} else {
mgprev->mg_next = mgnext;
mgnext->mg_prev = mgprev;
}
for (int i = 0; i < spa->spa_alloc_count; i++) {
if (mc->mc_allocator[i].mca_rotor == mg)
mc->mc_allocator[i].mca_rotor = mgnext;
}
mg->mg_prev = NULL;
mg->mg_next = NULL;
}
boolean_t
metaslab_group_initialized(metaslab_group_t *mg)
{
vdev_t *vd = mg->mg_vd;
vdev_stat_t *vs = &vd->vdev_stat;
return (vs->vs_space != 0 && mg->mg_activation_count > 0);
}
uint64_t
metaslab_group_get_space(metaslab_group_t *mg)
{
/*
* Note that the number of nodes in mg_metaslab_tree may be one less
* than vdev_ms_count, due to the embedded log metaslab.
*/
mutex_enter(&mg->mg_lock);
uint64_t ms_count = avl_numnodes(&mg->mg_metaslab_tree);
mutex_exit(&mg->mg_lock);
return ((1ULL << mg->mg_vd->vdev_ms_shift) * ms_count);
}
void
metaslab_group_histogram_verify(metaslab_group_t *mg)
{
uint64_t *mg_hist;
avl_tree_t *t = &mg->mg_metaslab_tree;
uint64_t ashift = mg->mg_vd->vdev_ashift;
if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0)
return;
mg_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE,
KM_SLEEP);
ASSERT3U(RANGE_TREE_HISTOGRAM_SIZE, >=,
SPACE_MAP_HISTOGRAM_SIZE + ashift);
mutex_enter(&mg->mg_lock);
for (metaslab_t *msp = avl_first(t);
msp != NULL; msp = AVL_NEXT(t, msp)) {
VERIFY3P(msp->ms_group, ==, mg);
/* skip if not active */
if (msp->ms_sm == NULL)
continue;
for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
mg_hist[i + ashift] +=
msp->ms_sm->sm_phys->smp_histogram[i];
}
}
for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i ++)
VERIFY3U(mg_hist[i], ==, mg->mg_histogram[i]);
mutex_exit(&mg->mg_lock);
kmem_free(mg_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE);
}
static void
metaslab_group_histogram_add(metaslab_group_t *mg, metaslab_t *msp)
{
metaslab_class_t *mc = mg->mg_class;
uint64_t ashift = mg->mg_vd->vdev_ashift;
ASSERT(MUTEX_HELD(&msp->ms_lock));
if (msp->ms_sm == NULL)
return;
mutex_enter(&mg->mg_lock);
mutex_enter(&mc->mc_lock);
for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
IMPLY(mg == mg->mg_vd->vdev_log_mg,
mc == spa_embedded_log_class(mg->mg_vd->vdev_spa));
mg->mg_histogram[i + ashift] +=
msp->ms_sm->sm_phys->smp_histogram[i];
mc->mc_histogram[i + ashift] +=
msp->ms_sm->sm_phys->smp_histogram[i];
}
mutex_exit(&mc->mc_lock);
mutex_exit(&mg->mg_lock);
}
void
metaslab_group_histogram_remove(metaslab_group_t *mg, metaslab_t *msp)
{
metaslab_class_t *mc = mg->mg_class;
uint64_t ashift = mg->mg_vd->vdev_ashift;
ASSERT(MUTEX_HELD(&msp->ms_lock));
if (msp->ms_sm == NULL)
return;
mutex_enter(&mg->mg_lock);
mutex_enter(&mc->mc_lock);
for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
ASSERT3U(mg->mg_histogram[i + ashift], >=,
msp->ms_sm->sm_phys->smp_histogram[i]);
ASSERT3U(mc->mc_histogram[i + ashift], >=,
msp->ms_sm->sm_phys->smp_histogram[i]);
IMPLY(mg == mg->mg_vd->vdev_log_mg,
mc == spa_embedded_log_class(mg->mg_vd->vdev_spa));
mg->mg_histogram[i + ashift] -=
msp->ms_sm->sm_phys->smp_histogram[i];
mc->mc_histogram[i + ashift] -=
msp->ms_sm->sm_phys->smp_histogram[i];
}
mutex_exit(&mc->mc_lock);
mutex_exit(&mg->mg_lock);
}
static void
metaslab_group_add(metaslab_group_t *mg, metaslab_t *msp)
{
ASSERT(msp->ms_group == NULL);
mutex_enter(&mg->mg_lock);
msp->ms_group = mg;
msp->ms_weight = 0;
avl_add(&mg->mg_metaslab_tree, msp);
mutex_exit(&mg->mg_lock);
mutex_enter(&msp->ms_lock);
metaslab_group_histogram_add(mg, msp);
mutex_exit(&msp->ms_lock);
}
static void
metaslab_group_remove(metaslab_group_t *mg, metaslab_t *msp)
{
mutex_enter(&msp->ms_lock);
metaslab_group_histogram_remove(mg, msp);
mutex_exit(&msp->ms_lock);
mutex_enter(&mg->mg_lock);
ASSERT(msp->ms_group == mg);
avl_remove(&mg->mg_metaslab_tree, msp);
metaslab_class_t *mc = msp->ms_group->mg_class;
multilist_sublist_t *mls =
multilist_sublist_lock_obj(&mc->mc_metaslab_txg_list, msp);
if (multilist_link_active(&msp->ms_class_txg_node))
multilist_sublist_remove(mls, msp);
multilist_sublist_unlock(mls);
msp->ms_group = NULL;
mutex_exit(&mg->mg_lock);
}
static void
metaslab_group_sort_impl(metaslab_group_t *mg, metaslab_t *msp, uint64_t weight)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
ASSERT(MUTEX_HELD(&mg->mg_lock));
ASSERT(msp->ms_group == mg);
avl_remove(&mg->mg_metaslab_tree, msp);
msp->ms_weight = weight;
avl_add(&mg->mg_metaslab_tree, msp);
}
static void
metaslab_group_sort(metaslab_group_t *mg, metaslab_t *msp, uint64_t weight)
{
/*
* Although in principle the weight can be any value, in
* practice we do not use values in the range [1, 511].
*/
ASSERT(weight >= SPA_MINBLOCKSIZE || weight == 0);
ASSERT(MUTEX_HELD(&msp->ms_lock));
mutex_enter(&mg->mg_lock);
metaslab_group_sort_impl(mg, msp, weight);
mutex_exit(&mg->mg_lock);
}
/*
* Calculate the fragmentation for a given metaslab group. We can use
* a simple average here since all metaslabs within the group must have
* the same size. The return value will be a value between 0 and 100
* (inclusive), or ZFS_FRAG_INVALID if less than half of the metaslab in this
* group have a fragmentation metric.
*/
uint64_t
metaslab_group_fragmentation(metaslab_group_t *mg)
{
vdev_t *vd = mg->mg_vd;
uint64_t fragmentation = 0;
uint64_t valid_ms = 0;
for (int m = 0; m < vd->vdev_ms_count; m++) {
metaslab_t *msp = vd->vdev_ms[m];
if (msp->ms_fragmentation == ZFS_FRAG_INVALID)
continue;
if (msp->ms_group != mg)
continue;
valid_ms++;
fragmentation += msp->ms_fragmentation;
}
if (valid_ms <= mg->mg_vd->vdev_ms_count / 2)
return (ZFS_FRAG_INVALID);
fragmentation /= valid_ms;
ASSERT3U(fragmentation, <=, 100);
return (fragmentation);
}
/*
* Determine if a given metaslab group should skip allocations. A metaslab
* group should avoid allocations if its free capacity is less than the
* zfs_mg_noalloc_threshold or its fragmentation metric is greater than
* zfs_mg_fragmentation_threshold and there is at least one metaslab group
* that can still handle allocations. If the allocation throttle is enabled
* then we skip allocations to devices that have reached their maximum
* allocation queue depth unless the selected metaslab group is the only
* eligible group remaining.
*/
static boolean_t
metaslab_group_allocatable(metaslab_group_t *mg, metaslab_group_t *rotor,
uint64_t psize, int allocator, int d)
{
spa_t *spa = mg->mg_vd->vdev_spa;
metaslab_class_t *mc = mg->mg_class;
/*
* We can only consider skipping this metaslab group if it's
* in the normal metaslab class and there are other metaslab
* groups to select from. Otherwise, we always consider it eligible
* for allocations.
*/
if ((mc != spa_normal_class(spa) &&
mc != spa_special_class(spa) &&
mc != spa_dedup_class(spa)) ||
mc->mc_groups <= 1)
return (B_TRUE);
/*
* If the metaslab group's mg_allocatable flag is set (see comments
* in metaslab_group_alloc_update() for more information) and
* the allocation throttle is disabled then allow allocations to this
* device. However, if the allocation throttle is enabled then
* check if we have reached our allocation limit (mga_alloc_queue_depth)
* to determine if we should allow allocations to this metaslab group.
* If all metaslab groups are no longer considered allocatable
* (mc_alloc_groups == 0) or we're trying to allocate the smallest
* gang block size then we allow allocations on this metaslab group
* regardless of the mg_allocatable or throttle settings.
*/
if (mg->mg_allocatable) {
metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator];
int64_t qdepth;
uint64_t qmax = mga->mga_cur_max_alloc_queue_depth;
if (!mc->mc_alloc_throttle_enabled)
return (B_TRUE);
/*
* If this metaslab group does not have any free space, then
* there is no point in looking further.
*/
if (mg->mg_no_free_space)
return (B_FALSE);
/*
* Relax allocation throttling for ditto blocks. Due to
* random imbalances in allocation it tends to push copies
* to one vdev, that looks a bit better at the moment.
*/
qmax = qmax * (4 + d) / 4;
qdepth = zfs_refcount_count(&mga->mga_alloc_queue_depth);
/*
* If this metaslab group is below its qmax or it's
* the only allocatable metasable group, then attempt
* to allocate from it.
*/
if (qdepth < qmax || mc->mc_alloc_groups == 1)
return (B_TRUE);
ASSERT3U(mc->mc_alloc_groups, >, 1);
/*
* Since this metaslab group is at or over its qmax, we
* need to determine if there are metaslab groups after this
* one that might be able to handle this allocation. This is
* racy since we can't hold the locks for all metaslab
* groups at the same time when we make this check.
*/
for (metaslab_group_t *mgp = mg->mg_next;
mgp != rotor; mgp = mgp->mg_next) {
metaslab_group_allocator_t *mgap =
&mgp->mg_allocator[allocator];
qmax = mgap->mga_cur_max_alloc_queue_depth;
qmax = qmax * (4 + d) / 4;
qdepth =
zfs_refcount_count(&mgap->mga_alloc_queue_depth);
/*
* If there is another metaslab group that
* might be able to handle the allocation, then
* we return false so that we skip this group.
*/
if (qdepth < qmax && !mgp->mg_no_free_space)
return (B_FALSE);
}
/*
* We didn't find another group to handle the allocation
* so we can't skip this metaslab group even though
* we are at or over our qmax.
*/
return (B_TRUE);
} else if (mc->mc_alloc_groups == 0 || psize == SPA_MINBLOCKSIZE) {
return (B_TRUE);
}
return (B_FALSE);
}
/*
* ==========================================================================
* Range tree callbacks
* ==========================================================================
*/
/*
* Comparison function for the private size-ordered tree using 32-bit
* ranges. Tree is sorted by size, larger sizes at the end of the tree.
*/
static int
metaslab_rangesize32_compare(const void *x1, const void *x2)
{
const range_seg32_t *r1 = x1;
const range_seg32_t *r2 = x2;
uint64_t rs_size1 = r1->rs_end - r1->rs_start;
uint64_t rs_size2 = r2->rs_end - r2->rs_start;
int cmp = TREE_CMP(rs_size1, rs_size2);
if (likely(cmp))
return (cmp);
return (TREE_CMP(r1->rs_start, r2->rs_start));
}
/*
* Comparison function for the private size-ordered tree using 64-bit
* ranges. Tree is sorted by size, larger sizes at the end of the tree.
*/
static int
metaslab_rangesize64_compare(const void *x1, const void *x2)
{
const range_seg64_t *r1 = x1;
const range_seg64_t *r2 = x2;
uint64_t rs_size1 = r1->rs_end - r1->rs_start;
uint64_t rs_size2 = r2->rs_end - r2->rs_start;
int cmp = TREE_CMP(rs_size1, rs_size2);
if (likely(cmp))
return (cmp);
return (TREE_CMP(r1->rs_start, r2->rs_start));
}
typedef struct metaslab_rt_arg {
zfs_btree_t *mra_bt;
uint32_t mra_floor_shift;
} metaslab_rt_arg_t;
struct mssa_arg {
range_tree_t *rt;
metaslab_rt_arg_t *mra;
};
static void
metaslab_size_sorted_add(void *arg, uint64_t start, uint64_t size)
{
struct mssa_arg *mssap = arg;
range_tree_t *rt = mssap->rt;
metaslab_rt_arg_t *mrap = mssap->mra;
range_seg_max_t seg = {0};
rs_set_start(&seg, rt, start);
rs_set_end(&seg, rt, start + size);
metaslab_rt_add(rt, &seg, mrap);
}
static void
metaslab_size_tree_full_load(range_tree_t *rt)
{
metaslab_rt_arg_t *mrap = rt->rt_arg;
METASLABSTAT_BUMP(metaslabstat_reload_tree);
ASSERT0(zfs_btree_numnodes(mrap->mra_bt));
mrap->mra_floor_shift = 0;
struct mssa_arg arg = {0};
arg.rt = rt;
arg.mra = mrap;
range_tree_walk(rt, metaslab_size_sorted_add, &arg);
}
/*
* Create any block allocator specific components. The current allocators
* rely on using both a size-ordered range_tree_t and an array of uint64_t's.
*/
/* ARGSUSED */
static void
metaslab_rt_create(range_tree_t *rt, void *arg)
{
metaslab_rt_arg_t *mrap = arg;
zfs_btree_t *size_tree = mrap->mra_bt;
size_t size;
int (*compare) (const void *, const void *);
switch (rt->rt_type) {
case RANGE_SEG32:
size = sizeof (range_seg32_t);
compare = metaslab_rangesize32_compare;
break;
case RANGE_SEG64:
size = sizeof (range_seg64_t);
compare = metaslab_rangesize64_compare;
break;
default:
panic("Invalid range seg type %d", rt->rt_type);
}
zfs_btree_create(size_tree, compare, size);
mrap->mra_floor_shift = metaslab_by_size_min_shift;
}
/* ARGSUSED */
static void
metaslab_rt_destroy(range_tree_t *rt, void *arg)
{
metaslab_rt_arg_t *mrap = arg;
zfs_btree_t *size_tree = mrap->mra_bt;
zfs_btree_destroy(size_tree);
kmem_free(mrap, sizeof (*mrap));
}
/* ARGSUSED */
static void
metaslab_rt_add(range_tree_t *rt, range_seg_t *rs, void *arg)
{
metaslab_rt_arg_t *mrap = arg;
zfs_btree_t *size_tree = mrap->mra_bt;
if (rs_get_end(rs, rt) - rs_get_start(rs, rt) <
(1 << mrap->mra_floor_shift))
return;
zfs_btree_add(size_tree, rs);
}
/* ARGSUSED */
static void
metaslab_rt_remove(range_tree_t *rt, range_seg_t *rs, void *arg)
{
metaslab_rt_arg_t *mrap = arg;
zfs_btree_t *size_tree = mrap->mra_bt;
if (rs_get_end(rs, rt) - rs_get_start(rs, rt) < (1 <<
mrap->mra_floor_shift))
return;
zfs_btree_remove(size_tree, rs);
}
/* ARGSUSED */
static void
metaslab_rt_vacate(range_tree_t *rt, void *arg)
{
metaslab_rt_arg_t *mrap = arg;
zfs_btree_t *size_tree = mrap->mra_bt;
zfs_btree_clear(size_tree);
zfs_btree_destroy(size_tree);
metaslab_rt_create(rt, arg);
}
static range_tree_ops_t metaslab_rt_ops = {
.rtop_create = metaslab_rt_create,
.rtop_destroy = metaslab_rt_destroy,
.rtop_add = metaslab_rt_add,
.rtop_remove = metaslab_rt_remove,
.rtop_vacate = metaslab_rt_vacate
};
/*
* ==========================================================================
* Common allocator routines
* ==========================================================================
*/
/*
* Return the maximum contiguous segment within the metaslab.
*/
uint64_t
metaslab_largest_allocatable(metaslab_t *msp)
{
zfs_btree_t *t = &msp->ms_allocatable_by_size;
range_seg_t *rs;
if (t == NULL)
return (0);
if (zfs_btree_numnodes(t) == 0)
metaslab_size_tree_full_load(msp->ms_allocatable);
rs = zfs_btree_last(t, NULL);
if (rs == NULL)
return (0);
return (rs_get_end(rs, msp->ms_allocatable) - rs_get_start(rs,
msp->ms_allocatable));
}
/*
* Return the maximum contiguous segment within the unflushed frees of this
* metaslab.
*/
static uint64_t
metaslab_largest_unflushed_free(metaslab_t *msp)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
if (msp->ms_unflushed_frees == NULL)
return (0);
if (zfs_btree_numnodes(&msp->ms_unflushed_frees_by_size) == 0)
metaslab_size_tree_full_load(msp->ms_unflushed_frees);
range_seg_t *rs = zfs_btree_last(&msp->ms_unflushed_frees_by_size,
NULL);
if (rs == NULL)
return (0);
/*
* When a range is freed from the metaslab, that range is added to
* both the unflushed frees and the deferred frees. While the block
* will eventually be usable, if the metaslab were loaded the range
* would not be added to the ms_allocatable tree until TXG_DEFER_SIZE
* txgs had passed. As a result, when attempting to estimate an upper
* bound for the largest currently-usable free segment in the
* metaslab, we need to not consider any ranges currently in the defer
* trees. This algorithm approximates the largest available chunk in
* the largest range in the unflushed_frees tree by taking the first
* chunk. While this may be a poor estimate, it should only remain so
* briefly and should eventually self-correct as frees are no longer
* deferred. Similar logic applies to the ms_freed tree. See
* metaslab_load() for more details.
*
* There are two primary sources of inaccuracy in this estimate. Both
* are tolerated for performance reasons. The first source is that we
* only check the largest segment for overlaps. Smaller segments may
* have more favorable overlaps with the other trees, resulting in
* larger usable chunks. Second, we only look at the first chunk in
* the largest segment; there may be other usable chunks in the
* largest segment, but we ignore them.
*/
uint64_t rstart = rs_get_start(rs, msp->ms_unflushed_frees);
uint64_t rsize = rs_get_end(rs, msp->ms_unflushed_frees) - rstart;
for (int t = 0; t < TXG_DEFER_SIZE; t++) {
uint64_t start = 0;
uint64_t size = 0;
boolean_t found = range_tree_find_in(msp->ms_defer[t], rstart,
rsize, &start, &size);
if (found) {
if (rstart == start)
return (0);
rsize = start - rstart;
}
}
uint64_t start = 0;
uint64_t size = 0;
boolean_t found = range_tree_find_in(msp->ms_freed, rstart,
rsize, &start, &size);
if (found)
rsize = start - rstart;
return (rsize);
}
static range_seg_t *
metaslab_block_find(zfs_btree_t *t, range_tree_t *rt, uint64_t start,
uint64_t size, zfs_btree_index_t *where)
{
range_seg_t *rs;
range_seg_max_t rsearch;
rs_set_start(&rsearch, rt, start);
rs_set_end(&rsearch, rt, start + size);
rs = zfs_btree_find(t, &rsearch, where);
if (rs == NULL) {
rs = zfs_btree_next(t, where, where);
}
return (rs);
}
#if defined(WITH_DF_BLOCK_ALLOCATOR) || \
defined(WITH_CF_BLOCK_ALLOCATOR)
/*
* This is a helper function that can be used by the allocator to find a
* suitable block to allocate. This will search the specified B-tree looking
* for a block that matches the specified criteria.
*/
static uint64_t
metaslab_block_picker(range_tree_t *rt, uint64_t *cursor, uint64_t size,
uint64_t max_search)
{
if (*cursor == 0)
*cursor = rt->rt_start;
zfs_btree_t *bt = &rt->rt_root;
zfs_btree_index_t where;
range_seg_t *rs = metaslab_block_find(bt, rt, *cursor, size, &where);
uint64_t first_found;
int count_searched = 0;
if (rs != NULL)
first_found = rs_get_start(rs, rt);
while (rs != NULL && (rs_get_start(rs, rt) - first_found <=
max_search || count_searched < metaslab_min_search_count)) {
uint64_t offset = rs_get_start(rs, rt);
if (offset + size <= rs_get_end(rs, rt)) {
*cursor = offset + size;
return (offset);
}
rs = zfs_btree_next(bt, &where, &where);
count_searched++;
}
*cursor = 0;
return (-1ULL);
}
#endif /* WITH_DF/CF_BLOCK_ALLOCATOR */
#if defined(WITH_DF_BLOCK_ALLOCATOR)
/*
* ==========================================================================
* Dynamic Fit (df) block allocator
*
* Search for a free chunk of at least this size, starting from the last
* offset (for this alignment of block) looking for up to
* metaslab_df_max_search bytes (16MB). If a large enough free chunk is not
* found within 16MB, then return a free chunk of exactly the requested size (or
* larger).
*
* If it seems like searching from the last offset will be unproductive, skip
* that and just return a free chunk of exactly the requested size (or larger).
* This is based on metaslab_df_alloc_threshold and metaslab_df_free_pct. This
* mechanism is probably not very useful and may be removed in the future.
*
* The behavior when not searching can be changed to return the largest free
* chunk, instead of a free chunk of exactly the requested size, by setting
* metaslab_df_use_largest_segment.
* ==========================================================================
*/
static uint64_t
metaslab_df_alloc(metaslab_t *msp, uint64_t size)
{
/*
* Find the largest power of 2 block size that evenly divides the
* requested size. This is used to try to allocate blocks with similar
* alignment from the same area of the metaslab (i.e. same cursor
* bucket) but it does not guarantee that other allocations sizes
* may exist in the same region.
*/
uint64_t align = size & -size;
uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1];
range_tree_t *rt = msp->ms_allocatable;
int free_pct = range_tree_space(rt) * 100 / msp->ms_size;
uint64_t offset;
ASSERT(MUTEX_HELD(&msp->ms_lock));
/*
* If we're running low on space, find a segment based on size,
* rather than iterating based on offset.
*/
if (metaslab_largest_allocatable(msp) < metaslab_df_alloc_threshold ||
free_pct < metaslab_df_free_pct) {
offset = -1;
} else {
offset = metaslab_block_picker(rt,
cursor, size, metaslab_df_max_search);
}
if (offset == -1) {
range_seg_t *rs;
if (zfs_btree_numnodes(&msp->ms_allocatable_by_size) == 0)
metaslab_size_tree_full_load(msp->ms_allocatable);
if (metaslab_df_use_largest_segment) {
/* use largest free segment */
rs = zfs_btree_last(&msp->ms_allocatable_by_size, NULL);
} else {
zfs_btree_index_t where;
/* use segment of this size, or next largest */
rs = metaslab_block_find(&msp->ms_allocatable_by_size,
rt, msp->ms_start, size, &where);
}
if (rs != NULL && rs_get_start(rs, rt) + size <= rs_get_end(rs,
rt)) {
offset = rs_get_start(rs, rt);
*cursor = offset + size;
}
}
return (offset);
}
static metaslab_ops_t metaslab_df_ops = {
metaslab_df_alloc
};
metaslab_ops_t *zfs_metaslab_ops = &metaslab_df_ops;
#endif /* WITH_DF_BLOCK_ALLOCATOR */
#if defined(WITH_CF_BLOCK_ALLOCATOR)
/*
* ==========================================================================
* Cursor fit block allocator -
* Select the largest region in the metaslab, set the cursor to the beginning
* of the range and the cursor_end to the end of the range. As allocations
* are made advance the cursor. Continue allocating from the cursor until
* the range is exhausted and then find a new range.
* ==========================================================================
*/
static uint64_t
metaslab_cf_alloc(metaslab_t *msp, uint64_t size)
{
range_tree_t *rt = msp->ms_allocatable;
zfs_btree_t *t = &msp->ms_allocatable_by_size;
uint64_t *cursor = &msp->ms_lbas[0];
uint64_t *cursor_end = &msp->ms_lbas[1];
uint64_t offset = 0;
ASSERT(MUTEX_HELD(&msp->ms_lock));
ASSERT3U(*cursor_end, >=, *cursor);
if ((*cursor + size) > *cursor_end) {
range_seg_t *rs;
if (zfs_btree_numnodes(t) == 0)
metaslab_size_tree_full_load(msp->ms_allocatable);
rs = zfs_btree_last(t, NULL);
if (rs == NULL || (rs_get_end(rs, rt) - rs_get_start(rs, rt)) <
size)
return (-1ULL);
*cursor = rs_get_start(rs, rt);
*cursor_end = rs_get_end(rs, rt);
}
offset = *cursor;
*cursor += size;
return (offset);
}
static metaslab_ops_t metaslab_cf_ops = {
metaslab_cf_alloc
};
metaslab_ops_t *zfs_metaslab_ops = &metaslab_cf_ops;
#endif /* WITH_CF_BLOCK_ALLOCATOR */
#if defined(WITH_NDF_BLOCK_ALLOCATOR)
/*
* ==========================================================================
* New dynamic fit allocator -
* Select a region that is large enough to allocate 2^metaslab_ndf_clump_shift
* contiguous blocks. If no region is found then just use the largest segment
* that remains.
* ==========================================================================
*/
/*
* Determines desired number of contiguous blocks (2^metaslab_ndf_clump_shift)
* to request from the allocator.
*/
uint64_t metaslab_ndf_clump_shift = 4;
static uint64_t
metaslab_ndf_alloc(metaslab_t *msp, uint64_t size)
{
zfs_btree_t *t = &msp->ms_allocatable->rt_root;
range_tree_t *rt = msp->ms_allocatable;
zfs_btree_index_t where;
range_seg_t *rs;
range_seg_max_t rsearch;
uint64_t hbit = highbit64(size);
uint64_t *cursor = &msp->ms_lbas[hbit - 1];
uint64_t max_size = metaslab_largest_allocatable(msp);
ASSERT(MUTEX_HELD(&msp->ms_lock));
if (max_size < size)
return (-1ULL);
rs_set_start(&rsearch, rt, *cursor);
rs_set_end(&rsearch, rt, *cursor + size);
rs = zfs_btree_find(t, &rsearch, &where);
if (rs == NULL || (rs_get_end(rs, rt) - rs_get_start(rs, rt)) < size) {
t = &msp->ms_allocatable_by_size;
rs_set_start(&rsearch, rt, 0);
rs_set_end(&rsearch, rt, MIN(max_size, 1ULL << (hbit +
metaslab_ndf_clump_shift)));
rs = zfs_btree_find(t, &rsearch, &where);
if (rs == NULL)
rs = zfs_btree_next(t, &where, &where);
ASSERT(rs != NULL);
}
if ((rs_get_end(rs, rt) - rs_get_start(rs, rt)) >= size) {
*cursor = rs_get_start(rs, rt) + size;
return (rs_get_start(rs, rt));
}
return (-1ULL);
}
static metaslab_ops_t metaslab_ndf_ops = {
metaslab_ndf_alloc
};
metaslab_ops_t *zfs_metaslab_ops = &metaslab_ndf_ops;
#endif /* WITH_NDF_BLOCK_ALLOCATOR */
/*
* ==========================================================================
* Metaslabs
* ==========================================================================
*/
/*
* Wait for any in-progress metaslab loads to complete.
*/
static void
metaslab_load_wait(metaslab_t *msp)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
while (msp->ms_loading) {
ASSERT(!msp->ms_loaded);
cv_wait(&msp->ms_load_cv, &msp->ms_lock);
}
}
/*
* Wait for any in-progress flushing to complete.
*/
static void
metaslab_flush_wait(metaslab_t *msp)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
while (msp->ms_flushing)
cv_wait(&msp->ms_flush_cv, &msp->ms_lock);
}
static unsigned int
metaslab_idx_func(multilist_t *ml, void *arg)
{
metaslab_t *msp = arg;
- return (msp->ms_id % multilist_get_num_sublists(ml));
+
+ /*
+ * ms_id values are allocated sequentially, so full 64bit
+ * division would be a waste of time, so limit it to 32 bits.
+ */
+ return ((unsigned int)msp->ms_id % multilist_get_num_sublists(ml));
}
uint64_t
metaslab_allocated_space(metaslab_t *msp)
{
return (msp->ms_allocated_space);
}
/*
* Verify that the space accounting on disk matches the in-core range_trees.
*/
static void
metaslab_verify_space(metaslab_t *msp, uint64_t txg)
{
spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
uint64_t allocating = 0;
uint64_t sm_free_space, msp_free_space;
ASSERT(MUTEX_HELD(&msp->ms_lock));
ASSERT(!msp->ms_condensing);
if ((zfs_flags & ZFS_DEBUG_METASLAB_VERIFY) == 0)
return;
/*
* We can only verify the metaslab space when we're called
* from syncing context with a loaded metaslab that has an
* allocated space map. Calling this in non-syncing context
* does not provide a consistent view of the metaslab since
* we're performing allocations in the future.
*/
if (txg != spa_syncing_txg(spa) || msp->ms_sm == NULL ||
!msp->ms_loaded)
return;
/*
* Even though the smp_alloc field can get negative,
* when it comes to a metaslab's space map, that should
* never be the case.
*/
ASSERT3S(space_map_allocated(msp->ms_sm), >=, 0);
ASSERT3U(space_map_allocated(msp->ms_sm), >=,
range_tree_space(msp->ms_unflushed_frees));
ASSERT3U(metaslab_allocated_space(msp), ==,
space_map_allocated(msp->ms_sm) +
range_tree_space(msp->ms_unflushed_allocs) -
range_tree_space(msp->ms_unflushed_frees));
sm_free_space = msp->ms_size - metaslab_allocated_space(msp);
/*
* Account for future allocations since we would have
* already deducted that space from the ms_allocatable.
*/
for (int t = 0; t < TXG_CONCURRENT_STATES; t++) {
allocating +=
range_tree_space(msp->ms_allocating[(txg + t) & TXG_MASK]);
}
ASSERT3U(allocating + msp->ms_allocated_this_txg, ==,
msp->ms_allocating_total);
ASSERT3U(msp->ms_deferspace, ==,
range_tree_space(msp->ms_defer[0]) +
range_tree_space(msp->ms_defer[1]));
msp_free_space = range_tree_space(msp->ms_allocatable) + allocating +
msp->ms_deferspace + range_tree_space(msp->ms_freed);
VERIFY3U(sm_free_space, ==, msp_free_space);
}
static void
metaslab_aux_histograms_clear(metaslab_t *msp)
{
/*
* Auxiliary histograms are only cleared when resetting them,
* which can only happen while the metaslab is loaded.
*/
ASSERT(msp->ms_loaded);
bzero(msp->ms_synchist, sizeof (msp->ms_synchist));
for (int t = 0; t < TXG_DEFER_SIZE; t++)
bzero(msp->ms_deferhist[t], sizeof (msp->ms_deferhist[t]));
}
static void
metaslab_aux_histogram_add(uint64_t *histogram, uint64_t shift,
range_tree_t *rt)
{
/*
* This is modeled after space_map_histogram_add(), so refer to that
* function for implementation details. We want this to work like
* the space map histogram, and not the range tree histogram, as we
* are essentially constructing a delta that will be later subtracted
* from the space map histogram.
*/
int idx = 0;
for (int i = shift; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
ASSERT3U(i, >=, idx + shift);
histogram[idx] += rt->rt_histogram[i] << (i - idx - shift);
if (idx < SPACE_MAP_HISTOGRAM_SIZE - 1) {
ASSERT3U(idx + shift, ==, i);
idx++;
ASSERT3U(idx, <, SPACE_MAP_HISTOGRAM_SIZE);
}
}
}
/*
* Called at every sync pass that the metaslab gets synced.
*
* The reason is that we want our auxiliary histograms to be updated
* wherever the metaslab's space map histogram is updated. This way
* we stay consistent on which parts of the metaslab space map's
* histogram are currently not available for allocations (e.g because
* they are in the defer, freed, and freeing trees).
*/
static void
metaslab_aux_histograms_update(metaslab_t *msp)
{
space_map_t *sm = msp->ms_sm;
ASSERT(sm != NULL);
/*
* This is similar to the metaslab's space map histogram updates
* that take place in metaslab_sync(). The only difference is that
* we only care about segments that haven't made it into the
* ms_allocatable tree yet.
*/
if (msp->ms_loaded) {
metaslab_aux_histograms_clear(msp);
metaslab_aux_histogram_add(msp->ms_synchist,
sm->sm_shift, msp->ms_freed);
for (int t = 0; t < TXG_DEFER_SIZE; t++) {
metaslab_aux_histogram_add(msp->ms_deferhist[t],
sm->sm_shift, msp->ms_defer[t]);
}
}
metaslab_aux_histogram_add(msp->ms_synchist,
sm->sm_shift, msp->ms_freeing);
}
/*
* Called every time we are done syncing (writing to) the metaslab,
* i.e. at the end of each sync pass.
* [see the comment in metaslab_impl.h for ms_synchist, ms_deferhist]
*/
static void
metaslab_aux_histograms_update_done(metaslab_t *msp, boolean_t defer_allowed)
{
spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
space_map_t *sm = msp->ms_sm;
if (sm == NULL) {
/*
* We came here from metaslab_init() when creating/opening a
* pool, looking at a metaslab that hasn't had any allocations
* yet.
*/
return;
}
/*
* This is similar to the actions that we take for the ms_freed
* and ms_defer trees in metaslab_sync_done().
*/
uint64_t hist_index = spa_syncing_txg(spa) % TXG_DEFER_SIZE;
if (defer_allowed) {
bcopy(msp->ms_synchist, msp->ms_deferhist[hist_index],
sizeof (msp->ms_synchist));
} else {
bzero(msp->ms_deferhist[hist_index],
sizeof (msp->ms_deferhist[hist_index]));
}
bzero(msp->ms_synchist, sizeof (msp->ms_synchist));
}
/*
* Ensure that the metaslab's weight and fragmentation are consistent
* with the contents of the histogram (either the range tree's histogram
* or the space map's depending whether the metaslab is loaded).
*/
static void
metaslab_verify_weight_and_frag(metaslab_t *msp)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
if ((zfs_flags & ZFS_DEBUG_METASLAB_VERIFY) == 0)
return;
/*
* We can end up here from vdev_remove_complete(), in which case we
* cannot do these assertions because we hold spa config locks and
* thus we are not allowed to read from the DMU.
*
* We check if the metaslab group has been removed and if that's
* the case we return immediately as that would mean that we are
* here from the aforementioned code path.
*/
if (msp->ms_group == NULL)
return;
/*
* Devices being removed always return a weight of 0 and leave
* fragmentation and ms_max_size as is - there is nothing for
* us to verify here.
*/
vdev_t *vd = msp->ms_group->mg_vd;
if (vd->vdev_removing)
return;
/*
* If the metaslab is dirty it probably means that we've done
* some allocations or frees that have changed our histograms
* and thus the weight.
*/
for (int t = 0; t < TXG_SIZE; t++) {
if (txg_list_member(&vd->vdev_ms_list, msp, t))
return;
}
/*
* This verification checks that our in-memory state is consistent
* with what's on disk. If the pool is read-only then there aren't
* any changes and we just have the initially-loaded state.
*/
if (!spa_writeable(msp->ms_group->mg_vd->vdev_spa))
return;
/* some extra verification for in-core tree if you can */
if (msp->ms_loaded) {
range_tree_stat_verify(msp->ms_allocatable);
VERIFY(space_map_histogram_verify(msp->ms_sm,
msp->ms_allocatable));
}
uint64_t weight = msp->ms_weight;
uint64_t was_active = msp->ms_weight & METASLAB_ACTIVE_MASK;
boolean_t space_based = WEIGHT_IS_SPACEBASED(msp->ms_weight);
uint64_t frag = msp->ms_fragmentation;
uint64_t max_segsize = msp->ms_max_size;
msp->ms_weight = 0;
msp->ms_fragmentation = 0;
/*
* This function is used for verification purposes and thus should
* not introduce any side-effects/mutations on the system's state.
*
* Regardless of whether metaslab_weight() thinks this metaslab
* should be active or not, we want to ensure that the actual weight
* (and therefore the value of ms_weight) would be the same if it
* was to be recalculated at this point.
*
* In addition we set the nodirty flag so metaslab_weight() does
* not dirty the metaslab for future TXGs (e.g. when trying to
* force condensing to upgrade the metaslab spacemaps).
*/
msp->ms_weight = metaslab_weight(msp, B_TRUE) | was_active;
VERIFY3U(max_segsize, ==, msp->ms_max_size);
/*
* If the weight type changed then there is no point in doing
* verification. Revert fields to their original values.
*/
if ((space_based && !WEIGHT_IS_SPACEBASED(msp->ms_weight)) ||
(!space_based && WEIGHT_IS_SPACEBASED(msp->ms_weight))) {
msp->ms_fragmentation = frag;
msp->ms_weight = weight;
return;
}
VERIFY3U(msp->ms_fragmentation, ==, frag);
VERIFY3U(msp->ms_weight, ==, weight);
}
/*
* If we're over the zfs_metaslab_mem_limit, select the loaded metaslab from
* this class that was used longest ago, and attempt to unload it. We don't
* want to spend too much time in this loop to prevent performance
* degradation, and we expect that most of the time this operation will
* succeed. Between that and the normal unloading processing during txg sync,
* we expect this to keep the metaslab memory usage under control.
*/
static void
metaslab_potentially_evict(metaslab_class_t *mc)
{
#ifdef _KERNEL
uint64_t allmem = arc_all_memory();
uint64_t inuse = spl_kmem_cache_inuse(zfs_btree_leaf_cache);
uint64_t size = spl_kmem_cache_entry_size(zfs_btree_leaf_cache);
int tries = 0;
for (; allmem * zfs_metaslab_mem_limit / 100 < inuse * size &&
tries < multilist_get_num_sublists(&mc->mc_metaslab_txg_list) * 2;
tries++) {
unsigned int idx = multilist_get_random_index(
&mc->mc_metaslab_txg_list);
multilist_sublist_t *mls =
multilist_sublist_lock(&mc->mc_metaslab_txg_list, idx);
metaslab_t *msp = multilist_sublist_head(mls);
multilist_sublist_unlock(mls);
while (msp != NULL && allmem * zfs_metaslab_mem_limit / 100 <
inuse * size) {
VERIFY3P(mls, ==, multilist_sublist_lock(
&mc->mc_metaslab_txg_list, idx));
ASSERT3U(idx, ==,
metaslab_idx_func(&mc->mc_metaslab_txg_list, msp));
if (!multilist_link_active(&msp->ms_class_txg_node)) {
multilist_sublist_unlock(mls);
break;
}
metaslab_t *next_msp = multilist_sublist_next(mls, msp);
multilist_sublist_unlock(mls);
/*
* If the metaslab is currently loading there are two
* cases. If it's the metaslab we're evicting, we
* can't continue on or we'll panic when we attempt to
* recursively lock the mutex. If it's another
* metaslab that's loading, it can be safely skipped,
* since we know it's very new and therefore not a
* good eviction candidate. We check later once the
* lock is held that the metaslab is fully loaded
* before actually unloading it.
*/
if (msp->ms_loading) {
msp = next_msp;
inuse =
spl_kmem_cache_inuse(zfs_btree_leaf_cache);
continue;
}
/*
* We can't unload metaslabs with no spacemap because
* they're not ready to be unloaded yet. We can't
* unload metaslabs with outstanding allocations
* because doing so could cause the metaslab's weight
* to decrease while it's unloaded, which violates an
* invariant that we use to prevent unnecessary
* loading. We also don't unload metaslabs that are
* currently active because they are high-weight
* metaslabs that are likely to be used in the near
* future.
*/
mutex_enter(&msp->ms_lock);
if (msp->ms_allocator == -1 && msp->ms_sm != NULL &&
msp->ms_allocating_total == 0) {
metaslab_unload(msp);
}
mutex_exit(&msp->ms_lock);
msp = next_msp;
inuse = spl_kmem_cache_inuse(zfs_btree_leaf_cache);
}
}
#endif
}
static int
metaslab_load_impl(metaslab_t *msp)
{
int error = 0;
ASSERT(MUTEX_HELD(&msp->ms_lock));
ASSERT(msp->ms_loading);
ASSERT(!msp->ms_condensing);
/*
* We temporarily drop the lock to unblock other operations while we
* are reading the space map. Therefore, metaslab_sync() and
* metaslab_sync_done() can run at the same time as we do.
*
* If we are using the log space maps, metaslab_sync() can't write to
* the metaslab's space map while we are loading as we only write to
* it when we are flushing the metaslab, and that can't happen while
* we are loading it.
*
* If we are not using log space maps though, metaslab_sync() can
* append to the space map while we are loading. Therefore we load
* only entries that existed when we started the load. Additionally,
* metaslab_sync_done() has to wait for the load to complete because
* there are potential races like metaslab_load() loading parts of the
* space map that are currently being appended by metaslab_sync(). If
* we didn't, the ms_allocatable would have entries that
* metaslab_sync_done() would try to re-add later.
*
* That's why before dropping the lock we remember the synced length
* of the metaslab and read up to that point of the space map,
* ignoring entries appended by metaslab_sync() that happen after we
* drop the lock.
*/
uint64_t length = msp->ms_synced_length;
mutex_exit(&msp->ms_lock);
hrtime_t load_start = gethrtime();
metaslab_rt_arg_t *mrap;
if (msp->ms_allocatable->rt_arg == NULL) {
mrap = kmem_zalloc(sizeof (*mrap), KM_SLEEP);
} else {
mrap = msp->ms_allocatable->rt_arg;
msp->ms_allocatable->rt_ops = NULL;
msp->ms_allocatable->rt_arg = NULL;
}
mrap->mra_bt = &msp->ms_allocatable_by_size;
mrap->mra_floor_shift = metaslab_by_size_min_shift;
if (msp->ms_sm != NULL) {
error = space_map_load_length(msp->ms_sm, msp->ms_allocatable,
SM_FREE, length);
/* Now, populate the size-sorted tree. */
metaslab_rt_create(msp->ms_allocatable, mrap);
msp->ms_allocatable->rt_ops = &metaslab_rt_ops;
msp->ms_allocatable->rt_arg = mrap;
struct mssa_arg arg = {0};
arg.rt = msp->ms_allocatable;
arg.mra = mrap;
range_tree_walk(msp->ms_allocatable, metaslab_size_sorted_add,
&arg);
} else {
/*
* Add the size-sorted tree first, since we don't need to load
* the metaslab from the spacemap.
*/
metaslab_rt_create(msp->ms_allocatable, mrap);
msp->ms_allocatable->rt_ops = &metaslab_rt_ops;
msp->ms_allocatable->rt_arg = mrap;
/*
* The space map has not been allocated yet, so treat
* all the space in the metaslab as free and add it to the
* ms_allocatable tree.
*/
range_tree_add(msp->ms_allocatable,
msp->ms_start, msp->ms_size);
if (msp->ms_new) {
/*
* If the ms_sm doesn't exist, this means that this
* metaslab hasn't gone through metaslab_sync() and
* thus has never been dirtied. So we shouldn't
* expect any unflushed allocs or frees from previous
* TXGs.
*/
ASSERT(range_tree_is_empty(msp->ms_unflushed_allocs));
ASSERT(range_tree_is_empty(msp->ms_unflushed_frees));
}
}
/*
* We need to grab the ms_sync_lock to prevent metaslab_sync() from
* changing the ms_sm (or log_sm) and the metaslab's range trees
* while we are about to use them and populate the ms_allocatable.
* The ms_lock is insufficient for this because metaslab_sync() doesn't
* hold the ms_lock while writing the ms_checkpointing tree to disk.
*/
mutex_enter(&msp->ms_sync_lock);
mutex_enter(&msp->ms_lock);
ASSERT(!msp->ms_condensing);
ASSERT(!msp->ms_flushing);
if (error != 0) {
mutex_exit(&msp->ms_sync_lock);
return (error);
}
ASSERT3P(msp->ms_group, !=, NULL);
msp->ms_loaded = B_TRUE;
/*
* Apply all the unflushed changes to ms_allocatable right
* away so any manipulations we do below have a clear view
* of what is allocated and what is free.
*/
range_tree_walk(msp->ms_unflushed_allocs,
range_tree_remove, msp->ms_allocatable);
range_tree_walk(msp->ms_unflushed_frees,
range_tree_add, msp->ms_allocatable);
ASSERT3P(msp->ms_group, !=, NULL);
spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
if (spa_syncing_log_sm(spa) != NULL) {
ASSERT(spa_feature_is_enabled(spa,
SPA_FEATURE_LOG_SPACEMAP));
/*
* If we use a log space map we add all the segments
* that are in ms_unflushed_frees so they are available
* for allocation.
*
* ms_allocatable needs to contain all free segments
* that are ready for allocations (thus not segments
* from ms_freeing, ms_freed, and the ms_defer trees).
* But if we grab the lock in this code path at a sync
* pass later that 1, then it also contains the
* segments of ms_freed (they were added to it earlier
* in this path through ms_unflushed_frees). So we
* need to remove all the segments that exist in
* ms_freed from ms_allocatable as they will be added
* later in metaslab_sync_done().
*
* When there's no log space map, the ms_allocatable
* correctly doesn't contain any segments that exist
* in ms_freed [see ms_synced_length].
*/
range_tree_walk(msp->ms_freed,
range_tree_remove, msp->ms_allocatable);
}
/*
* If we are not using the log space map, ms_allocatable
* contains the segments that exist in the ms_defer trees
* [see ms_synced_length]. Thus we need to remove them
* from ms_allocatable as they will be added again in
* metaslab_sync_done().
*
* If we are using the log space map, ms_allocatable still
* contains the segments that exist in the ms_defer trees.
* Not because it read them through the ms_sm though. But
* because these segments are part of ms_unflushed_frees
* whose segments we add to ms_allocatable earlier in this
* code path.
*/
for (int t = 0; t < TXG_DEFER_SIZE; t++) {
range_tree_walk(msp->ms_defer[t],
range_tree_remove, msp->ms_allocatable);
}
/*
* Call metaslab_recalculate_weight_and_sort() now that the
* metaslab is loaded so we get the metaslab's real weight.
*
* Unless this metaslab was created with older software and
* has not yet been converted to use segment-based weight, we
* expect the new weight to be better or equal to the weight
* that the metaslab had while it was not loaded. This is
* because the old weight does not take into account the
* consolidation of adjacent segments between TXGs. [see
* comment for ms_synchist and ms_deferhist[] for more info]
*/
uint64_t weight = msp->ms_weight;
uint64_t max_size = msp->ms_max_size;
metaslab_recalculate_weight_and_sort(msp);
if (!WEIGHT_IS_SPACEBASED(weight))
ASSERT3U(weight, <=, msp->ms_weight);
msp->ms_max_size = metaslab_largest_allocatable(msp);
ASSERT3U(max_size, <=, msp->ms_max_size);
hrtime_t load_end = gethrtime();
msp->ms_load_time = load_end;
zfs_dbgmsg("metaslab_load: txg %llu, spa %s, vdev_id %llu, "
"ms_id %llu, smp_length %llu, "
"unflushed_allocs %llu, unflushed_frees %llu, "
"freed %llu, defer %llu + %llu, unloaded time %llu ms, "
"loading_time %lld ms, ms_max_size %llu, "
"max size error %lld, "
"old_weight %llx, new_weight %llx",
(u_longlong_t)spa_syncing_txg(spa), spa_name(spa),
(u_longlong_t)msp->ms_group->mg_vd->vdev_id,
(u_longlong_t)msp->ms_id,
(u_longlong_t)space_map_length(msp->ms_sm),
(u_longlong_t)range_tree_space(msp->ms_unflushed_allocs),
(u_longlong_t)range_tree_space(msp->ms_unflushed_frees),
(u_longlong_t)range_tree_space(msp->ms_freed),
(u_longlong_t)range_tree_space(msp->ms_defer[0]),
(u_longlong_t)range_tree_space(msp->ms_defer[1]),
(longlong_t)((load_start - msp->ms_unload_time) / 1000000),
(longlong_t)((load_end - load_start) / 1000000),
(u_longlong_t)msp->ms_max_size,
(u_longlong_t)msp->ms_max_size - max_size,
(u_longlong_t)weight, (u_longlong_t)msp->ms_weight);
metaslab_verify_space(msp, spa_syncing_txg(spa));
mutex_exit(&msp->ms_sync_lock);
return (0);
}
int
metaslab_load(metaslab_t *msp)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
/*
* There may be another thread loading the same metaslab, if that's
* the case just wait until the other thread is done and return.
*/
metaslab_load_wait(msp);
if (msp->ms_loaded)
return (0);
VERIFY(!msp->ms_loading);
ASSERT(!msp->ms_condensing);
/*
* We set the loading flag BEFORE potentially dropping the lock to
* wait for an ongoing flush (see ms_flushing below). This way other
* threads know that there is already a thread that is loading this
* metaslab.
*/
msp->ms_loading = B_TRUE;
/*
* Wait for any in-progress flushing to finish as we drop the ms_lock
* both here (during space_map_load()) and in metaslab_flush() (when
* we flush our changes to the ms_sm).
*/
if (msp->ms_flushing)
metaslab_flush_wait(msp);
/*
* In the possibility that we were waiting for the metaslab to be
* flushed (where we temporarily dropped the ms_lock), ensure that
* no one else loaded the metaslab somehow.
*/
ASSERT(!msp->ms_loaded);
/*
* If we're loading a metaslab in the normal class, consider evicting
* another one to keep our memory usage under the limit defined by the
* zfs_metaslab_mem_limit tunable.
*/
if (spa_normal_class(msp->ms_group->mg_class->mc_spa) ==
msp->ms_group->mg_class) {
metaslab_potentially_evict(msp->ms_group->mg_class);
}
int error = metaslab_load_impl(msp);
ASSERT(MUTEX_HELD(&msp->ms_lock));
msp->ms_loading = B_FALSE;
cv_broadcast(&msp->ms_load_cv);
return (error);
}
void
metaslab_unload(metaslab_t *msp)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
/*
* This can happen if a metaslab is selected for eviction (in
* metaslab_potentially_evict) and then unloaded during spa_sync (via
* metaslab_class_evict_old).
*/
if (!msp->ms_loaded)
return;
range_tree_vacate(msp->ms_allocatable, NULL, NULL);
msp->ms_loaded = B_FALSE;
msp->ms_unload_time = gethrtime();
msp->ms_activation_weight = 0;
msp->ms_weight &= ~METASLAB_ACTIVE_MASK;
if (msp->ms_group != NULL) {
metaslab_class_t *mc = msp->ms_group->mg_class;
multilist_sublist_t *mls =
multilist_sublist_lock_obj(&mc->mc_metaslab_txg_list, msp);
if (multilist_link_active(&msp->ms_class_txg_node))
multilist_sublist_remove(mls, msp);
multilist_sublist_unlock(mls);
spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
zfs_dbgmsg("metaslab_unload: txg %llu, spa %s, vdev_id %llu, "
"ms_id %llu, weight %llx, "
"selected txg %llu (%llu ms ago), alloc_txg %llu, "
"loaded %llu ms ago, max_size %llu",
(u_longlong_t)spa_syncing_txg(spa), spa_name(spa),
(u_longlong_t)msp->ms_group->mg_vd->vdev_id,
(u_longlong_t)msp->ms_id,
(u_longlong_t)msp->ms_weight,
(u_longlong_t)msp->ms_selected_txg,
(u_longlong_t)(msp->ms_unload_time -
msp->ms_selected_time) / 1000 / 1000,
(u_longlong_t)msp->ms_alloc_txg,
(u_longlong_t)(msp->ms_unload_time -
msp->ms_load_time) / 1000 / 1000,
(u_longlong_t)msp->ms_max_size);
}
/*
* We explicitly recalculate the metaslab's weight based on its space
* map (as it is now not loaded). We want unload metaslabs to always
* have their weights calculated from the space map histograms, while
* loaded ones have it calculated from their in-core range tree
* [see metaslab_load()]. This way, the weight reflects the information
* available in-core, whether it is loaded or not.
*
* If ms_group == NULL means that we came here from metaslab_fini(),
* at which point it doesn't make sense for us to do the recalculation
* and the sorting.
*/
if (msp->ms_group != NULL)
metaslab_recalculate_weight_and_sort(msp);
}
/*
* We want to optimize the memory use of the per-metaslab range
* trees. To do this, we store the segments in the range trees in
* units of sectors, zero-indexing from the start of the metaslab. If
* the vdev_ms_shift - the vdev_ashift is less than 32, we can store
* the ranges using two uint32_ts, rather than two uint64_ts.
*/
range_seg_type_t
metaslab_calculate_range_tree_type(vdev_t *vdev, metaslab_t *msp,
uint64_t *start, uint64_t *shift)
{
if (vdev->vdev_ms_shift - vdev->vdev_ashift < 32 &&
!zfs_metaslab_force_large_segs) {
*shift = vdev->vdev_ashift;
*start = msp->ms_start;
return (RANGE_SEG32);
} else {
*shift = 0;
*start = 0;
return (RANGE_SEG64);
}
}
void
metaslab_set_selected_txg(metaslab_t *msp, uint64_t txg)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
metaslab_class_t *mc = msp->ms_group->mg_class;
multilist_sublist_t *mls =
multilist_sublist_lock_obj(&mc->mc_metaslab_txg_list, msp);
if (multilist_link_active(&msp->ms_class_txg_node))
multilist_sublist_remove(mls, msp);
msp->ms_selected_txg = txg;
msp->ms_selected_time = gethrtime();
multilist_sublist_insert_tail(mls, msp);
multilist_sublist_unlock(mls);
}
void
metaslab_space_update(vdev_t *vd, metaslab_class_t *mc, int64_t alloc_delta,
int64_t defer_delta, int64_t space_delta)
{
vdev_space_update(vd, alloc_delta, defer_delta, space_delta);
ASSERT3P(vd->vdev_spa->spa_root_vdev, ==, vd->vdev_parent);
ASSERT(vd->vdev_ms_count != 0);
metaslab_class_space_update(mc, alloc_delta, defer_delta, space_delta,
vdev_deflated_space(vd, space_delta));
}
int
metaslab_init(metaslab_group_t *mg, uint64_t id, uint64_t object,
uint64_t txg, metaslab_t **msp)
{
vdev_t *vd = mg->mg_vd;
spa_t *spa = vd->vdev_spa;
objset_t *mos = spa->spa_meta_objset;
metaslab_t *ms;
int error;
ms = kmem_zalloc(sizeof (metaslab_t), KM_SLEEP);
mutex_init(&ms->ms_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&ms->ms_sync_lock, NULL, MUTEX_DEFAULT, NULL);
cv_init(&ms->ms_load_cv, NULL, CV_DEFAULT, NULL);
cv_init(&ms->ms_flush_cv, NULL, CV_DEFAULT, NULL);
multilist_link_init(&ms->ms_class_txg_node);
ms->ms_id = id;
ms->ms_start = id << vd->vdev_ms_shift;
ms->ms_size = 1ULL << vd->vdev_ms_shift;
ms->ms_allocator = -1;
ms->ms_new = B_TRUE;
vdev_ops_t *ops = vd->vdev_ops;
if (ops->vdev_op_metaslab_init != NULL)
ops->vdev_op_metaslab_init(vd, &ms->ms_start, &ms->ms_size);
/*
* We only open space map objects that already exist. All others
* will be opened when we finally allocate an object for it.
*
* Note:
* When called from vdev_expand(), we can't call into the DMU as
* we are holding the spa_config_lock as a writer and we would
* deadlock [see relevant comment in vdev_metaslab_init()]. in
* that case, the object parameter is zero though, so we won't
* call into the DMU.
*/
if (object != 0) {
error = space_map_open(&ms->ms_sm, mos, object, ms->ms_start,
ms->ms_size, vd->vdev_ashift);
if (error != 0) {
kmem_free(ms, sizeof (metaslab_t));
return (error);
}
ASSERT(ms->ms_sm != NULL);
ms->ms_allocated_space = space_map_allocated(ms->ms_sm);
}
uint64_t shift, start;
range_seg_type_t type =
metaslab_calculate_range_tree_type(vd, ms, &start, &shift);
ms->ms_allocatable = range_tree_create(NULL, type, NULL, start, shift);
for (int t = 0; t < TXG_SIZE; t++) {
ms->ms_allocating[t] = range_tree_create(NULL, type,
NULL, start, shift);
}
ms->ms_freeing = range_tree_create(NULL, type, NULL, start, shift);
ms->ms_freed = range_tree_create(NULL, type, NULL, start, shift);
for (int t = 0; t < TXG_DEFER_SIZE; t++) {
ms->ms_defer[t] = range_tree_create(NULL, type, NULL,
start, shift);
}
ms->ms_checkpointing =
range_tree_create(NULL, type, NULL, start, shift);
ms->ms_unflushed_allocs =
range_tree_create(NULL, type, NULL, start, shift);
metaslab_rt_arg_t *mrap = kmem_zalloc(sizeof (*mrap), KM_SLEEP);
mrap->mra_bt = &ms->ms_unflushed_frees_by_size;
mrap->mra_floor_shift = metaslab_by_size_min_shift;
ms->ms_unflushed_frees = range_tree_create(&metaslab_rt_ops,
type, mrap, start, shift);
ms->ms_trim = range_tree_create(NULL, type, NULL, start, shift);
metaslab_group_add(mg, ms);
metaslab_set_fragmentation(ms, B_FALSE);
/*
* If we're opening an existing pool (txg == 0) or creating
* a new one (txg == TXG_INITIAL), all space is available now.
* If we're adding space to an existing pool, the new space
* does not become available until after this txg has synced.
* The metaslab's weight will also be initialized when we sync
* out this txg. This ensures that we don't attempt to allocate
* from it before we have initialized it completely.
*/
if (txg <= TXG_INITIAL) {
metaslab_sync_done(ms, 0);
metaslab_space_update(vd, mg->mg_class,
metaslab_allocated_space(ms), 0, 0);
}
if (txg != 0) {
vdev_dirty(vd, 0, NULL, txg);
vdev_dirty(vd, VDD_METASLAB, ms, txg);
}
*msp = ms;
return (0);
}
static void
metaslab_fini_flush_data(metaslab_t *msp)
{
spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
if (metaslab_unflushed_txg(msp) == 0) {
ASSERT3P(avl_find(&spa->spa_metaslabs_by_flushed, msp, NULL),
==, NULL);
return;
}
ASSERT(spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP));
mutex_enter(&spa->spa_flushed_ms_lock);
avl_remove(&spa->spa_metaslabs_by_flushed, msp);
mutex_exit(&spa->spa_flushed_ms_lock);
spa_log_sm_decrement_mscount(spa, metaslab_unflushed_txg(msp));
spa_log_summary_decrement_mscount(spa, metaslab_unflushed_txg(msp));
}
uint64_t
metaslab_unflushed_changes_memused(metaslab_t *ms)
{
return ((range_tree_numsegs(ms->ms_unflushed_allocs) +
range_tree_numsegs(ms->ms_unflushed_frees)) *
ms->ms_unflushed_allocs->rt_root.bt_elem_size);
}
void
metaslab_fini(metaslab_t *msp)
{
metaslab_group_t *mg = msp->ms_group;
vdev_t *vd = mg->mg_vd;
spa_t *spa = vd->vdev_spa;
metaslab_fini_flush_data(msp);
metaslab_group_remove(mg, msp);
mutex_enter(&msp->ms_lock);
VERIFY(msp->ms_group == NULL);
/*
* If this metaslab hasn't been through metaslab_sync_done() yet its
* space hasn't been accounted for in its vdev and doesn't need to be
* subtracted.
*/
if (!msp->ms_new) {
metaslab_space_update(vd, mg->mg_class,
-metaslab_allocated_space(msp), 0, -msp->ms_size);
}
space_map_close(msp->ms_sm);
msp->ms_sm = NULL;
metaslab_unload(msp);
range_tree_destroy(msp->ms_allocatable);
range_tree_destroy(msp->ms_freeing);
range_tree_destroy(msp->ms_freed);
ASSERT3U(spa->spa_unflushed_stats.sus_memused, >=,
metaslab_unflushed_changes_memused(msp));
spa->spa_unflushed_stats.sus_memused -=
metaslab_unflushed_changes_memused(msp);
range_tree_vacate(msp->ms_unflushed_allocs, NULL, NULL);
range_tree_destroy(msp->ms_unflushed_allocs);
range_tree_destroy(msp->ms_checkpointing);
range_tree_vacate(msp->ms_unflushed_frees, NULL, NULL);
range_tree_destroy(msp->ms_unflushed_frees);
for (int t = 0; t < TXG_SIZE; t++) {
range_tree_destroy(msp->ms_allocating[t]);
}
for (int t = 0; t < TXG_DEFER_SIZE; t++) {
range_tree_destroy(msp->ms_defer[t]);
}
ASSERT0(msp->ms_deferspace);
for (int t = 0; t < TXG_SIZE; t++)
ASSERT(!txg_list_member(&vd->vdev_ms_list, msp, t));
range_tree_vacate(msp->ms_trim, NULL, NULL);
range_tree_destroy(msp->ms_trim);
mutex_exit(&msp->ms_lock);
cv_destroy(&msp->ms_load_cv);
cv_destroy(&msp->ms_flush_cv);
mutex_destroy(&msp->ms_lock);
mutex_destroy(&msp->ms_sync_lock);
ASSERT3U(msp->ms_allocator, ==, -1);
kmem_free(msp, sizeof (metaslab_t));
}
#define FRAGMENTATION_TABLE_SIZE 17
/*
* This table defines a segment size based fragmentation metric that will
* allow each metaslab to derive its own fragmentation value. This is done
* by calculating the space in each bucket of the spacemap histogram and
* multiplying that by the fragmentation metric in this table. Doing
* this for all buckets and dividing it by the total amount of free
* space in this metaslab (i.e. the total free space in all buckets) gives
* us the fragmentation metric. This means that a high fragmentation metric
* equates to most of the free space being comprised of small segments.
* Conversely, if the metric is low, then most of the free space is in
* large segments. A 10% change in fragmentation equates to approximately
* double the number of segments.
*
* This table defines 0% fragmented space using 16MB segments. Testing has
* shown that segments that are greater than or equal to 16MB do not suffer
* from drastic performance problems. Using this value, we derive the rest
* of the table. Since the fragmentation value is never stored on disk, it
* is possible to change these calculations in the future.
*/
int zfs_frag_table[FRAGMENTATION_TABLE_SIZE] = {
100, /* 512B */
100, /* 1K */
98, /* 2K */
95, /* 4K */
90, /* 8K */
80, /* 16K */
70, /* 32K */
60, /* 64K */
50, /* 128K */
40, /* 256K */
30, /* 512K */
20, /* 1M */
15, /* 2M */
10, /* 4M */
5, /* 8M */
0 /* 16M */
};
/*
* Calculate the metaslab's fragmentation metric and set ms_fragmentation.
* Setting this value to ZFS_FRAG_INVALID means that the metaslab has not
* been upgraded and does not support this metric. Otherwise, the return
* value should be in the range [0, 100].
*/
static void
metaslab_set_fragmentation(metaslab_t *msp, boolean_t nodirty)
{
spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
uint64_t fragmentation = 0;
uint64_t total = 0;
boolean_t feature_enabled = spa_feature_is_enabled(spa,
SPA_FEATURE_SPACEMAP_HISTOGRAM);
if (!feature_enabled) {
msp->ms_fragmentation = ZFS_FRAG_INVALID;
return;
}
/*
* A null space map means that the entire metaslab is free
* and thus is not fragmented.
*/
if (msp->ms_sm == NULL) {
msp->ms_fragmentation = 0;
return;
}
/*
* If this metaslab's space map has not been upgraded, flag it
* so that we upgrade next time we encounter it.
*/
if (msp->ms_sm->sm_dbuf->db_size != sizeof (space_map_phys_t)) {
uint64_t txg = spa_syncing_txg(spa);
vdev_t *vd = msp->ms_group->mg_vd;
/*
* If we've reached the final dirty txg, then we must
* be shutting down the pool. We don't want to dirty
* any data past this point so skip setting the condense
* flag. We can retry this action the next time the pool
* is imported. We also skip marking this metaslab for
* condensing if the caller has explicitly set nodirty.
*/
if (!nodirty &&
spa_writeable(spa) && txg < spa_final_dirty_txg(spa)) {
msp->ms_condense_wanted = B_TRUE;
vdev_dirty(vd, VDD_METASLAB, msp, txg + 1);
zfs_dbgmsg("txg %llu, requesting force condense: "
"ms_id %llu, vdev_id %llu", (u_longlong_t)txg,
(u_longlong_t)msp->ms_id,
(u_longlong_t)vd->vdev_id);
}
msp->ms_fragmentation = ZFS_FRAG_INVALID;
return;
}
for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
uint64_t space = 0;
uint8_t shift = msp->ms_sm->sm_shift;
int idx = MIN(shift - SPA_MINBLOCKSHIFT + i,
FRAGMENTATION_TABLE_SIZE - 1);
if (msp->ms_sm->sm_phys->smp_histogram[i] == 0)
continue;
space = msp->ms_sm->sm_phys->smp_histogram[i] << (i + shift);
total += space;
ASSERT3U(idx, <, FRAGMENTATION_TABLE_SIZE);
fragmentation += space * zfs_frag_table[idx];
}
if (total > 0)
fragmentation /= total;
ASSERT3U(fragmentation, <=, 100);
msp->ms_fragmentation = fragmentation;
}
/*
* Compute a weight -- a selection preference value -- for the given metaslab.
* This is based on the amount of free space, the level of fragmentation,
* the LBA range, and whether the metaslab is loaded.
*/
static uint64_t
metaslab_space_weight(metaslab_t *msp)
{
metaslab_group_t *mg = msp->ms_group;
vdev_t *vd = mg->mg_vd;
uint64_t weight, space;
ASSERT(MUTEX_HELD(&msp->ms_lock));
/*
* The baseline weight is the metaslab's free space.
*/
space = msp->ms_size - metaslab_allocated_space(msp);
if (metaslab_fragmentation_factor_enabled &&
msp->ms_fragmentation != ZFS_FRAG_INVALID) {
/*
* Use the fragmentation information to inversely scale
* down the baseline weight. We need to ensure that we
* don't exclude this metaslab completely when it's 100%
* fragmented. To avoid this we reduce the fragmented value
* by 1.
*/
space = (space * (100 - (msp->ms_fragmentation - 1))) / 100;
/*
* If space < SPA_MINBLOCKSIZE, then we will not allocate from
* this metaslab again. The fragmentation metric may have
* decreased the space to something smaller than
* SPA_MINBLOCKSIZE, so reset the space to SPA_MINBLOCKSIZE
* so that we can consume any remaining space.
*/
if (space > 0 && space < SPA_MINBLOCKSIZE)
space = SPA_MINBLOCKSIZE;
}
weight = space;
/*
* Modern disks have uniform bit density and constant angular velocity.
* Therefore, the outer recording zones are faster (higher bandwidth)
* than the inner zones by the ratio of outer to inner track diameter,
* which is typically around 2:1. We account for this by assigning
* higher weight to lower metaslabs (multiplier ranging from 2x to 1x).
* In effect, this means that we'll select the metaslab with the most
* free bandwidth rather than simply the one with the most free space.
*/
if (!vd->vdev_nonrot && metaslab_lba_weighting_enabled) {
weight = 2 * weight - (msp->ms_id * weight) / vd->vdev_ms_count;
ASSERT(weight >= space && weight <= 2 * space);
}
/*
* If this metaslab is one we're actively using, adjust its
* weight to make it preferable to any inactive metaslab so
* we'll polish it off. If the fragmentation on this metaslab
* has exceed our threshold, then don't mark it active.
*/
if (msp->ms_loaded && msp->ms_fragmentation != ZFS_FRAG_INVALID &&
msp->ms_fragmentation <= zfs_metaslab_fragmentation_threshold) {
weight |= (msp->ms_weight & METASLAB_ACTIVE_MASK);
}
WEIGHT_SET_SPACEBASED(weight);
return (weight);
}
/*
* Return the weight of the specified metaslab, according to the segment-based
* weighting algorithm. The metaslab must be loaded. This function can
* be called within a sync pass since it relies only on the metaslab's
* range tree which is always accurate when the metaslab is loaded.
*/
static uint64_t
metaslab_weight_from_range_tree(metaslab_t *msp)
{
uint64_t weight = 0;
uint32_t segments = 0;
ASSERT(msp->ms_loaded);
for (int i = RANGE_TREE_HISTOGRAM_SIZE - 1; i >= SPA_MINBLOCKSHIFT;
i--) {
uint8_t shift = msp->ms_group->mg_vd->vdev_ashift;
int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1;
segments <<= 1;
segments += msp->ms_allocatable->rt_histogram[i];
/*
* The range tree provides more precision than the space map
* and must be downgraded so that all values fit within the
* space map's histogram. This allows us to compare loaded
* vs. unloaded metaslabs to determine which metaslab is
* considered "best".
*/
if (i > max_idx)
continue;
if (segments != 0) {
WEIGHT_SET_COUNT(weight, segments);
WEIGHT_SET_INDEX(weight, i);
WEIGHT_SET_ACTIVE(weight, 0);
break;
}
}
return (weight);
}
/*
* Calculate the weight based on the on-disk histogram. Should be applied
* only to unloaded metaslabs (i.e no incoming allocations) in-order to
* give results consistent with the on-disk state
*/
static uint64_t
metaslab_weight_from_spacemap(metaslab_t *msp)
{
space_map_t *sm = msp->ms_sm;
ASSERT(!msp->ms_loaded);
ASSERT(sm != NULL);
ASSERT3U(space_map_object(sm), !=, 0);
ASSERT3U(sm->sm_dbuf->db_size, ==, sizeof (space_map_phys_t));
/*
* Create a joint histogram from all the segments that have made
* it to the metaslab's space map histogram, that are not yet
* available for allocation because they are still in the freeing
* pipeline (e.g. freeing, freed, and defer trees). Then subtract
* these segments from the space map's histogram to get a more
* accurate weight.
*/
uint64_t deferspace_histogram[SPACE_MAP_HISTOGRAM_SIZE] = {0};
for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++)
deferspace_histogram[i] += msp->ms_synchist[i];
for (int t = 0; t < TXG_DEFER_SIZE; t++) {
for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
deferspace_histogram[i] += msp->ms_deferhist[t][i];
}
}
uint64_t weight = 0;
for (int i = SPACE_MAP_HISTOGRAM_SIZE - 1; i >= 0; i--) {
ASSERT3U(sm->sm_phys->smp_histogram[i], >=,
deferspace_histogram[i]);
uint64_t count =
sm->sm_phys->smp_histogram[i] - deferspace_histogram[i];
if (count != 0) {
WEIGHT_SET_COUNT(weight, count);
WEIGHT_SET_INDEX(weight, i + sm->sm_shift);
WEIGHT_SET_ACTIVE(weight, 0);
break;
}
}
return (weight);
}
/*
* Compute a segment-based weight for the specified metaslab. The weight
* is determined by highest bucket in the histogram. The information
* for the highest bucket is encoded into the weight value.
*/
static uint64_t
metaslab_segment_weight(metaslab_t *msp)
{
metaslab_group_t *mg = msp->ms_group;
uint64_t weight = 0;
uint8_t shift = mg->mg_vd->vdev_ashift;
ASSERT(MUTEX_HELD(&msp->ms_lock));
/*
* The metaslab is completely free.
*/
if (metaslab_allocated_space(msp) == 0) {
int idx = highbit64(msp->ms_size) - 1;
int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1;
if (idx < max_idx) {
WEIGHT_SET_COUNT(weight, 1ULL);
WEIGHT_SET_INDEX(weight, idx);
} else {
WEIGHT_SET_COUNT(weight, 1ULL << (idx - max_idx));
WEIGHT_SET_INDEX(weight, max_idx);
}
WEIGHT_SET_ACTIVE(weight, 0);
ASSERT(!WEIGHT_IS_SPACEBASED(weight));
return (weight);
}
ASSERT3U(msp->ms_sm->sm_dbuf->db_size, ==, sizeof (space_map_phys_t));
/*
* If the metaslab is fully allocated then just make the weight 0.
*/
if (metaslab_allocated_space(msp) == msp->ms_size)
return (0);
/*
* If the metaslab is already loaded, then use the range tree to
* determine the weight. Otherwise, we rely on the space map information
* to generate the weight.
*/
if (msp->ms_loaded) {
weight = metaslab_weight_from_range_tree(msp);
} else {
weight = metaslab_weight_from_spacemap(msp);
}
/*
* If the metaslab was active the last time we calculated its weight
* then keep it active. We want to consume the entire region that
* is associated with this weight.
*/
if (msp->ms_activation_weight != 0 && weight != 0)
WEIGHT_SET_ACTIVE(weight, WEIGHT_GET_ACTIVE(msp->ms_weight));
return (weight);
}
/*
* Determine if we should attempt to allocate from this metaslab. If the
* metaslab is loaded, then we can determine if the desired allocation
* can be satisfied by looking at the size of the maximum free segment
* on that metaslab. Otherwise, we make our decision based on the metaslab's
* weight. For segment-based weighting we can determine the maximum
* allocation based on the index encoded in its value. For space-based
* weights we rely on the entire weight (excluding the weight-type bit).
*/
static boolean_t
metaslab_should_allocate(metaslab_t *msp, uint64_t asize, boolean_t try_hard)
{
/*
* If the metaslab is loaded, ms_max_size is definitive and we can use
* the fast check. If it's not, the ms_max_size is a lower bound (once
* set), and we should use the fast check as long as we're not in
* try_hard and it's been less than zfs_metaslab_max_size_cache_sec
* seconds since the metaslab was unloaded.
*/
if (msp->ms_loaded ||
(msp->ms_max_size != 0 && !try_hard && gethrtime() <
msp->ms_unload_time + SEC2NSEC(zfs_metaslab_max_size_cache_sec)))
return (msp->ms_max_size >= asize);
boolean_t should_allocate;
if (!WEIGHT_IS_SPACEBASED(msp->ms_weight)) {
/*
* The metaslab segment weight indicates segments in the
* range [2^i, 2^(i+1)), where i is the index in the weight.
* Since the asize might be in the middle of the range, we
* should attempt the allocation if asize < 2^(i+1).
*/
should_allocate = (asize <
1ULL << (WEIGHT_GET_INDEX(msp->ms_weight) + 1));
} else {
should_allocate = (asize <=
(msp->ms_weight & ~METASLAB_WEIGHT_TYPE));
}
return (should_allocate);
}
static uint64_t
metaslab_weight(metaslab_t *msp, boolean_t nodirty)
{
vdev_t *vd = msp->ms_group->mg_vd;
spa_t *spa = vd->vdev_spa;
uint64_t weight;
ASSERT(MUTEX_HELD(&msp->ms_lock));
metaslab_set_fragmentation(msp, nodirty);
/*
* Update the maximum size. If the metaslab is loaded, this will
* ensure that we get an accurate maximum size if newly freed space
* has been added back into the free tree. If the metaslab is
* unloaded, we check if there's a larger free segment in the
* unflushed frees. This is a lower bound on the largest allocatable
* segment size. Coalescing of adjacent entries may reveal larger
* allocatable segments, but we aren't aware of those until loading
* the space map into a range tree.
*/
if (msp->ms_loaded) {
msp->ms_max_size = metaslab_largest_allocatable(msp);
} else {
msp->ms_max_size = MAX(msp->ms_max_size,
metaslab_largest_unflushed_free(msp));
}
/*
* Segment-based weighting requires space map histogram support.
*/
if (zfs_metaslab_segment_weight_enabled &&
spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM) &&
(msp->ms_sm == NULL || msp->ms_sm->sm_dbuf->db_size ==
sizeof (space_map_phys_t))) {
weight = metaslab_segment_weight(msp);
} else {
weight = metaslab_space_weight(msp);
}
return (weight);
}
void
metaslab_recalculate_weight_and_sort(metaslab_t *msp)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
/* note: we preserve the mask (e.g. indication of primary, etc..) */
uint64_t was_active = msp->ms_weight & METASLAB_ACTIVE_MASK;
metaslab_group_sort(msp->ms_group, msp,
metaslab_weight(msp, B_FALSE) | was_active);
}
static int
metaslab_activate_allocator(metaslab_group_t *mg, metaslab_t *msp,
int allocator, uint64_t activation_weight)
{
metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator];
ASSERT(MUTEX_HELD(&msp->ms_lock));
/*
* If we're activating for the claim code, we don't want to actually
* set the metaslab up for a specific allocator.
*/
if (activation_weight == METASLAB_WEIGHT_CLAIM) {
ASSERT0(msp->ms_activation_weight);
msp->ms_activation_weight = msp->ms_weight;
metaslab_group_sort(mg, msp, msp->ms_weight |
activation_weight);
return (0);
}
metaslab_t **mspp = (activation_weight == METASLAB_WEIGHT_PRIMARY ?
&mga->mga_primary : &mga->mga_secondary);
mutex_enter(&mg->mg_lock);
if (*mspp != NULL) {
mutex_exit(&mg->mg_lock);
return (EEXIST);
}
*mspp = msp;
ASSERT3S(msp->ms_allocator, ==, -1);
msp->ms_allocator = allocator;
msp->ms_primary = (activation_weight == METASLAB_WEIGHT_PRIMARY);
ASSERT0(msp->ms_activation_weight);
msp->ms_activation_weight = msp->ms_weight;
metaslab_group_sort_impl(mg, msp,
msp->ms_weight | activation_weight);
mutex_exit(&mg->mg_lock);
return (0);
}
static int
metaslab_activate(metaslab_t *msp, int allocator, uint64_t activation_weight)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
/*
* The current metaslab is already activated for us so there
* is nothing to do. Already activated though, doesn't mean
* that this metaslab is activated for our allocator nor our
* requested activation weight. The metaslab could have started
* as an active one for our allocator but changed allocators
* while we were waiting to grab its ms_lock or we stole it
* [see find_valid_metaslab()]. This means that there is a
* possibility of passivating a metaslab of another allocator
* or from a different activation mask, from this thread.
*/
if ((msp->ms_weight & METASLAB_ACTIVE_MASK) != 0) {
ASSERT(msp->ms_loaded);
return (0);
}
int error = metaslab_load(msp);
if (error != 0) {
metaslab_group_sort(msp->ms_group, msp, 0);
return (error);
}
/*
* When entering metaslab_load() we may have dropped the
* ms_lock because we were loading this metaslab, or we
* were waiting for another thread to load it for us. In
* that scenario, we recheck the weight of the metaslab
* to see if it was activated by another thread.
*
* If the metaslab was activated for another allocator or
* it was activated with a different activation weight (e.g.
* we wanted to make it a primary but it was activated as
* secondary) we return error (EBUSY).
*
* If the metaslab was activated for the same allocator
* and requested activation mask, skip activating it.
*/
if ((msp->ms_weight & METASLAB_ACTIVE_MASK) != 0) {
if (msp->ms_allocator != allocator)
return (EBUSY);
if ((msp->ms_weight & activation_weight) == 0)
return (SET_ERROR(EBUSY));
EQUIV((activation_weight == METASLAB_WEIGHT_PRIMARY),
msp->ms_primary);
return (0);
}
/*
* If the metaslab has literally 0 space, it will have weight 0. In
* that case, don't bother activating it. This can happen if the
* metaslab had space during find_valid_metaslab, but another thread
* loaded it and used all that space while we were waiting to grab the
* lock.
*/
if (msp->ms_weight == 0) {
ASSERT0(range_tree_space(msp->ms_allocatable));
return (SET_ERROR(ENOSPC));
}
if ((error = metaslab_activate_allocator(msp->ms_group, msp,
allocator, activation_weight)) != 0) {
return (error);
}
ASSERT(msp->ms_loaded);
ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK);
return (0);
}
static void
metaslab_passivate_allocator(metaslab_group_t *mg, metaslab_t *msp,
uint64_t weight)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
ASSERT(msp->ms_loaded);
if (msp->ms_weight & METASLAB_WEIGHT_CLAIM) {
metaslab_group_sort(mg, msp, weight);
return;
}
mutex_enter(&mg->mg_lock);
ASSERT3P(msp->ms_group, ==, mg);
ASSERT3S(0, <=, msp->ms_allocator);
ASSERT3U(msp->ms_allocator, <, mg->mg_allocators);
metaslab_group_allocator_t *mga = &mg->mg_allocator[msp->ms_allocator];
if (msp->ms_primary) {
ASSERT3P(mga->mga_primary, ==, msp);
ASSERT(msp->ms_weight & METASLAB_WEIGHT_PRIMARY);
mga->mga_primary = NULL;
} else {
ASSERT3P(mga->mga_secondary, ==, msp);
ASSERT(msp->ms_weight & METASLAB_WEIGHT_SECONDARY);
mga->mga_secondary = NULL;
}
msp->ms_allocator = -1;
metaslab_group_sort_impl(mg, msp, weight);
mutex_exit(&mg->mg_lock);
}
static void
metaslab_passivate(metaslab_t *msp, uint64_t weight)
{
uint64_t size __maybe_unused = weight & ~METASLAB_WEIGHT_TYPE;
/*
* If size < SPA_MINBLOCKSIZE, then we will not allocate from
* this metaslab again. In that case, it had better be empty,
* or we would be leaving space on the table.
*/
ASSERT(!WEIGHT_IS_SPACEBASED(msp->ms_weight) ||
size >= SPA_MINBLOCKSIZE ||
range_tree_space(msp->ms_allocatable) == 0);
ASSERT0(weight & METASLAB_ACTIVE_MASK);
ASSERT(msp->ms_activation_weight != 0);
msp->ms_activation_weight = 0;
metaslab_passivate_allocator(msp->ms_group, msp, weight);
ASSERT0(msp->ms_weight & METASLAB_ACTIVE_MASK);
}
/*
* Segment-based metaslabs are activated once and remain active until
* we either fail an allocation attempt (similar to space-based metaslabs)
* or have exhausted the free space in zfs_metaslab_switch_threshold
* buckets since the metaslab was activated. This function checks to see
* if we've exhausted the zfs_metaslab_switch_threshold buckets in the
* metaslab and passivates it proactively. This will allow us to select a
* metaslab with a larger contiguous region, if any, remaining within this
* metaslab group. If we're in sync pass > 1, then we continue using this
* metaslab so that we don't dirty more block and cause more sync passes.
*/
static void
metaslab_segment_may_passivate(metaslab_t *msp)
{
spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
if (WEIGHT_IS_SPACEBASED(msp->ms_weight) || spa_sync_pass(spa) > 1)
return;
/*
* Since we are in the middle of a sync pass, the most accurate
* information that is accessible to us is the in-core range tree
* histogram; calculate the new weight based on that information.
*/
uint64_t weight = metaslab_weight_from_range_tree(msp);
int activation_idx = WEIGHT_GET_INDEX(msp->ms_activation_weight);
int current_idx = WEIGHT_GET_INDEX(weight);
if (current_idx <= activation_idx - zfs_metaslab_switch_threshold)
metaslab_passivate(msp, weight);
}
static void
metaslab_preload(void *arg)
{
metaslab_t *msp = arg;
metaslab_class_t *mc = msp->ms_group->mg_class;
spa_t *spa = mc->mc_spa;
fstrans_cookie_t cookie = spl_fstrans_mark();
ASSERT(!MUTEX_HELD(&msp->ms_group->mg_lock));
mutex_enter(&msp->ms_lock);
(void) metaslab_load(msp);
metaslab_set_selected_txg(msp, spa_syncing_txg(spa));
mutex_exit(&msp->ms_lock);
spl_fstrans_unmark(cookie);
}
static void
metaslab_group_preload(metaslab_group_t *mg)
{
spa_t *spa = mg->mg_vd->vdev_spa;
metaslab_t *msp;
avl_tree_t *t = &mg->mg_metaslab_tree;
int m = 0;
if (spa_shutting_down(spa) || !metaslab_preload_enabled) {
taskq_wait_outstanding(mg->mg_taskq, 0);
return;
}
mutex_enter(&mg->mg_lock);
/*
* Load the next potential metaslabs
*/
for (msp = avl_first(t); msp != NULL; msp = AVL_NEXT(t, msp)) {
ASSERT3P(msp->ms_group, ==, mg);
/*
* We preload only the maximum number of metaslabs specified
* by metaslab_preload_limit. If a metaslab is being forced
* to condense then we preload it too. This will ensure
* that force condensing happens in the next txg.
*/
if (++m > metaslab_preload_limit && !msp->ms_condense_wanted) {
continue;
}
VERIFY(taskq_dispatch(mg->mg_taskq, metaslab_preload,
msp, TQ_SLEEP) != TASKQID_INVALID);
}
mutex_exit(&mg->mg_lock);
}
/*
* Determine if the space map's on-disk footprint is past our tolerance for
* inefficiency. We would like to use the following criteria to make our
* decision:
*
* 1. Do not condense if the size of the space map object would dramatically
* increase as a result of writing out the free space range tree.
*
* 2. Condense if the on on-disk space map representation is at least
* zfs_condense_pct/100 times the size of the optimal representation
* (i.e. zfs_condense_pct = 110 and in-core = 1MB, optimal = 1.1MB).
*
* 3. Do not condense if the on-disk size of the space map does not actually
* decrease.
*
* Unfortunately, we cannot compute the on-disk size of the space map in this
* context because we cannot accurately compute the effects of compression, etc.
* Instead, we apply the heuristic described in the block comment for
* zfs_metaslab_condense_block_threshold - we only condense if the space used
* is greater than a threshold number of blocks.
*/
static boolean_t
metaslab_should_condense(metaslab_t *msp)
{
space_map_t *sm = msp->ms_sm;
vdev_t *vd = msp->ms_group->mg_vd;
uint64_t vdev_blocksize = 1 << vd->vdev_ashift;
ASSERT(MUTEX_HELD(&msp->ms_lock));
ASSERT(msp->ms_loaded);
ASSERT(sm != NULL);
ASSERT3U(spa_sync_pass(vd->vdev_spa), ==, 1);
/*
* We always condense metaslabs that are empty and metaslabs for
* which a condense request has been made.
*/
if (range_tree_numsegs(msp->ms_allocatable) == 0 ||
msp->ms_condense_wanted)
return (B_TRUE);
uint64_t record_size = MAX(sm->sm_blksz, vdev_blocksize);
uint64_t object_size = space_map_length(sm);
uint64_t optimal_size = space_map_estimate_optimal_size(sm,
msp->ms_allocatable, SM_NO_VDEVID);
return (object_size >= (optimal_size * zfs_condense_pct / 100) &&
object_size > zfs_metaslab_condense_block_threshold * record_size);
}
/*
* Condense the on-disk space map representation to its minimized form.
* The minimized form consists of a small number of allocations followed
* by the entries of the free range tree (ms_allocatable). The condensed
* spacemap contains all the entries of previous TXGs (including those in
* the pool-wide log spacemaps; thus this is effectively a superset of
* metaslab_flush()), but this TXG's entries still need to be written.
*/
static void
metaslab_condense(metaslab_t *msp, dmu_tx_t *tx)
{
range_tree_t *condense_tree;
space_map_t *sm = msp->ms_sm;
uint64_t txg = dmu_tx_get_txg(tx);
spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
ASSERT(MUTEX_HELD(&msp->ms_lock));
ASSERT(msp->ms_loaded);
ASSERT(msp->ms_sm != NULL);
/*
* In order to condense the space map, we need to change it so it
* only describes which segments are currently allocated and free.
*
* All the current free space resides in the ms_allocatable, all
* the ms_defer trees, and all the ms_allocating trees. We ignore
* ms_freed because it is empty because we're in sync pass 1. We
* ignore ms_freeing because these changes are not yet reflected
* in the spacemap (they will be written later this txg).
*
* So to truncate the space map to represent all the entries of
* previous TXGs we do the following:
*
* 1] We create a range tree (condense tree) that is 100% empty.
* 2] We add to it all segments found in the ms_defer trees
* as those segments are marked as free in the original space
* map. We do the same with the ms_allocating trees for the same
* reason. Adding these segments should be a relatively
* inexpensive operation since we expect these trees to have a
* small number of nodes.
* 3] We vacate any unflushed allocs, since they are not frees we
* need to add to the condense tree. Then we vacate any
* unflushed frees as they should already be part of ms_allocatable.
* 4] At this point, we would ideally like to add all segments
* in the ms_allocatable tree from the condense tree. This way
* we would write all the entries of the condense tree as the
* condensed space map, which would only contain freed
* segments with everything else assumed to be allocated.
*
* Doing so can be prohibitively expensive as ms_allocatable can
* be large, and therefore computationally expensive to add to
* the condense_tree. Instead we first sync out an entry marking
* everything as allocated, then the condense_tree and then the
* ms_allocatable, in the condensed space map. While this is not
* optimal, it is typically close to optimal and more importantly
* much cheaper to compute.
*
* 5] Finally, as both of the unflushed trees were written to our
* new and condensed metaslab space map, we basically flushed
* all the unflushed changes to disk, thus we call
* metaslab_flush_update().
*/
ASSERT3U(spa_sync_pass(spa), ==, 1);
ASSERT(range_tree_is_empty(msp->ms_freed)); /* since it is pass 1 */
zfs_dbgmsg("condensing: txg %llu, msp[%llu] %px, vdev id %llu, "
"spa %s, smp size %llu, segments %llu, forcing condense=%s",
(u_longlong_t)txg, (u_longlong_t)msp->ms_id, msp,
(u_longlong_t)msp->ms_group->mg_vd->vdev_id,
spa->spa_name, (u_longlong_t)space_map_length(msp->ms_sm),
(u_longlong_t)range_tree_numsegs(msp->ms_allocatable),
msp->ms_condense_wanted ? "TRUE" : "FALSE");
msp->ms_condense_wanted = B_FALSE;
range_seg_type_t type;
uint64_t shift, start;
type = metaslab_calculate_range_tree_type(msp->ms_group->mg_vd, msp,
&start, &shift);
condense_tree = range_tree_create(NULL, type, NULL, start, shift);
for (int t = 0; t < TXG_DEFER_SIZE; t++) {
range_tree_walk(msp->ms_defer[t],
range_tree_add, condense_tree);
}
for (int t = 0; t < TXG_CONCURRENT_STATES; t++) {
range_tree_walk(msp->ms_allocating[(txg + t) & TXG_MASK],
range_tree_add, condense_tree);
}
ASSERT3U(spa->spa_unflushed_stats.sus_memused, >=,
metaslab_unflushed_changes_memused(msp));
spa->spa_unflushed_stats.sus_memused -=
metaslab_unflushed_changes_memused(msp);
range_tree_vacate(msp->ms_unflushed_allocs, NULL, NULL);
range_tree_vacate(msp->ms_unflushed_frees, NULL, NULL);
/*
* We're about to drop the metaslab's lock thus allowing other
* consumers to change it's content. Set the metaslab's ms_condensing
* flag to ensure that allocations on this metaslab do not occur
* while we're in the middle of committing it to disk. This is only
* critical for ms_allocatable as all other range trees use per TXG
* views of their content.
*/
msp->ms_condensing = B_TRUE;
mutex_exit(&msp->ms_lock);
uint64_t object = space_map_object(msp->ms_sm);
space_map_truncate(sm,
spa_feature_is_enabled(spa, SPA_FEATURE_LOG_SPACEMAP) ?
zfs_metaslab_sm_blksz_with_log : zfs_metaslab_sm_blksz_no_log, tx);
/*
* space_map_truncate() may have reallocated the spacemap object.
* If so, update the vdev_ms_array.
*/
if (space_map_object(msp->ms_sm) != object) {
object = space_map_object(msp->ms_sm);
dmu_write(spa->spa_meta_objset,
msp->ms_group->mg_vd->vdev_ms_array, sizeof (uint64_t) *
msp->ms_id, sizeof (uint64_t), &object, tx);
}
/*
* Note:
* When the log space map feature is enabled, each space map will
* always have ALLOCS followed by FREES for each sync pass. This is
* typically true even when the log space map feature is disabled,
* except from the case where a metaslab goes through metaslab_sync()
* and gets condensed. In that case the metaslab's space map will have
* ALLOCS followed by FREES (due to condensing) followed by ALLOCS
* followed by FREES (due to space_map_write() in metaslab_sync()) for
* sync pass 1.
*/
range_tree_t *tmp_tree = range_tree_create(NULL, type, NULL, start,
shift);
range_tree_add(tmp_tree, msp->ms_start, msp->ms_size);
space_map_write(sm, tmp_tree, SM_ALLOC, SM_NO_VDEVID, tx);
space_map_write(sm, msp->ms_allocatable, SM_FREE, SM_NO_VDEVID, tx);
space_map_write(sm, condense_tree, SM_FREE, SM_NO_VDEVID, tx);
range_tree_vacate(condense_tree, NULL, NULL);
range_tree_destroy(condense_tree);
range_tree_vacate(tmp_tree, NULL, NULL);
range_tree_destroy(tmp_tree);
mutex_enter(&msp->ms_lock);
msp->ms_condensing = B_FALSE;
metaslab_flush_update(msp, tx);
}
/*
* Called when the metaslab has been flushed (its own spacemap now reflects
* all the contents of the pool-wide spacemap log). Updates the metaslab's
* metadata and any pool-wide related log space map data (e.g. summary,
* obsolete logs, etc..) to reflect that.
*/
static void
metaslab_flush_update(metaslab_t *msp, dmu_tx_t *tx)
{
metaslab_group_t *mg = msp->ms_group;
spa_t *spa = mg->mg_vd->vdev_spa;
ASSERT(MUTEX_HELD(&msp->ms_lock));
ASSERT3U(spa_sync_pass(spa), ==, 1);
ASSERT(range_tree_is_empty(msp->ms_unflushed_allocs));
ASSERT(range_tree_is_empty(msp->ms_unflushed_frees));
/*
* Just because a metaslab got flushed, that doesn't mean that
* it will pass through metaslab_sync_done(). Thus, make sure to
* update ms_synced_length here in case it doesn't.
*/
msp->ms_synced_length = space_map_length(msp->ms_sm);
/*
* We may end up here from metaslab_condense() without the
* feature being active. In that case this is a no-op.
*/
if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
return;
ASSERT(spa_syncing_log_sm(spa) != NULL);
ASSERT(msp->ms_sm != NULL);
ASSERT(metaslab_unflushed_txg(msp) != 0);
ASSERT3P(avl_find(&spa->spa_metaslabs_by_flushed, msp, NULL), ==, msp);
VERIFY3U(tx->tx_txg, <=, spa_final_dirty_txg(spa));
/* update metaslab's position in our flushing tree */
uint64_t ms_prev_flushed_txg = metaslab_unflushed_txg(msp);
mutex_enter(&spa->spa_flushed_ms_lock);
avl_remove(&spa->spa_metaslabs_by_flushed, msp);
metaslab_set_unflushed_txg(msp, spa_syncing_txg(spa), tx);
avl_add(&spa->spa_metaslabs_by_flushed, msp);
mutex_exit(&spa->spa_flushed_ms_lock);
/* update metaslab counts of spa_log_sm_t nodes */
spa_log_sm_decrement_mscount(spa, ms_prev_flushed_txg);
spa_log_sm_increment_current_mscount(spa);
/* cleanup obsolete logs if any */
uint64_t log_blocks_before = spa_log_sm_nblocks(spa);
spa_cleanup_old_sm_logs(spa, tx);
uint64_t log_blocks_after = spa_log_sm_nblocks(spa);
VERIFY3U(log_blocks_after, <=, log_blocks_before);
/* update log space map summary */
uint64_t blocks_gone = log_blocks_before - log_blocks_after;
spa_log_summary_add_flushed_metaslab(spa);
spa_log_summary_decrement_mscount(spa, ms_prev_flushed_txg);
spa_log_summary_decrement_blkcount(spa, blocks_gone);
}
boolean_t
metaslab_flush(metaslab_t *msp, dmu_tx_t *tx)
{
spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
ASSERT(MUTEX_HELD(&msp->ms_lock));
ASSERT3U(spa_sync_pass(spa), ==, 1);
ASSERT(spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP));
ASSERT(msp->ms_sm != NULL);
ASSERT(metaslab_unflushed_txg(msp) != 0);
ASSERT(avl_find(&spa->spa_metaslabs_by_flushed, msp, NULL) != NULL);
/*
* There is nothing wrong with flushing the same metaslab twice, as
* this codepath should work on that case. However, the current
* flushing scheme makes sure to avoid this situation as we would be
* making all these calls without having anything meaningful to write
* to disk. We assert this behavior here.
*/
ASSERT3U(metaslab_unflushed_txg(msp), <, dmu_tx_get_txg(tx));
/*
* We can not flush while loading, because then we would
* not load the ms_unflushed_{allocs,frees}.
*/
if (msp->ms_loading)
return (B_FALSE);
metaslab_verify_space(msp, dmu_tx_get_txg(tx));
metaslab_verify_weight_and_frag(msp);
/*
* Metaslab condensing is effectively flushing. Therefore if the
* metaslab can be condensed we can just condense it instead of
* flushing it.
*
* Note that metaslab_condense() does call metaslab_flush_update()
* so we can just return immediately after condensing. We also
* don't need to care about setting ms_flushing or broadcasting
* ms_flush_cv, even if we temporarily drop the ms_lock in
* metaslab_condense(), as the metaslab is already loaded.
*/
if (msp->ms_loaded && metaslab_should_condense(msp)) {
metaslab_group_t *mg = msp->ms_group;
/*
* For all histogram operations below refer to the
* comments of metaslab_sync() where we follow a
* similar procedure.
*/
metaslab_group_histogram_verify(mg);
metaslab_class_histogram_verify(mg->mg_class);
metaslab_group_histogram_remove(mg, msp);
metaslab_condense(msp, tx);
space_map_histogram_clear(msp->ms_sm);
space_map_histogram_add(msp->ms_sm, msp->ms_allocatable, tx);
ASSERT(range_tree_is_empty(msp->ms_freed));
for (int t = 0; t < TXG_DEFER_SIZE; t++) {
space_map_histogram_add(msp->ms_sm,
msp->ms_defer[t], tx);
}
metaslab_aux_histograms_update(msp);
metaslab_group_histogram_add(mg, msp);
metaslab_group_histogram_verify(mg);
metaslab_class_histogram_verify(mg->mg_class);
metaslab_verify_space(msp, dmu_tx_get_txg(tx));
/*
* Since we recreated the histogram (and potentially
* the ms_sm too while condensing) ensure that the
* weight is updated too because we are not guaranteed
* that this metaslab is dirty and will go through
* metaslab_sync_done().
*/
metaslab_recalculate_weight_and_sort(msp);
return (B_TRUE);
}
msp->ms_flushing = B_TRUE;
uint64_t sm_len_before = space_map_length(msp->ms_sm);
mutex_exit(&msp->ms_lock);
space_map_write(msp->ms_sm, msp->ms_unflushed_allocs, SM_ALLOC,
SM_NO_VDEVID, tx);
space_map_write(msp->ms_sm, msp->ms_unflushed_frees, SM_FREE,
SM_NO_VDEVID, tx);
mutex_enter(&msp->ms_lock);
uint64_t sm_len_after = space_map_length(msp->ms_sm);
if (zfs_flags & ZFS_DEBUG_LOG_SPACEMAP) {
zfs_dbgmsg("flushing: txg %llu, spa %s, vdev_id %llu, "
"ms_id %llu, unflushed_allocs %llu, unflushed_frees %llu, "
"appended %llu bytes", (u_longlong_t)dmu_tx_get_txg(tx),
spa_name(spa),
(u_longlong_t)msp->ms_group->mg_vd->vdev_id,
(u_longlong_t)msp->ms_id,
(u_longlong_t)range_tree_space(msp->ms_unflushed_allocs),
(u_longlong_t)range_tree_space(msp->ms_unflushed_frees),
(u_longlong_t)(sm_len_after - sm_len_before));
}
ASSERT3U(spa->spa_unflushed_stats.sus_memused, >=,
metaslab_unflushed_changes_memused(msp));
spa->spa_unflushed_stats.sus_memused -=
metaslab_unflushed_changes_memused(msp);
range_tree_vacate(msp->ms_unflushed_allocs, NULL, NULL);
range_tree_vacate(msp->ms_unflushed_frees, NULL, NULL);
metaslab_verify_space(msp, dmu_tx_get_txg(tx));
metaslab_verify_weight_and_frag(msp);
metaslab_flush_update(msp, tx);
metaslab_verify_space(msp, dmu_tx_get_txg(tx));
metaslab_verify_weight_and_frag(msp);
msp->ms_flushing = B_FALSE;
cv_broadcast(&msp->ms_flush_cv);
return (B_TRUE);
}
/*
* Write a metaslab to disk in the context of the specified transaction group.
*/
void
metaslab_sync(metaslab_t *msp, uint64_t txg)
{
metaslab_group_t *mg = msp->ms_group;
vdev_t *vd = mg->mg_vd;
spa_t *spa = vd->vdev_spa;
objset_t *mos = spa_meta_objset(spa);
range_tree_t *alloctree = msp->ms_allocating[txg & TXG_MASK];
dmu_tx_t *tx;
ASSERT(!vd->vdev_ishole);
/*
* This metaslab has just been added so there's no work to do now.
*/
if (msp->ms_new) {
ASSERT0(range_tree_space(alloctree));
ASSERT0(range_tree_space(msp->ms_freeing));
ASSERT0(range_tree_space(msp->ms_freed));
ASSERT0(range_tree_space(msp->ms_checkpointing));
ASSERT0(range_tree_space(msp->ms_trim));
return;
}
/*
* Normally, we don't want to process a metaslab if there are no
* allocations or frees to perform. However, if the metaslab is being
* forced to condense, it's loaded and we're not beyond the final
* dirty txg, we need to let it through. Not condensing beyond the
* final dirty txg prevents an issue where metaslabs that need to be
* condensed but were loaded for other reasons could cause a panic
* here. By only checking the txg in that branch of the conditional,
* we preserve the utility of the VERIFY statements in all other
* cases.
*/
if (range_tree_is_empty(alloctree) &&
range_tree_is_empty(msp->ms_freeing) &&
range_tree_is_empty(msp->ms_checkpointing) &&
!(msp->ms_loaded && msp->ms_condense_wanted &&
txg <= spa_final_dirty_txg(spa)))
return;
VERIFY3U(txg, <=, spa_final_dirty_txg(spa));
/*
* The only state that can actually be changing concurrently
* with metaslab_sync() is the metaslab's ms_allocatable. No
* other thread can be modifying this txg's alloc, freeing,
* freed, or space_map_phys_t. We drop ms_lock whenever we
* could call into the DMU, because the DMU can call down to
* us (e.g. via zio_free()) at any time.
*
* The spa_vdev_remove_thread() can be reading metaslab state
* concurrently, and it is locked out by the ms_sync_lock.
* Note that the ms_lock is insufficient for this, because it
* is dropped by space_map_write().
*/
tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
/*
* Generate a log space map if one doesn't exist already.
*/
spa_generate_syncing_log_sm(spa, tx);
if (msp->ms_sm == NULL) {
uint64_t new_object = space_map_alloc(mos,
spa_feature_is_enabled(spa, SPA_FEATURE_LOG_SPACEMAP) ?
zfs_metaslab_sm_blksz_with_log :
zfs_metaslab_sm_blksz_no_log, tx);
VERIFY3U(new_object, !=, 0);
dmu_write(mos, vd->vdev_ms_array, sizeof (uint64_t) *
msp->ms_id, sizeof (uint64_t), &new_object, tx);
VERIFY0(space_map_open(&msp->ms_sm, mos, new_object,
msp->ms_start, msp->ms_size, vd->vdev_ashift));
ASSERT(msp->ms_sm != NULL);
ASSERT(range_tree_is_empty(msp->ms_unflushed_allocs));
ASSERT(range_tree_is_empty(msp->ms_unflushed_frees));
ASSERT0(metaslab_allocated_space(msp));
}
if (metaslab_unflushed_txg(msp) == 0 &&
spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
ASSERT(spa_syncing_log_sm(spa) != NULL);
metaslab_set_unflushed_txg(msp, spa_syncing_txg(spa), tx);
spa_log_sm_increment_current_mscount(spa);
spa_log_summary_add_flushed_metaslab(spa);
ASSERT(msp->ms_sm != NULL);
mutex_enter(&spa->spa_flushed_ms_lock);
avl_add(&spa->spa_metaslabs_by_flushed, msp);
mutex_exit(&spa->spa_flushed_ms_lock);
ASSERT(range_tree_is_empty(msp->ms_unflushed_allocs));
ASSERT(range_tree_is_empty(msp->ms_unflushed_frees));
}
if (!range_tree_is_empty(msp->ms_checkpointing) &&
vd->vdev_checkpoint_sm == NULL) {
ASSERT(spa_has_checkpoint(spa));
uint64_t new_object = space_map_alloc(mos,
zfs_vdev_standard_sm_blksz, tx);
VERIFY3U(new_object, !=, 0);
VERIFY0(space_map_open(&vd->vdev_checkpoint_sm,
mos, new_object, 0, vd->vdev_asize, vd->vdev_ashift));
ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
/*
* We save the space map object as an entry in vdev_top_zap
* so it can be retrieved when the pool is reopened after an
* export or through zdb.
*/
VERIFY0(zap_add(vd->vdev_spa->spa_meta_objset,
vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
sizeof (new_object), 1, &new_object, tx));
}
mutex_enter(&msp->ms_sync_lock);
mutex_enter(&msp->ms_lock);
/*
* Note: metaslab_condense() clears the space map's histogram.
* Therefore we must verify and remove this histogram before
* condensing.
*/
metaslab_group_histogram_verify(mg);
metaslab_class_histogram_verify(mg->mg_class);
metaslab_group_histogram_remove(mg, msp);
if (spa->spa_sync_pass == 1 && msp->ms_loaded &&
metaslab_should_condense(msp))
metaslab_condense(msp, tx);
/*
* We'll be going to disk to sync our space accounting, thus we
* drop the ms_lock during that time so allocations coming from
* open-context (ZIL) for future TXGs do not block.
*/
mutex_exit(&msp->ms_lock);
space_map_t *log_sm = spa_syncing_log_sm(spa);
if (log_sm != NULL) {
ASSERT(spa_feature_is_enabled(spa, SPA_FEATURE_LOG_SPACEMAP));
space_map_write(log_sm, alloctree, SM_ALLOC,
vd->vdev_id, tx);
space_map_write(log_sm, msp->ms_freeing, SM_FREE,
vd->vdev_id, tx);
mutex_enter(&msp->ms_lock);
ASSERT3U(spa->spa_unflushed_stats.sus_memused, >=,
metaslab_unflushed_changes_memused(msp));
spa->spa_unflushed_stats.sus_memused -=
metaslab_unflushed_changes_memused(msp);
range_tree_remove_xor_add(alloctree,
msp->ms_unflushed_frees, msp->ms_unflushed_allocs);
range_tree_remove_xor_add(msp->ms_freeing,
msp->ms_unflushed_allocs, msp->ms_unflushed_frees);
spa->spa_unflushed_stats.sus_memused +=
metaslab_unflushed_changes_memused(msp);
} else {
ASSERT(!spa_feature_is_enabled(spa, SPA_FEATURE_LOG_SPACEMAP));
space_map_write(msp->ms_sm, alloctree, SM_ALLOC,
SM_NO_VDEVID, tx);
space_map_write(msp->ms_sm, msp->ms_freeing, SM_FREE,
SM_NO_VDEVID, tx);
mutex_enter(&msp->ms_lock);
}
msp->ms_allocated_space += range_tree_space(alloctree);
ASSERT3U(msp->ms_allocated_space, >=,
range_tree_space(msp->ms_freeing));
msp->ms_allocated_space -= range_tree_space(msp->ms_freeing);
if (!range_tree_is_empty(msp->ms_checkpointing)) {
ASSERT(spa_has_checkpoint(spa));
ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL);
/*
* Since we are doing writes to disk and the ms_checkpointing
* tree won't be changing during that time, we drop the
* ms_lock while writing to the checkpoint space map, for the
* same reason mentioned above.
*/
mutex_exit(&msp->ms_lock);
space_map_write(vd->vdev_checkpoint_sm,
msp->ms_checkpointing, SM_FREE, SM_NO_VDEVID, tx);
mutex_enter(&msp->ms_lock);
spa->spa_checkpoint_info.sci_dspace +=
range_tree_space(msp->ms_checkpointing);
vd->vdev_stat.vs_checkpoint_space +=
range_tree_space(msp->ms_checkpointing);
ASSERT3U(vd->vdev_stat.vs_checkpoint_space, ==,
-space_map_allocated(vd->vdev_checkpoint_sm));
range_tree_vacate(msp->ms_checkpointing, NULL, NULL);
}
if (msp->ms_loaded) {
/*
* When the space map is loaded, we have an accurate
* histogram in the range tree. This gives us an opportunity
* to bring the space map's histogram up-to-date so we clear
* it first before updating it.
*/
space_map_histogram_clear(msp->ms_sm);
space_map_histogram_add(msp->ms_sm, msp->ms_allocatable, tx);
/*
* Since we've cleared the histogram we need to add back
* any free space that has already been processed, plus
* any deferred space. This allows the on-disk histogram
* to accurately reflect all free space even if some space
* is not yet available for allocation (i.e. deferred).
*/
space_map_histogram_add(msp->ms_sm, msp->ms_freed, tx);
/*
* Add back any deferred free space that has not been
* added back into the in-core free tree yet. This will
* ensure that we don't end up with a space map histogram
* that is completely empty unless the metaslab is fully
* allocated.
*/
for (int t = 0; t < TXG_DEFER_SIZE; t++) {
space_map_histogram_add(msp->ms_sm,
msp->ms_defer[t], tx);
}
}
/*
* Always add the free space from this sync pass to the space
* map histogram. We want to make sure that the on-disk histogram
* accounts for all free space. If the space map is not loaded,
* then we will lose some accuracy but will correct it the next
* time we load the space map.
*/
space_map_histogram_add(msp->ms_sm, msp->ms_freeing, tx);
metaslab_aux_histograms_update(msp);
metaslab_group_histogram_add(mg, msp);
metaslab_group_histogram_verify(mg);
metaslab_class_histogram_verify(mg->mg_class);
/*
* For sync pass 1, we avoid traversing this txg's free range tree
* and instead will just swap the pointers for freeing and freed.
* We can safely do this since the freed_tree is guaranteed to be
* empty on the initial pass.
*
* Keep in mind that even if we are currently using a log spacemap
* we want current frees to end up in the ms_allocatable (but not
* get appended to the ms_sm) so their ranges can be reused as usual.
*/
if (spa_sync_pass(spa) == 1) {
range_tree_swap(&msp->ms_freeing, &msp->ms_freed);
ASSERT0(msp->ms_allocated_this_txg);
} else {
range_tree_vacate(msp->ms_freeing,
range_tree_add, msp->ms_freed);
}
msp->ms_allocated_this_txg += range_tree_space(alloctree);
range_tree_vacate(alloctree, NULL, NULL);
ASSERT0(range_tree_space(msp->ms_allocating[txg & TXG_MASK]));
ASSERT0(range_tree_space(msp->ms_allocating[TXG_CLEAN(txg)
& TXG_MASK]));
ASSERT0(range_tree_space(msp->ms_freeing));
ASSERT0(range_tree_space(msp->ms_checkpointing));
mutex_exit(&msp->ms_lock);
/*
* Verify that the space map object ID has been recorded in the
* vdev_ms_array.
*/
uint64_t object;
VERIFY0(dmu_read(mos, vd->vdev_ms_array,
msp->ms_id * sizeof (uint64_t), sizeof (uint64_t), &object, 0));
VERIFY3U(object, ==, space_map_object(msp->ms_sm));
mutex_exit(&msp->ms_sync_lock);
dmu_tx_commit(tx);
}
static void
metaslab_evict(metaslab_t *msp, uint64_t txg)
{
if (!msp->ms_loaded || msp->ms_disabled != 0)
return;
for (int t = 1; t < TXG_CONCURRENT_STATES; t++) {
VERIFY0(range_tree_space(
msp->ms_allocating[(txg + t) & TXG_MASK]));
}
if (msp->ms_allocator != -1)
metaslab_passivate(msp, msp->ms_weight & ~METASLAB_ACTIVE_MASK);
if (!metaslab_debug_unload)
metaslab_unload(msp);
}
/*
* Called after a transaction group has completely synced to mark
* all of the metaslab's free space as usable.
*/
void
metaslab_sync_done(metaslab_t *msp, uint64_t txg)
{
metaslab_group_t *mg = msp->ms_group;
vdev_t *vd = mg->mg_vd;
spa_t *spa = vd->vdev_spa;
range_tree_t **defer_tree;
int64_t alloc_delta, defer_delta;
boolean_t defer_allowed = B_TRUE;
ASSERT(!vd->vdev_ishole);
mutex_enter(&msp->ms_lock);
if (msp->ms_new) {
/* this is a new metaslab, add its capacity to the vdev */
metaslab_space_update(vd, mg->mg_class, 0, 0, msp->ms_size);
/* there should be no allocations nor frees at this point */
VERIFY0(msp->ms_allocated_this_txg);
VERIFY0(range_tree_space(msp->ms_freed));
}
ASSERT0(range_tree_space(msp->ms_freeing));
ASSERT0(range_tree_space(msp->ms_checkpointing));
defer_tree = &msp->ms_defer[txg % TXG_DEFER_SIZE];
uint64_t free_space = metaslab_class_get_space(spa_normal_class(spa)) -
metaslab_class_get_alloc(spa_normal_class(spa));
if (free_space <= spa_get_slop_space(spa) || vd->vdev_removing) {
defer_allowed = B_FALSE;
}
defer_delta = 0;
alloc_delta = msp->ms_allocated_this_txg -
range_tree_space(msp->ms_freed);
if (defer_allowed) {
defer_delta = range_tree_space(msp->ms_freed) -
range_tree_space(*defer_tree);
} else {
defer_delta -= range_tree_space(*defer_tree);
}
metaslab_space_update(vd, mg->mg_class, alloc_delta + defer_delta,
defer_delta, 0);
if (spa_syncing_log_sm(spa) == NULL) {
/*
* If there's a metaslab_load() in progress and we don't have
* a log space map, it means that we probably wrote to the
* metaslab's space map. If this is the case, we need to
* make sure that we wait for the load to complete so that we
* have a consistent view at the in-core side of the metaslab.
*/
metaslab_load_wait(msp);
} else {
ASSERT(spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP));
}
/*
* When auto-trimming is enabled, free ranges which are added to
* ms_allocatable are also be added to ms_trim. The ms_trim tree is
* periodically consumed by the vdev_autotrim_thread() which issues
* trims for all ranges and then vacates the tree. The ms_trim tree
* can be discarded at any time with the sole consequence of recent
* frees not being trimmed.
*/
if (spa_get_autotrim(spa) == SPA_AUTOTRIM_ON) {
range_tree_walk(*defer_tree, range_tree_add, msp->ms_trim);
if (!defer_allowed) {
range_tree_walk(msp->ms_freed, range_tree_add,
msp->ms_trim);
}
} else {
range_tree_vacate(msp->ms_trim, NULL, NULL);
}
/*
* Move the frees from the defer_tree back to the free
* range tree (if it's loaded). Swap the freed_tree and
* the defer_tree -- this is safe to do because we've
* just emptied out the defer_tree.
*/
range_tree_vacate(*defer_tree,
msp->ms_loaded ? range_tree_add : NULL, msp->ms_allocatable);
if (defer_allowed) {
range_tree_swap(&msp->ms_freed, defer_tree);
} else {
range_tree_vacate(msp->ms_freed,
msp->ms_loaded ? range_tree_add : NULL,
msp->ms_allocatable);
}
msp->ms_synced_length = space_map_length(msp->ms_sm);
msp->ms_deferspace += defer_delta;
ASSERT3S(msp->ms_deferspace, >=, 0);
ASSERT3S(msp->ms_deferspace, <=, msp->ms_size);
if (msp->ms_deferspace != 0) {
/*
* Keep syncing this metaslab until all deferred frees
* are back in circulation.
*/
vdev_dirty(vd, VDD_METASLAB, msp, txg + 1);
}
metaslab_aux_histograms_update_done(msp, defer_allowed);
if (msp->ms_new) {
msp->ms_new = B_FALSE;
mutex_enter(&mg->mg_lock);
mg->mg_ms_ready++;
mutex_exit(&mg->mg_lock);
}
/*
* Re-sort metaslab within its group now that we've adjusted
* its allocatable space.
*/
metaslab_recalculate_weight_and_sort(msp);
ASSERT0(range_tree_space(msp->ms_allocating[txg & TXG_MASK]));
ASSERT0(range_tree_space(msp->ms_freeing));
ASSERT0(range_tree_space(msp->ms_freed));
ASSERT0(range_tree_space(msp->ms_checkpointing));
msp->ms_allocating_total -= msp->ms_allocated_this_txg;
msp->ms_allocated_this_txg = 0;
mutex_exit(&msp->ms_lock);
}
void
metaslab_sync_reassess(metaslab_group_t *mg)
{
spa_t *spa = mg->mg_class->mc_spa;
spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER);
metaslab_group_alloc_update(mg);
mg->mg_fragmentation = metaslab_group_fragmentation(mg);
/*
* Preload the next potential metaslabs but only on active
* metaslab groups. We can get into a state where the metaslab
* is no longer active since we dirty metaslabs as we remove a
* a device, thus potentially making the metaslab group eligible
* for preloading.
*/
if (mg->mg_activation_count > 0) {
metaslab_group_preload(mg);
}
spa_config_exit(spa, SCL_ALLOC, FTAG);
}
/*
* When writing a ditto block (i.e. more than one DVA for a given BP) on
* the same vdev as an existing DVA of this BP, then try to allocate it
* on a different metaslab than existing DVAs (i.e. a unique metaslab).
*/
static boolean_t
metaslab_is_unique(metaslab_t *msp, dva_t *dva)
{
uint64_t dva_ms_id;
if (DVA_GET_ASIZE(dva) == 0)
return (B_TRUE);
if (msp->ms_group->mg_vd->vdev_id != DVA_GET_VDEV(dva))
return (B_TRUE);
dva_ms_id = DVA_GET_OFFSET(dva) >> msp->ms_group->mg_vd->vdev_ms_shift;
return (msp->ms_id != dva_ms_id);
}
/*
* ==========================================================================
* Metaslab allocation tracing facility
* ==========================================================================
*/
/*
* Add an allocation trace element to the allocation tracing list.
*/
static void
metaslab_trace_add(zio_alloc_list_t *zal, metaslab_group_t *mg,
metaslab_t *msp, uint64_t psize, uint32_t dva_id, uint64_t offset,
int allocator)
{
metaslab_alloc_trace_t *mat;
if (!metaslab_trace_enabled)
return;
/*
* When the tracing list reaches its maximum we remove
* the second element in the list before adding a new one.
* By removing the second element we preserve the original
* entry as a clue to what allocations steps have already been
* performed.
*/
if (zal->zal_size == metaslab_trace_max_entries) {
metaslab_alloc_trace_t *mat_next;
#ifdef ZFS_DEBUG
panic("too many entries in allocation list");
#endif
METASLABSTAT_BUMP(metaslabstat_trace_over_limit);
zal->zal_size--;
mat_next = list_next(&zal->zal_list, list_head(&zal->zal_list));
list_remove(&zal->zal_list, mat_next);
kmem_cache_free(metaslab_alloc_trace_cache, mat_next);
}
mat = kmem_cache_alloc(metaslab_alloc_trace_cache, KM_SLEEP);
list_link_init(&mat->mat_list_node);
mat->mat_mg = mg;
mat->mat_msp = msp;
mat->mat_size = psize;
mat->mat_dva_id = dva_id;
mat->mat_offset = offset;
mat->mat_weight = 0;
mat->mat_allocator = allocator;
if (msp != NULL)
mat->mat_weight = msp->ms_weight;
/*
* The list is part of the zio so locking is not required. Only
* a single thread will perform allocations for a given zio.
*/
list_insert_tail(&zal->zal_list, mat);
zal->zal_size++;
ASSERT3U(zal->zal_size, <=, metaslab_trace_max_entries);
}
void
metaslab_trace_init(zio_alloc_list_t *zal)
{
list_create(&zal->zal_list, sizeof (metaslab_alloc_trace_t),
offsetof(metaslab_alloc_trace_t, mat_list_node));
zal->zal_size = 0;
}
void
metaslab_trace_fini(zio_alloc_list_t *zal)
{
metaslab_alloc_trace_t *mat;
while ((mat = list_remove_head(&zal->zal_list)) != NULL)
kmem_cache_free(metaslab_alloc_trace_cache, mat);
list_destroy(&zal->zal_list);
zal->zal_size = 0;
}
/*
* ==========================================================================
* Metaslab block operations
* ==========================================================================
*/
static void
metaslab_group_alloc_increment(spa_t *spa, uint64_t vdev, void *tag, int flags,
int allocator)
{
if (!(flags & METASLAB_ASYNC_ALLOC) ||
(flags & METASLAB_DONT_THROTTLE))
return;
metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg;
if (!mg->mg_class->mc_alloc_throttle_enabled)
return;
metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator];
(void) zfs_refcount_add(&mga->mga_alloc_queue_depth, tag);
}
static void
metaslab_group_increment_qdepth(metaslab_group_t *mg, int allocator)
{
metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator];
metaslab_class_allocator_t *mca =
&mg->mg_class->mc_allocator[allocator];
uint64_t max = mg->mg_max_alloc_queue_depth;
uint64_t cur = mga->mga_cur_max_alloc_queue_depth;
while (cur < max) {
if (atomic_cas_64(&mga->mga_cur_max_alloc_queue_depth,
cur, cur + 1) == cur) {
atomic_inc_64(&mca->mca_alloc_max_slots);
return;
}
cur = mga->mga_cur_max_alloc_queue_depth;
}
}
void
metaslab_group_alloc_decrement(spa_t *spa, uint64_t vdev, void *tag, int flags,
int allocator, boolean_t io_complete)
{
if (!(flags & METASLAB_ASYNC_ALLOC) ||
(flags & METASLAB_DONT_THROTTLE))
return;
metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg;
if (!mg->mg_class->mc_alloc_throttle_enabled)
return;
metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator];
(void) zfs_refcount_remove(&mga->mga_alloc_queue_depth, tag);
if (io_complete)
metaslab_group_increment_qdepth(mg, allocator);
}
void
metaslab_group_alloc_verify(spa_t *spa, const blkptr_t *bp, void *tag,
int allocator)
{
#ifdef ZFS_DEBUG
const dva_t *dva = bp->blk_dva;
int ndvas = BP_GET_NDVAS(bp);
for (int d = 0; d < ndvas; d++) {
uint64_t vdev = DVA_GET_VDEV(&dva[d]);
metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg;
metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator];
VERIFY(zfs_refcount_not_held(&mga->mga_alloc_queue_depth, tag));
}
#endif
}
static uint64_t
metaslab_block_alloc(metaslab_t *msp, uint64_t size, uint64_t txg)
{
uint64_t start;
range_tree_t *rt = msp->ms_allocatable;
metaslab_class_t *mc = msp->ms_group->mg_class;
ASSERT(MUTEX_HELD(&msp->ms_lock));
VERIFY(!msp->ms_condensing);
VERIFY0(msp->ms_disabled);
start = mc->mc_ops->msop_alloc(msp, size);
if (start != -1ULL) {
metaslab_group_t *mg = msp->ms_group;
vdev_t *vd = mg->mg_vd;
VERIFY0(P2PHASE(start, 1ULL << vd->vdev_ashift));
VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
VERIFY3U(range_tree_space(rt) - size, <=, msp->ms_size);
range_tree_remove(rt, start, size);
range_tree_clear(msp->ms_trim, start, size);
if (range_tree_is_empty(msp->ms_allocating[txg & TXG_MASK]))
vdev_dirty(mg->mg_vd, VDD_METASLAB, msp, txg);
range_tree_add(msp->ms_allocating[txg & TXG_MASK], start, size);
msp->ms_allocating_total += size;
/* Track the last successful allocation */
msp->ms_alloc_txg = txg;
metaslab_verify_space(msp, txg);
}
/*
* Now that we've attempted the allocation we need to update the
* metaslab's maximum block size since it may have changed.
*/
msp->ms_max_size = metaslab_largest_allocatable(msp);
return (start);
}
/*
* Find the metaslab with the highest weight that is less than what we've
* already tried. In the common case, this means that we will examine each
* metaslab at most once. Note that concurrent callers could reorder metaslabs
* by activation/passivation once we have dropped the mg_lock. If a metaslab is
* activated by another thread, and we fail to allocate from the metaslab we
* have selected, we may not try the newly-activated metaslab, and instead
* activate another metaslab. This is not optimal, but generally does not cause
* any problems (a possible exception being if every metaslab is completely full
* except for the newly-activated metaslab which we fail to examine).
*/
static metaslab_t *
find_valid_metaslab(metaslab_group_t *mg, uint64_t activation_weight,
dva_t *dva, int d, boolean_t want_unique, uint64_t asize, int allocator,
boolean_t try_hard, zio_alloc_list_t *zal, metaslab_t *search,
boolean_t *was_active)
{
avl_index_t idx;
avl_tree_t *t = &mg->mg_metaslab_tree;
metaslab_t *msp = avl_find(t, search, &idx);
if (msp == NULL)
msp = avl_nearest(t, idx, AVL_AFTER);
int tries = 0;
for (; msp != NULL; msp = AVL_NEXT(t, msp)) {
int i;
if (!try_hard && tries > zfs_metaslab_find_max_tries) {
METASLABSTAT_BUMP(metaslabstat_too_many_tries);
return (NULL);
}
tries++;
if (!metaslab_should_allocate(msp, asize, try_hard)) {
metaslab_trace_add(zal, mg, msp, asize, d,
TRACE_TOO_SMALL, allocator);
continue;
}
/*
* If the selected metaslab is condensing or disabled,
* skip it.
*/
if (msp->ms_condensing || msp->ms_disabled > 0)
continue;
*was_active = msp->ms_allocator != -1;
/*
* If we're activating as primary, this is our first allocation
* from this disk, so we don't need to check how close we are.
* If the metaslab under consideration was already active,
* we're getting desperate enough to steal another allocator's
* metaslab, so we still don't care about distances.
*/
if (activation_weight == METASLAB_WEIGHT_PRIMARY || *was_active)
break;
for (i = 0; i < d; i++) {
if (want_unique &&
!metaslab_is_unique(msp, &dva[i]))
break; /* try another metaslab */
}
if (i == d)
break;
}
if (msp != NULL) {
search->ms_weight = msp->ms_weight;
search->ms_start = msp->ms_start + 1;
search->ms_allocator = msp->ms_allocator;
search->ms_primary = msp->ms_primary;
}
return (msp);
}
static void
metaslab_active_mask_verify(metaslab_t *msp)
{
ASSERT(MUTEX_HELD(&msp->ms_lock));
if ((zfs_flags & ZFS_DEBUG_METASLAB_VERIFY) == 0)
return;
if ((msp->ms_weight & METASLAB_ACTIVE_MASK) == 0)
return;
if (msp->ms_weight & METASLAB_WEIGHT_PRIMARY) {
VERIFY0(msp->ms_weight & METASLAB_WEIGHT_SECONDARY);
VERIFY0(msp->ms_weight & METASLAB_WEIGHT_CLAIM);
VERIFY3S(msp->ms_allocator, !=, -1);
VERIFY(msp->ms_primary);
return;
}
if (msp->ms_weight & METASLAB_WEIGHT_SECONDARY) {
VERIFY0(msp->ms_weight & METASLAB_WEIGHT_PRIMARY);
VERIFY0(msp->ms_weight & METASLAB_WEIGHT_CLAIM);
VERIFY3S(msp->ms_allocator, !=, -1);
VERIFY(!msp->ms_primary);
return;
}
if (msp->ms_weight & METASLAB_WEIGHT_CLAIM) {
VERIFY0(msp->ms_weight & METASLAB_WEIGHT_PRIMARY);
VERIFY0(msp->ms_weight & METASLAB_WEIGHT_SECONDARY);
VERIFY3S(msp->ms_allocator, ==, -1);
return;
}
}
/* ARGSUSED */
static uint64_t
metaslab_group_alloc_normal(metaslab_group_t *mg, zio_alloc_list_t *zal,
uint64_t asize, uint64_t txg, boolean_t want_unique, dva_t *dva, int d,
int allocator, boolean_t try_hard)
{
metaslab_t *msp = NULL;
uint64_t offset = -1ULL;
uint64_t activation_weight = METASLAB_WEIGHT_PRIMARY;
for (int i = 0; i < d; i++) {
if (activation_weight == METASLAB_WEIGHT_PRIMARY &&
DVA_GET_VDEV(&dva[i]) == mg->mg_vd->vdev_id) {
activation_weight = METASLAB_WEIGHT_SECONDARY;
} else if (activation_weight == METASLAB_WEIGHT_SECONDARY &&
DVA_GET_VDEV(&dva[i]) == mg->mg_vd->vdev_id) {
activation_weight = METASLAB_WEIGHT_CLAIM;
break;
}
}
/*
* If we don't have enough metaslabs active to fill the entire array, we
* just use the 0th slot.
*/
if (mg->mg_ms_ready < mg->mg_allocators * 3)
allocator = 0;
metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator];
ASSERT3U(mg->mg_vd->vdev_ms_count, >=, 2);
metaslab_t *search = kmem_alloc(sizeof (*search), KM_SLEEP);
search->ms_weight = UINT64_MAX;
search->ms_start = 0;
/*
* At the end of the metaslab tree are the already-active metaslabs,
* first the primaries, then the secondaries. When we resume searching
* through the tree, we need to consider ms_allocator and ms_primary so
* we start in the location right after where we left off, and don't
* accidentally loop forever considering the same metaslabs.
*/
search->ms_allocator = -1;
search->ms_primary = B_TRUE;
for (;;) {
boolean_t was_active = B_FALSE;
mutex_enter(&mg->mg_lock);
if (activation_weight == METASLAB_WEIGHT_PRIMARY &&
mga->mga_primary != NULL) {
msp = mga->mga_primary;
/*
* Even though we don't hold the ms_lock for the
* primary metaslab, those fields should not
* change while we hold the mg_lock. Thus it is
* safe to make assertions on them.
*/
ASSERT(msp->ms_primary);
ASSERT3S(msp->ms_allocator, ==, allocator);
ASSERT(msp->ms_loaded);
was_active = B_TRUE;
ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK);
} else if (activation_weight == METASLAB_WEIGHT_SECONDARY &&
mga->mga_secondary != NULL) {
msp = mga->mga_secondary;
/*
* See comment above about the similar assertions
* for the primary metaslab.
*/
ASSERT(!msp->ms_primary);
ASSERT3S(msp->ms_allocator, ==, allocator);
ASSERT(msp->ms_loaded);
was_active = B_TRUE;
ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK);
} else {
msp = find_valid_metaslab(mg, activation_weight, dva, d,
want_unique, asize, allocator, try_hard, zal,
search, &was_active);
}
mutex_exit(&mg->mg_lock);
if (msp == NULL) {
kmem_free(search, sizeof (*search));
return (-1ULL);
}
mutex_enter(&msp->ms_lock);
metaslab_active_mask_verify(msp);
/*
* This code is disabled out because of issues with
* tracepoints in non-gpl kernel modules.
*/
#if 0
DTRACE_PROBE3(ms__activation__attempt,
metaslab_t *, msp, uint64_t, activation_weight,
boolean_t, was_active);
#endif
/*
* Ensure that the metaslab we have selected is still
* capable of handling our request. It's possible that
* another thread may have changed the weight while we
* were blocked on the metaslab lock. We check the
* active status first to see if we need to set_selected_txg
* a new metaslab.
*/
if (was_active && !(msp->ms_weight & METASLAB_ACTIVE_MASK)) {
ASSERT3S(msp->ms_allocator, ==, -1);
mutex_exit(&msp->ms_lock);
continue;
}
/*
* If the metaslab was activated for another allocator
* while we were waiting in the ms_lock above, or it's
* a primary and we're seeking a secondary (or vice versa),
* we go back and select a new metaslab.
*/
if (!was_active && (msp->ms_weight & METASLAB_ACTIVE_MASK) &&
(msp->ms_allocator != -1) &&
(msp->ms_allocator != allocator || ((activation_weight ==
METASLAB_WEIGHT_PRIMARY) != msp->ms_primary))) {
ASSERT(msp->ms_loaded);
ASSERT((msp->ms_weight & METASLAB_WEIGHT_CLAIM) ||
msp->ms_allocator != -1);
mutex_exit(&msp->ms_lock);
continue;
}
/*
* This metaslab was used for claiming regions allocated
* by the ZIL during pool import. Once these regions are
* claimed we don't need to keep the CLAIM bit set
* anymore. Passivate this metaslab to zero its activation
* mask.
*/
if (msp->ms_weight & METASLAB_WEIGHT_CLAIM &&
activation_weight != METASLAB_WEIGHT_CLAIM) {
ASSERT(msp->ms_loaded);
ASSERT3S(msp->ms_allocator, ==, -1);
metaslab_passivate(msp, msp->ms_weight &
~METASLAB_WEIGHT_CLAIM);
mutex_exit(&msp->ms_lock);
continue;
}
metaslab_set_selected_txg(msp, txg);
int activation_error =
metaslab_activate(msp, allocator, activation_weight);
metaslab_active_mask_verify(msp);
/*
* If the metaslab was activated by another thread for
* another allocator or activation_weight (EBUSY), or it
* failed because another metaslab was assigned as primary
* for this allocator (EEXIST) we continue using this
* metaslab for our allocation, rather than going on to a
* worse metaslab (we waited for that metaslab to be loaded
* after all).
*
* If the activation failed due to an I/O error or ENOSPC we
* skip to the next metaslab.
*/
boolean_t activated;
if (activation_error == 0) {
activated = B_TRUE;
} else if (activation_error == EBUSY ||
activation_error == EEXIST) {
activated = B_FALSE;
} else {
mutex_exit(&msp->ms_lock);
continue;
}
ASSERT(msp->ms_loaded);
/*
* Now that we have the lock, recheck to see if we should
* continue to use this metaslab for this allocation. The
* the metaslab is now loaded so metaslab_should_allocate()
* can accurately determine if the allocation attempt should
* proceed.
*/
if (!metaslab_should_allocate(msp, asize, try_hard)) {
/* Passivate this metaslab and select a new one. */
metaslab_trace_add(zal, mg, msp, asize, d,
TRACE_TOO_SMALL, allocator);
goto next;
}
/*
* If this metaslab is currently condensing then pick again
* as we can't manipulate this metaslab until it's committed
* to disk. If this metaslab is being initialized, we shouldn't
* allocate from it since the allocated region might be
* overwritten after allocation.
*/
if (msp->ms_condensing) {
metaslab_trace_add(zal, mg, msp, asize, d,
TRACE_CONDENSING, allocator);
if (activated) {
metaslab_passivate(msp, msp->ms_weight &
~METASLAB_ACTIVE_MASK);
}
mutex_exit(&msp->ms_lock);
continue;
} else if (msp->ms_disabled > 0) {
metaslab_trace_add(zal, mg, msp, asize, d,
TRACE_DISABLED, allocator);
if (activated) {
metaslab_passivate(msp, msp->ms_weight &
~METASLAB_ACTIVE_MASK);
}
mutex_exit(&msp->ms_lock);
continue;
}
offset = metaslab_block_alloc(msp, asize, txg);
metaslab_trace_add(zal, mg, msp, asize, d, offset, allocator);
if (offset != -1ULL) {
/* Proactively passivate the metaslab, if needed */
if (activated)
metaslab_segment_may_passivate(msp);
break;
}
next:
ASSERT(msp->ms_loaded);
/*
* This code is disabled out because of issues with
* tracepoints in non-gpl kernel modules.
*/
#if 0
DTRACE_PROBE2(ms__alloc__failure, metaslab_t *, msp,
uint64_t, asize);
#endif
/*
* We were unable to allocate from this metaslab so determine
* a new weight for this metaslab. Now that we have loaded
* the metaslab we can provide a better hint to the metaslab
* selector.
*
* For space-based metaslabs, we use the maximum block size.
* This information is only available when the metaslab
* is loaded and is more accurate than the generic free
* space weight that was calculated by metaslab_weight().
* This information allows us to quickly compare the maximum
* available allocation in the metaslab to the allocation
* size being requested.
*
* For segment-based metaslabs, determine the new weight
* based on the highest bucket in the range tree. We
* explicitly use the loaded segment weight (i.e. the range
* tree histogram) since it contains the space that is
* currently available for allocation and is accurate
* even within a sync pass.
*/
uint64_t weight;
if (WEIGHT_IS_SPACEBASED(msp->ms_weight)) {
weight = metaslab_largest_allocatable(msp);
WEIGHT_SET_SPACEBASED(weight);
} else {
weight = metaslab_weight_from_range_tree(msp);
}
if (activated) {
metaslab_passivate(msp, weight);
} else {
/*
* For the case where we use the metaslab that is
* active for another allocator we want to make
* sure that we retain the activation mask.
*
* Note that we could attempt to use something like
* metaslab_recalculate_weight_and_sort() that
* retains the activation mask here. That function
* uses metaslab_weight() to set the weight though
* which is not as accurate as the calculations
* above.
*/
weight |= msp->ms_weight & METASLAB_ACTIVE_MASK;
metaslab_group_sort(mg, msp, weight);
}
metaslab_active_mask_verify(msp);
/*
* We have just failed an allocation attempt, check
* that metaslab_should_allocate() agrees. Otherwise,
* we may end up in an infinite loop retrying the same
* metaslab.
*/
ASSERT(!metaslab_should_allocate(msp, asize, try_hard));
mutex_exit(&msp->ms_lock);
}
mutex_exit(&msp->ms_lock);
kmem_free(search, sizeof (*search));
return (offset);
}
static uint64_t
metaslab_group_alloc(metaslab_group_t *mg, zio_alloc_list_t *zal,
uint64_t asize, uint64_t txg, boolean_t want_unique, dva_t *dva, int d,
int allocator, boolean_t try_hard)
{
uint64_t offset;
ASSERT(mg->mg_initialized);
offset = metaslab_group_alloc_normal(mg, zal, asize, txg, want_unique,
dva, d, allocator, try_hard);
mutex_enter(&mg->mg_lock);
if (offset == -1ULL) {
mg->mg_failed_allocations++;
metaslab_trace_add(zal, mg, NULL, asize, d,
TRACE_GROUP_FAILURE, allocator);
if (asize == SPA_GANGBLOCKSIZE) {
/*
* This metaslab group was unable to allocate
* the minimum gang block size so it must be out of
* space. We must notify the allocation throttle
* to start skipping allocation attempts to this
* metaslab group until more space becomes available.
* Note: this failure cannot be caused by the
* allocation throttle since the allocation throttle
* is only responsible for skipping devices and
* not failing block allocations.
*/
mg->mg_no_free_space = B_TRUE;
}
}
mg->mg_allocations++;
mutex_exit(&mg->mg_lock);
return (offset);
}
/*
* Allocate a block for the specified i/o.
*/
int
metaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize,
dva_t *dva, int d, dva_t *hintdva, uint64_t txg, int flags,
zio_alloc_list_t *zal, int allocator)
{
metaslab_class_allocator_t *mca = &mc->mc_allocator[allocator];
metaslab_group_t *mg, *fast_mg, *rotor;
vdev_t *vd;
boolean_t try_hard = B_FALSE;
ASSERT(!DVA_IS_VALID(&dva[d]));
/*
* For testing, make some blocks above a certain size be gang blocks.
* This will result in more split blocks when using device removal,
* and a large number of split blocks coupled with ztest-induced
* damage can result in extremely long reconstruction times. This
* will also test spilling from special to normal.
*/
if (psize >= metaslab_force_ganging && (random_in_range(100) < 3)) {
metaslab_trace_add(zal, NULL, NULL, psize, d, TRACE_FORCE_GANG,
allocator);
return (SET_ERROR(ENOSPC));
}
/*
* Start at the rotor and loop through all mgs until we find something.
* Note that there's no locking on mca_rotor or mca_aliquot because
* nothing actually breaks if we miss a few updates -- we just won't
* allocate quite as evenly. It all balances out over time.
*
* If we are doing ditto or log blocks, try to spread them across
* consecutive vdevs. If we're forced to reuse a vdev before we've
* allocated all of our ditto blocks, then try and spread them out on
* that vdev as much as possible. If it turns out to not be possible,
* gradually lower our standards until anything becomes acceptable.
* Also, allocating on consecutive vdevs (as opposed to random vdevs)
* gives us hope of containing our fault domains to something we're
* able to reason about. Otherwise, any two top-level vdev failures
* will guarantee the loss of data. With consecutive allocation,
* only two adjacent top-level vdev failures will result in data loss.
*
* If we are doing gang blocks (hintdva is non-NULL), try to keep
* ourselves on the same vdev as our gang block header. That
* way, we can hope for locality in vdev_cache, plus it makes our
* fault domains something tractable.
*/
if (hintdva) {
vd = vdev_lookup_top(spa, DVA_GET_VDEV(&hintdva[d]));
/*
* It's possible the vdev we're using as the hint no
* longer exists or its mg has been closed (e.g. by
* device removal). Consult the rotor when
* all else fails.
*/
if (vd != NULL && vd->vdev_mg != NULL) {
mg = vdev_get_mg(vd, mc);
if (flags & METASLAB_HINTBP_AVOID &&
mg->mg_next != NULL)
mg = mg->mg_next;
} else {
mg = mca->mca_rotor;
}
} else if (d != 0) {
vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d - 1]));
mg = vd->vdev_mg->mg_next;
} else if (flags & METASLAB_FASTWRITE) {
mg = fast_mg = mca->mca_rotor;
do {
if (fast_mg->mg_vd->vdev_pending_fastwrite <
mg->mg_vd->vdev_pending_fastwrite)
mg = fast_mg;
} while ((fast_mg = fast_mg->mg_next) != mca->mca_rotor);
} else {
ASSERT(mca->mca_rotor != NULL);
mg = mca->mca_rotor;
}
/*
* If the hint put us into the wrong metaslab class, or into a
* metaslab group that has been passivated, just follow the rotor.
*/
if (mg->mg_class != mc || mg->mg_activation_count <= 0)
mg = mca->mca_rotor;
rotor = mg;
top:
do {
boolean_t allocatable;
ASSERT(mg->mg_activation_count == 1);
vd = mg->mg_vd;
/*
* Don't allocate from faulted devices.
*/
if (try_hard) {
spa_config_enter(spa, SCL_ZIO, FTAG, RW_READER);
allocatable = vdev_allocatable(vd);
spa_config_exit(spa, SCL_ZIO, FTAG);
} else {
allocatable = vdev_allocatable(vd);
}
/*
* Determine if the selected metaslab group is eligible
* for allocations. If we're ganging then don't allow
* this metaslab group to skip allocations since that would
* inadvertently return ENOSPC and suspend the pool
* even though space is still available.
*/
if (allocatable && !GANG_ALLOCATION(flags) && !try_hard) {
allocatable = metaslab_group_allocatable(mg, rotor,
psize, allocator, d);
}
if (!allocatable) {
metaslab_trace_add(zal, mg, NULL, psize, d,
TRACE_NOT_ALLOCATABLE, allocator);
goto next;
}
ASSERT(mg->mg_initialized);
/*
* Avoid writing single-copy data to a failing,
* non-redundant vdev, unless we've already tried all
* other vdevs.
*/
if ((vd->vdev_stat.vs_write_errors > 0 ||
vd->vdev_state < VDEV_STATE_HEALTHY) &&
d == 0 && !try_hard && vd->vdev_children == 0) {
metaslab_trace_add(zal, mg, NULL, psize, d,
TRACE_VDEV_ERROR, allocator);
goto next;
}
ASSERT(mg->mg_class == mc);
uint64_t asize = vdev_psize_to_asize(vd, psize);
ASSERT(P2PHASE(asize, 1ULL << vd->vdev_ashift) == 0);
/*
* If we don't need to try hard, then require that the
* block be on a different metaslab from any other DVAs
* in this BP (unique=true). If we are trying hard, then
* allow any metaslab to be used (unique=false).
*/
uint64_t offset = metaslab_group_alloc(mg, zal, asize, txg,
!try_hard, dva, d, allocator, try_hard);
if (offset != -1ULL) {
/*
* If we've just selected this metaslab group,
* figure out whether the corresponding vdev is
* over- or under-used relative to the pool,
* and set an allocation bias to even it out.
*
* Bias is also used to compensate for unequally
* sized vdevs so that space is allocated fairly.
*/
if (mca->mca_aliquot == 0 && metaslab_bias_enabled) {
vdev_stat_t *vs = &vd->vdev_stat;
int64_t vs_free = vs->vs_space - vs->vs_alloc;
int64_t mc_free = mc->mc_space - mc->mc_alloc;
int64_t ratio;
/*
* Calculate how much more or less we should
* try to allocate from this device during
* this iteration around the rotor.
*
* This basically introduces a zero-centered
* bias towards the devices with the most
* free space, while compensating for vdev
* size differences.
*
* Examples:
* vdev V1 = 16M/128M
* vdev V2 = 16M/128M
* ratio(V1) = 100% ratio(V2) = 100%
*
* vdev V1 = 16M/128M
* vdev V2 = 64M/128M
* ratio(V1) = 127% ratio(V2) = 72%
*
* vdev V1 = 16M/128M
* vdev V2 = 64M/512M
* ratio(V1) = 40% ratio(V2) = 160%
*/
ratio = (vs_free * mc->mc_alloc_groups * 100) /
(mc_free + 1);
mg->mg_bias = ((ratio - 100) *
(int64_t)mg->mg_aliquot) / 100;
} else if (!metaslab_bias_enabled) {
mg->mg_bias = 0;
}
if ((flags & METASLAB_FASTWRITE) ||
atomic_add_64_nv(&mca->mca_aliquot, asize) >=
mg->mg_aliquot + mg->mg_bias) {
mca->mca_rotor = mg->mg_next;
mca->mca_aliquot = 0;
}
DVA_SET_VDEV(&dva[d], vd->vdev_id);
DVA_SET_OFFSET(&dva[d], offset);
DVA_SET_GANG(&dva[d],
((flags & METASLAB_GANG_HEADER) ? 1 : 0));
DVA_SET_ASIZE(&dva[d], asize);
if (flags & METASLAB_FASTWRITE) {
atomic_add_64(&vd->vdev_pending_fastwrite,
psize);
}
return (0);
}
next:
mca->mca_rotor = mg->mg_next;
mca->mca_aliquot = 0;
} while ((mg = mg->mg_next) != rotor);
/*
* If we haven't tried hard, perhaps do so now.
*/
if (!try_hard && (zfs_metaslab_try_hard_before_gang ||
GANG_ALLOCATION(flags) || (flags & METASLAB_ZIL) != 0 ||
psize <= 1 << spa->spa_min_ashift)) {
METASLABSTAT_BUMP(metaslabstat_try_hard);
try_hard = B_TRUE;
goto top;
}
bzero(&dva[d], sizeof (dva_t));
metaslab_trace_add(zal, rotor, NULL, psize, d, TRACE_ENOSPC, allocator);
return (SET_ERROR(ENOSPC));
}
void
metaslab_free_concrete(vdev_t *vd, uint64_t offset, uint64_t asize,
boolean_t checkpoint)
{
metaslab_t *msp;
spa_t *spa = vd->vdev_spa;
ASSERT(vdev_is_concrete(vd));
ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0);
ASSERT3U(offset >> vd->vdev_ms_shift, <, vd->vdev_ms_count);
msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
VERIFY(!msp->ms_condensing);
VERIFY3U(offset, >=, msp->ms_start);
VERIFY3U(offset + asize, <=, msp->ms_start + msp->ms_size);
VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift));
VERIFY0(P2PHASE(asize, 1ULL << vd->vdev_ashift));
metaslab_check_free_impl(vd, offset, asize);
mutex_enter(&msp->ms_lock);
if (range_tree_is_empty(msp->ms_freeing) &&
range_tree_is_empty(msp->ms_checkpointing)) {
vdev_dirty(vd, VDD_METASLAB, msp, spa_syncing_txg(spa));
}
if (checkpoint) {
ASSERT(spa_has_checkpoint(spa));
range_tree_add(msp->ms_checkpointing, offset, asize);
} else {
range_tree_add(msp->ms_freeing, offset, asize);
}
mutex_exit(&msp->ms_lock);
}
/* ARGSUSED */
void
metaslab_free_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset,
uint64_t size, void *arg)
{
boolean_t *checkpoint = arg;
ASSERT3P(checkpoint, !=, NULL);
if (vd->vdev_ops->vdev_op_remap != NULL)
vdev_indirect_mark_obsolete(vd, offset, size);
else
metaslab_free_impl(vd, offset, size, *checkpoint);
}
static void
metaslab_free_impl(vdev_t *vd, uint64_t offset, uint64_t size,
boolean_t checkpoint)
{
spa_t *spa = vd->vdev_spa;
ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0);
if (spa_syncing_txg(spa) > spa_freeze_txg(spa))
return;
if (spa->spa_vdev_removal != NULL &&
spa->spa_vdev_removal->svr_vdev_id == vd->vdev_id &&
vdev_is_concrete(vd)) {
/*
* Note: we check if the vdev is concrete because when
* we complete the removal, we first change the vdev to be
* an indirect vdev (in open context), and then (in syncing
* context) clear spa_vdev_removal.
*/
free_from_removing_vdev(vd, offset, size);
} else if (vd->vdev_ops->vdev_op_remap != NULL) {
vdev_indirect_mark_obsolete(vd, offset, size);
vd->vdev_ops->vdev_op_remap(vd, offset, size,
metaslab_free_impl_cb, &checkpoint);
} else {
metaslab_free_concrete(vd, offset, size, checkpoint);
}
}
typedef struct remap_blkptr_cb_arg {
blkptr_t *rbca_bp;
spa_remap_cb_t rbca_cb;
vdev_t *rbca_remap_vd;
uint64_t rbca_remap_offset;
void *rbca_cb_arg;
} remap_blkptr_cb_arg_t;
static void
remap_blkptr_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset,
uint64_t size, void *arg)
{
remap_blkptr_cb_arg_t *rbca = arg;
blkptr_t *bp = rbca->rbca_bp;
/* We can not remap split blocks. */
if (size != DVA_GET_ASIZE(&bp->blk_dva[0]))
return;
ASSERT0(inner_offset);
if (rbca->rbca_cb != NULL) {
/*
* At this point we know that we are not handling split
* blocks and we invoke the callback on the previous
* vdev which must be indirect.
*/
ASSERT3P(rbca->rbca_remap_vd->vdev_ops, ==, &vdev_indirect_ops);
rbca->rbca_cb(rbca->rbca_remap_vd->vdev_id,
rbca->rbca_remap_offset, size, rbca->rbca_cb_arg);
/* set up remap_blkptr_cb_arg for the next call */
rbca->rbca_remap_vd = vd;
rbca->rbca_remap_offset = offset;
}
/*
* The phys birth time is that of dva[0]. This ensures that we know
* when each dva was written, so that resilver can determine which
* blocks need to be scrubbed (i.e. those written during the time
* the vdev was offline). It also ensures that the key used in
* the ARC hash table is unique (i.e. dva[0] + phys_birth). If
* we didn't change the phys_birth, a lookup in the ARC for a
* remapped BP could find the data that was previously stored at
* this vdev + offset.
*/
vdev_t *oldvd = vdev_lookup_top(vd->vdev_spa,
DVA_GET_VDEV(&bp->blk_dva[0]));
vdev_indirect_births_t *vib = oldvd->vdev_indirect_births;
bp->blk_phys_birth = vdev_indirect_births_physbirth(vib,
DVA_GET_OFFSET(&bp->blk_dva[0]), DVA_GET_ASIZE(&bp->blk_dva[0]));
DVA_SET_VDEV(&bp->blk_dva[0], vd->vdev_id);
DVA_SET_OFFSET(&bp->blk_dva[0], offset);
}
/*
* If the block pointer contains any indirect DVAs, modify them to refer to
* concrete DVAs. Note that this will sometimes not be possible, leaving
* the indirect DVA in place. This happens if the indirect DVA spans multiple
* segments in the mapping (i.e. it is a "split block").
*
* If the BP was remapped, calls the callback on the original dva (note the
* callback can be called multiple times if the original indirect DVA refers
* to another indirect DVA, etc).
*
* Returns TRUE if the BP was remapped.
*/
boolean_t
spa_remap_blkptr(spa_t *spa, blkptr_t *bp, spa_remap_cb_t callback, void *arg)
{
remap_blkptr_cb_arg_t rbca;
if (!zfs_remap_blkptr_enable)
return (B_FALSE);
if (!spa_feature_is_enabled(spa, SPA_FEATURE_OBSOLETE_COUNTS))
return (B_FALSE);
/*
* Dedup BP's can not be remapped, because ddt_phys_select() depends
* on DVA[0] being the same in the BP as in the DDT (dedup table).
*/
if (BP_GET_DEDUP(bp))
return (B_FALSE);
/*
* Gang blocks can not be remapped, because
* zio_checksum_gang_verifier() depends on the DVA[0] that's in
* the BP used to read the gang block header (GBH) being the same
* as the DVA[0] that we allocated for the GBH.
*/
if (BP_IS_GANG(bp))
return (B_FALSE);
/*
* Embedded BP's have no DVA to remap.
*/
if (BP_GET_NDVAS(bp) < 1)
return (B_FALSE);
/*
* Note: we only remap dva[0]. If we remapped other dvas, we
* would no longer know what their phys birth txg is.
*/
dva_t *dva = &bp->blk_dva[0];
uint64_t offset = DVA_GET_OFFSET(dva);
uint64_t size = DVA_GET_ASIZE(dva);
vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
if (vd->vdev_ops->vdev_op_remap == NULL)
return (B_FALSE);
rbca.rbca_bp = bp;
rbca.rbca_cb = callback;
rbca.rbca_remap_vd = vd;
rbca.rbca_remap_offset = offset;
rbca.rbca_cb_arg = arg;
/*
* remap_blkptr_cb() will be called in order for each level of
* indirection, until a concrete vdev is reached or a split block is
* encountered. old_vd and old_offset are updated within the callback
* as we go from the one indirect vdev to the next one (either concrete
* or indirect again) in that order.
*/
vd->vdev_ops->vdev_op_remap(vd, offset, size, remap_blkptr_cb, &rbca);
/* Check if the DVA wasn't remapped because it is a split block */
if (DVA_GET_VDEV(&rbca.rbca_bp->blk_dva[0]) == vd->vdev_id)
return (B_FALSE);
return (B_TRUE);
}
/*
* Undo the allocation of a DVA which happened in the given transaction group.
*/
void
metaslab_unalloc_dva(spa_t *spa, const dva_t *dva, uint64_t txg)
{
metaslab_t *msp;
vdev_t *vd;
uint64_t vdev = DVA_GET_VDEV(dva);
uint64_t offset = DVA_GET_OFFSET(dva);
uint64_t size = DVA_GET_ASIZE(dva);
ASSERT(DVA_IS_VALID(dva));
ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0);
if (txg > spa_freeze_txg(spa))
return;
if ((vd = vdev_lookup_top(spa, vdev)) == NULL || !DVA_IS_VALID(dva) ||
(offset >> vd->vdev_ms_shift) >= vd->vdev_ms_count) {
zfs_panic_recover("metaslab_free_dva(): bad DVA %llu:%llu:%llu",
(u_longlong_t)vdev, (u_longlong_t)offset,
(u_longlong_t)size);
return;
}
ASSERT(!vd->vdev_removing);
ASSERT(vdev_is_concrete(vd));
ASSERT0(vd->vdev_indirect_config.vic_mapping_object);
ASSERT3P(vd->vdev_indirect_mapping, ==, NULL);
if (DVA_GET_GANG(dva))
size = vdev_gang_header_asize(vd);
msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
mutex_enter(&msp->ms_lock);
range_tree_remove(msp->ms_allocating[txg & TXG_MASK],
offset, size);
msp->ms_allocating_total -= size;
VERIFY(!msp->ms_condensing);
VERIFY3U(offset, >=, msp->ms_start);
VERIFY3U(offset + size, <=, msp->ms_start + msp->ms_size);
VERIFY3U(range_tree_space(msp->ms_allocatable) + size, <=,
msp->ms_size);
VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift));
VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
range_tree_add(msp->ms_allocatable, offset, size);
mutex_exit(&msp->ms_lock);
}
/*
* Free the block represented by the given DVA.
*/
void
metaslab_free_dva(spa_t *spa, const dva_t *dva, boolean_t checkpoint)
{
uint64_t vdev = DVA_GET_VDEV(dva);
uint64_t offset = DVA_GET_OFFSET(dva);
uint64_t size = DVA_GET_ASIZE(dva);
vdev_t *vd = vdev_lookup_top(spa, vdev);
ASSERT(DVA_IS_VALID(dva));
ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0);
if (DVA_GET_GANG(dva)) {
size = vdev_gang_header_asize(vd);
}
metaslab_free_impl(vd, offset, size, checkpoint);
}
/*
* Reserve some allocation slots. The reservation system must be called
* before we call into the allocator. If there aren't any available slots
* then the I/O will be throttled until an I/O completes and its slots are
* freed up. The function returns true if it was successful in placing
* the reservation.
*/
boolean_t
metaslab_class_throttle_reserve(metaslab_class_t *mc, int slots, int allocator,
zio_t *zio, int flags)
{
metaslab_class_allocator_t *mca = &mc->mc_allocator[allocator];
uint64_t available_slots = 0;
boolean_t slot_reserved = B_FALSE;
uint64_t max = mca->mca_alloc_max_slots;
ASSERT(mc->mc_alloc_throttle_enabled);
mutex_enter(&mc->mc_lock);
uint64_t reserved_slots = zfs_refcount_count(&mca->mca_alloc_slots);
if (reserved_slots < max)
available_slots = max - reserved_slots;
if (slots <= available_slots || GANG_ALLOCATION(flags) ||
flags & METASLAB_MUST_RESERVE) {
/*
* We reserve the slots individually so that we can unreserve
* them individually when an I/O completes.
*/
for (int d = 0; d < slots; d++)
zfs_refcount_add(&mca->mca_alloc_slots, zio);
zio->io_flags |= ZIO_FLAG_IO_ALLOCATING;
slot_reserved = B_TRUE;
}
mutex_exit(&mc->mc_lock);
return (slot_reserved);
}
void
metaslab_class_throttle_unreserve(metaslab_class_t *mc, int slots,
int allocator, zio_t *zio)
{
metaslab_class_allocator_t *mca = &mc->mc_allocator[allocator];
ASSERT(mc->mc_alloc_throttle_enabled);
mutex_enter(&mc->mc_lock);
for (int d = 0; d < slots; d++)
zfs_refcount_remove(&mca->mca_alloc_slots, zio);
mutex_exit(&mc->mc_lock);
}
static int
metaslab_claim_concrete(vdev_t *vd, uint64_t offset, uint64_t size,
uint64_t txg)
{
metaslab_t *msp;
spa_t *spa = vd->vdev_spa;
int error = 0;
if (offset >> vd->vdev_ms_shift >= vd->vdev_ms_count)
return (SET_ERROR(ENXIO));
ASSERT3P(vd->vdev_ms, !=, NULL);
msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
mutex_enter(&msp->ms_lock);
if ((txg != 0 && spa_writeable(spa)) || !msp->ms_loaded) {
error = metaslab_activate(msp, 0, METASLAB_WEIGHT_CLAIM);
if (error == EBUSY) {
ASSERT(msp->ms_loaded);
ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK);
error = 0;
}
}
if (error == 0 &&
!range_tree_contains(msp->ms_allocatable, offset, size))
error = SET_ERROR(ENOENT);
if (error || txg == 0) { /* txg == 0 indicates dry run */
mutex_exit(&msp->ms_lock);
return (error);
}
VERIFY(!msp->ms_condensing);
VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift));
VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
VERIFY3U(range_tree_space(msp->ms_allocatable) - size, <=,
msp->ms_size);
range_tree_remove(msp->ms_allocatable, offset, size);
range_tree_clear(msp->ms_trim, offset, size);
if (spa_writeable(spa)) { /* don't dirty if we're zdb(8) */
metaslab_class_t *mc = msp->ms_group->mg_class;
multilist_sublist_t *mls =
multilist_sublist_lock_obj(&mc->mc_metaslab_txg_list, msp);
if (!multilist_link_active(&msp->ms_class_txg_node)) {
msp->ms_selected_txg = txg;
multilist_sublist_insert_head(mls, msp);
}
multilist_sublist_unlock(mls);
if (range_tree_is_empty(msp->ms_allocating[txg & TXG_MASK]))
vdev_dirty(vd, VDD_METASLAB, msp, txg);
range_tree_add(msp->ms_allocating[txg & TXG_MASK],
offset, size);
msp->ms_allocating_total += size;
}
mutex_exit(&msp->ms_lock);
return (0);
}
typedef struct metaslab_claim_cb_arg_t {
uint64_t mcca_txg;
int mcca_error;
} metaslab_claim_cb_arg_t;
/* ARGSUSED */
static void
metaslab_claim_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset,
uint64_t size, void *arg)
{
metaslab_claim_cb_arg_t *mcca_arg = arg;
if (mcca_arg->mcca_error == 0) {
mcca_arg->mcca_error = metaslab_claim_concrete(vd, offset,
size, mcca_arg->mcca_txg);
}
}
int
metaslab_claim_impl(vdev_t *vd, uint64_t offset, uint64_t size, uint64_t txg)
{
if (vd->vdev_ops->vdev_op_remap != NULL) {
metaslab_claim_cb_arg_t arg;
/*
* Only zdb(8) can claim on indirect vdevs. This is used
* to detect leaks of mapped space (that are not accounted
* for in the obsolete counts, spacemap, or bpobj).
*/
ASSERT(!spa_writeable(vd->vdev_spa));
arg.mcca_error = 0;
arg.mcca_txg = txg;
vd->vdev_ops->vdev_op_remap(vd, offset, size,
metaslab_claim_impl_cb, &arg);
if (arg.mcca_error == 0) {
arg.mcca_error = metaslab_claim_concrete(vd,
offset, size, txg);
}
return (arg.mcca_error);
} else {
return (metaslab_claim_concrete(vd, offset, size, txg));
}
}
/*
* Intent log support: upon opening the pool after a crash, notify the SPA
* of blocks that the intent log has allocated for immediate write, but
* which are still considered free by the SPA because the last transaction
* group didn't commit yet.
*/
static int
metaslab_claim_dva(spa_t *spa, const dva_t *dva, uint64_t txg)
{
uint64_t vdev = DVA_GET_VDEV(dva);
uint64_t offset = DVA_GET_OFFSET(dva);
uint64_t size = DVA_GET_ASIZE(dva);
vdev_t *vd;
if ((vd = vdev_lookup_top(spa, vdev)) == NULL) {
return (SET_ERROR(ENXIO));
}
ASSERT(DVA_IS_VALID(dva));
if (DVA_GET_GANG(dva))
size = vdev_gang_header_asize(vd);
return (metaslab_claim_impl(vd, offset, size, txg));
}
int
metaslab_alloc(spa_t *spa, metaslab_class_t *mc, uint64_t psize, blkptr_t *bp,
int ndvas, uint64_t txg, blkptr_t *hintbp, int flags,
zio_alloc_list_t *zal, zio_t *zio, int allocator)
{
dva_t *dva = bp->blk_dva;
dva_t *hintdva = (hintbp != NULL) ? hintbp->blk_dva : NULL;
int error = 0;
ASSERT(bp->blk_birth == 0);
ASSERT(BP_PHYSICAL_BIRTH(bp) == 0);
spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER);
if (mc->mc_allocator[allocator].mca_rotor == NULL) {
/* no vdevs in this class */
spa_config_exit(spa, SCL_ALLOC, FTAG);
return (SET_ERROR(ENOSPC));
}
ASSERT(ndvas > 0 && ndvas <= spa_max_replication(spa));
ASSERT(BP_GET_NDVAS(bp) == 0);
ASSERT(hintbp == NULL || ndvas <= BP_GET_NDVAS(hintbp));
ASSERT3P(zal, !=, NULL);
for (int d = 0; d < ndvas; d++) {
error = metaslab_alloc_dva(spa, mc, psize, dva, d, hintdva,
txg, flags, zal, allocator);
if (error != 0) {
for (d--; d >= 0; d--) {
metaslab_unalloc_dva(spa, &dva[d], txg);
metaslab_group_alloc_decrement(spa,
DVA_GET_VDEV(&dva[d]), zio, flags,
allocator, B_FALSE);
bzero(&dva[d], sizeof (dva_t));
}
spa_config_exit(spa, SCL_ALLOC, FTAG);
return (error);
} else {
/*
* Update the metaslab group's queue depth
* based on the newly allocated dva.
*/
metaslab_group_alloc_increment(spa,
DVA_GET_VDEV(&dva[d]), zio, flags, allocator);
}
}
ASSERT(error == 0);
ASSERT(BP_GET_NDVAS(bp) == ndvas);
spa_config_exit(spa, SCL_ALLOC, FTAG);
BP_SET_BIRTH(bp, txg, 0);
return (0);
}
void
metaslab_free(spa_t *spa, const blkptr_t *bp, uint64_t txg, boolean_t now)
{
const dva_t *dva = bp->blk_dva;
int ndvas = BP_GET_NDVAS(bp);
ASSERT(!BP_IS_HOLE(bp));
ASSERT(!now || bp->blk_birth >= spa_syncing_txg(spa));
/*
* If we have a checkpoint for the pool we need to make sure that
* the blocks that we free that are part of the checkpoint won't be
* reused until the checkpoint is discarded or we revert to it.
*
* The checkpoint flag is passed down the metaslab_free code path
* and is set whenever we want to add a block to the checkpoint's
* accounting. That is, we "checkpoint" blocks that existed at the
* time the checkpoint was created and are therefore referenced by
* the checkpointed uberblock.
*
* Note that, we don't checkpoint any blocks if the current
* syncing txg <= spa_checkpoint_txg. We want these frees to sync
* normally as they will be referenced by the checkpointed uberblock.
*/
boolean_t checkpoint = B_FALSE;
if (bp->blk_birth <= spa->spa_checkpoint_txg &&
spa_syncing_txg(spa) > spa->spa_checkpoint_txg) {
/*
* At this point, if the block is part of the checkpoint
* there is no way it was created in the current txg.
*/
ASSERT(!now);
ASSERT3U(spa_syncing_txg(spa), ==, txg);
checkpoint = B_TRUE;
}
spa_config_enter(spa, SCL_FREE, FTAG, RW_READER);
for (int d = 0; d < ndvas; d++) {
if (now) {
metaslab_unalloc_dva(spa, &dva[d], txg);
} else {
ASSERT3U(txg, ==, spa_syncing_txg(spa));
metaslab_free_dva(spa, &dva[d], checkpoint);
}
}
spa_config_exit(spa, SCL_FREE, FTAG);
}
int
metaslab_claim(spa_t *spa, const blkptr_t *bp, uint64_t txg)
{
const dva_t *dva = bp->blk_dva;
int ndvas = BP_GET_NDVAS(bp);
int error = 0;
ASSERT(!BP_IS_HOLE(bp));
if (txg != 0) {
/*
* First do a dry run to make sure all DVAs are claimable,
* so we don't have to unwind from partial failures below.
*/
if ((error = metaslab_claim(spa, bp, 0)) != 0)
return (error);
}
spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER);
for (int d = 0; d < ndvas; d++) {
error = metaslab_claim_dva(spa, &dva[d], txg);
if (error != 0)
break;
}
spa_config_exit(spa, SCL_ALLOC, FTAG);
ASSERT(error == 0 || txg == 0);
return (error);
}
void
metaslab_fastwrite_mark(spa_t *spa, const blkptr_t *bp)
{
const dva_t *dva = bp->blk_dva;
int ndvas = BP_GET_NDVAS(bp);
uint64_t psize = BP_GET_PSIZE(bp);
int d;
vdev_t *vd;
ASSERT(!BP_IS_HOLE(bp));
ASSERT(!BP_IS_EMBEDDED(bp));
ASSERT(psize > 0);
spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
for (d = 0; d < ndvas; d++) {
if ((vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d]))) == NULL)
continue;
atomic_add_64(&vd->vdev_pending_fastwrite, psize);
}
spa_config_exit(spa, SCL_VDEV, FTAG);
}
void
metaslab_fastwrite_unmark(spa_t *spa, const blkptr_t *bp)
{
const dva_t *dva = bp->blk_dva;
int ndvas = BP_GET_NDVAS(bp);
uint64_t psize = BP_GET_PSIZE(bp);
int d;
vdev_t *vd;
ASSERT(!BP_IS_HOLE(bp));
ASSERT(!BP_IS_EMBEDDED(bp));
ASSERT(psize > 0);
spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
for (d = 0; d < ndvas; d++) {
if ((vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d]))) == NULL)
continue;
ASSERT3U(vd->vdev_pending_fastwrite, >=, psize);
atomic_sub_64(&vd->vdev_pending_fastwrite, psize);
}
spa_config_exit(spa, SCL_VDEV, FTAG);
}
/* ARGSUSED */
static void
metaslab_check_free_impl_cb(uint64_t inner, vdev_t *vd, uint64_t offset,
uint64_t size, void *arg)
{
if (vd->vdev_ops == &vdev_indirect_ops)
return;
metaslab_check_free_impl(vd, offset, size);
}
static void
metaslab_check_free_impl(vdev_t *vd, uint64_t offset, uint64_t size)
{
metaslab_t *msp;
spa_t *spa __maybe_unused = vd->vdev_spa;
if ((zfs_flags & ZFS_DEBUG_ZIO_FREE) == 0)
return;
if (vd->vdev_ops->vdev_op_remap != NULL) {
vd->vdev_ops->vdev_op_remap(vd, offset, size,
metaslab_check_free_impl_cb, NULL);
return;
}
ASSERT(vdev_is_concrete(vd));
ASSERT3U(offset >> vd->vdev_ms_shift, <, vd->vdev_ms_count);
ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0);
msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
mutex_enter(&msp->ms_lock);
if (msp->ms_loaded) {
range_tree_verify_not_present(msp->ms_allocatable,
offset, size);
}
/*
* Check all segments that currently exist in the freeing pipeline.
*
* It would intuitively make sense to also check the current allocating
* tree since metaslab_unalloc_dva() exists for extents that are
* allocated and freed in the same sync pass within the same txg.
* Unfortunately there are places (e.g. the ZIL) where we allocate a
* segment but then we free part of it within the same txg
* [see zil_sync()]. Thus, we don't call range_tree_verify() in the
* current allocating tree.
*/
range_tree_verify_not_present(msp->ms_freeing, offset, size);
range_tree_verify_not_present(msp->ms_checkpointing, offset, size);
range_tree_verify_not_present(msp->ms_freed, offset, size);
for (int j = 0; j < TXG_DEFER_SIZE; j++)
range_tree_verify_not_present(msp->ms_defer[j], offset, size);
range_tree_verify_not_present(msp->ms_trim, offset, size);
mutex_exit(&msp->ms_lock);
}
void
metaslab_check_free(spa_t *spa, const blkptr_t *bp)
{
if ((zfs_flags & ZFS_DEBUG_ZIO_FREE) == 0)
return;
spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
for (int i = 0; i < BP_GET_NDVAS(bp); i++) {
uint64_t vdev = DVA_GET_VDEV(&bp->blk_dva[i]);
vdev_t *vd = vdev_lookup_top(spa, vdev);
uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]);
uint64_t size = DVA_GET_ASIZE(&bp->blk_dva[i]);
if (DVA_GET_GANG(&bp->blk_dva[i]))
size = vdev_gang_header_asize(vd);
ASSERT3P(vd, !=, NULL);
metaslab_check_free_impl(vd, offset, size);
}
spa_config_exit(spa, SCL_VDEV, FTAG);
}
static void
metaslab_group_disable_wait(metaslab_group_t *mg)
{
ASSERT(MUTEX_HELD(&mg->mg_ms_disabled_lock));
while (mg->mg_disabled_updating) {
cv_wait(&mg->mg_ms_disabled_cv, &mg->mg_ms_disabled_lock);
}
}
static void
metaslab_group_disabled_increment(metaslab_group_t *mg)
{
ASSERT(MUTEX_HELD(&mg->mg_ms_disabled_lock));
ASSERT(mg->mg_disabled_updating);
while (mg->mg_ms_disabled >= max_disabled_ms) {
cv_wait(&mg->mg_ms_disabled_cv, &mg->mg_ms_disabled_lock);
}
mg->mg_ms_disabled++;
ASSERT3U(mg->mg_ms_disabled, <=, max_disabled_ms);
}
/*
* Mark the metaslab as disabled to prevent any allocations on this metaslab.
* We must also track how many metaslabs are currently disabled within a
* metaslab group and limit them to prevent allocation failures from
* occurring because all metaslabs are disabled.
*/
void
metaslab_disable(metaslab_t *msp)
{
ASSERT(!MUTEX_HELD(&msp->ms_lock));
metaslab_group_t *mg = msp->ms_group;
mutex_enter(&mg->mg_ms_disabled_lock);
/*
* To keep an accurate count of how many threads have disabled
* a specific metaslab group, we only allow one thread to mark
* the metaslab group at a time. This ensures that the value of
* ms_disabled will be accurate when we decide to mark a metaslab
* group as disabled. To do this we force all other threads
* to wait till the metaslab's mg_disabled_updating flag is no
* longer set.
*/
metaslab_group_disable_wait(mg);
mg->mg_disabled_updating = B_TRUE;
if (msp->ms_disabled == 0) {
metaslab_group_disabled_increment(mg);
}
mutex_enter(&msp->ms_lock);
msp->ms_disabled++;
mutex_exit(&msp->ms_lock);
mg->mg_disabled_updating = B_FALSE;
cv_broadcast(&mg->mg_ms_disabled_cv);
mutex_exit(&mg->mg_ms_disabled_lock);
}
void
metaslab_enable(metaslab_t *msp, boolean_t sync, boolean_t unload)
{
metaslab_group_t *mg = msp->ms_group;
spa_t *spa = mg->mg_vd->vdev_spa;
/*
* Wait for the outstanding IO to be synced to prevent newly
* allocated blocks from being overwritten. This used by
* initialize and TRIM which are modifying unallocated space.
*/
if (sync)
txg_wait_synced(spa_get_dsl(spa), 0);
mutex_enter(&mg->mg_ms_disabled_lock);
mutex_enter(&msp->ms_lock);
if (--msp->ms_disabled == 0) {
mg->mg_ms_disabled--;
cv_broadcast(&mg->mg_ms_disabled_cv);
if (unload)
metaslab_unload(msp);
}
mutex_exit(&msp->ms_lock);
mutex_exit(&mg->mg_ms_disabled_lock);
}
static void
metaslab_update_ondisk_flush_data(metaslab_t *ms, dmu_tx_t *tx)
{
vdev_t *vd = ms->ms_group->mg_vd;
spa_t *spa = vd->vdev_spa;
objset_t *mos = spa_meta_objset(spa);
ASSERT(spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP));
metaslab_unflushed_phys_t entry = {
.msp_unflushed_txg = metaslab_unflushed_txg(ms),
};
uint64_t entry_size = sizeof (entry);
uint64_t entry_offset = ms->ms_id * entry_size;
uint64_t object = 0;
int err = zap_lookup(mos, vd->vdev_top_zap,
VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1,
&object);
if (err == ENOENT) {
object = dmu_object_alloc(mos, DMU_OTN_UINT64_METADATA,
SPA_OLD_MAXBLOCKSIZE, DMU_OT_NONE, 0, tx);
VERIFY0(zap_add(mos, vd->vdev_top_zap,
VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1,
&object, tx));
} else {
VERIFY0(err);
}
dmu_write(spa_meta_objset(spa), object, entry_offset, entry_size,
&entry, tx);
}
void
metaslab_set_unflushed_txg(metaslab_t *ms, uint64_t txg, dmu_tx_t *tx)
{
spa_t *spa = ms->ms_group->mg_vd->vdev_spa;
if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
return;
ms->ms_unflushed_txg = txg;
metaslab_update_ondisk_flush_data(ms, tx);
}
uint64_t
metaslab_unflushed_txg(metaslab_t *ms)
{
return (ms->ms_unflushed_txg);
}
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, aliquot, ULONG, ZMOD_RW,
"Allocation granularity (a.k.a. stripe size)");
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, debug_load, INT, ZMOD_RW,
"Load all metaslabs when pool is first opened");
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, debug_unload, INT, ZMOD_RW,
"Prevent metaslabs from being unloaded");
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, preload_enabled, INT, ZMOD_RW,
"Preload potential metaslabs during reassessment");
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, unload_delay, INT, ZMOD_RW,
"Delay in txgs after metaslab was last used before unloading");
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, unload_delay_ms, INT, ZMOD_RW,
"Delay in milliseconds after metaslab was last used before unloading");
/* BEGIN CSTYLED */
ZFS_MODULE_PARAM(zfs_mg, zfs_mg_, noalloc_threshold, INT, ZMOD_RW,
"Percentage of metaslab group size that should be free to make it "
"eligible for allocation");
ZFS_MODULE_PARAM(zfs_mg, zfs_mg_, fragmentation_threshold, INT, ZMOD_RW,
"Percentage of metaslab group size that should be considered eligible "
"for allocations unless all metaslab groups within the metaslab class "
"have also crossed this threshold");
ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, fragmentation_threshold, INT,
ZMOD_RW, "Fragmentation for metaslab to allow allocation");
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, fragmentation_factor_enabled, INT, ZMOD_RW,
"Use the fragmentation metric to prefer less fragmented metaslabs");
/* END CSTYLED */
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, lba_weighting_enabled, INT, ZMOD_RW,
"Prefer metaslabs with lower LBAs");
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, bias_enabled, INT, ZMOD_RW,
"Enable metaslab group biasing");
ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, segment_weight_enabled, INT,
ZMOD_RW, "Enable segment-based metaslab selection");
ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, switch_threshold, INT, ZMOD_RW,
"Segment-based metaslab selection maximum buckets before switching");
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, force_ganging, ULONG, ZMOD_RW,
"Blocks larger than this size are forced to be gang blocks");
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, df_max_search, INT, ZMOD_RW,
"Max distance (bytes) to search forward before using size tree");
ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, df_use_largest_segment, INT, ZMOD_RW,
"When looking in size tree, use largest segment instead of exact fit");
ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, max_size_cache_sec, ULONG,
ZMOD_RW, "How long to trust the cached max chunk size of a metaslab");
ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, mem_limit, INT, ZMOD_RW,
"Percentage of memory that can be used to store metaslab range trees");
ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, try_hard_before_gang, INT,
ZMOD_RW, "Try hard to allocate before ganging");
ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, find_max_tries, INT, ZMOD_RW,
"Normally only consider this many of the best metaslabs in each vdev");
diff --git a/sys/contrib/openzfs/module/zstd/zfs_zstd.c b/sys/contrib/openzfs/module/zstd/zfs_zstd.c
index 78616c08ba72..3f3983d8d868 100644
--- a/sys/contrib/openzfs/module/zstd/zfs_zstd.c
+++ b/sys/contrib/openzfs/module/zstd/zfs_zstd.c
@@ -1,780 +1,799 @@
/*
* BSD 3-Clause New License (https://spdx.org/licenses/BSD-3-Clause.html)
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its
* contributors may be used to endorse or promote products derived from this
* software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
/*
* Copyright (c) 2016-2018, Klara Inc.
* Copyright (c) 2016-2018, Allan Jude
* Copyright (c) 2018-2020, Sebastian Gottschall
* Copyright (c) 2019-2020, Michael Niewöhner
* Copyright (c) 2020, The FreeBSD Foundation [1]
*
* [1] Portions of this software were developed by Allan Jude
* under sponsorship from the FreeBSD Foundation.
*/
#include
#include
#include
#include
#include
#include
#define ZSTD_STATIC_LINKING_ONLY
#include "lib/zstd.h"
#include "lib/zstd_errors.h"
kstat_t *zstd_ksp = NULL;
typedef struct zstd_stats {
kstat_named_t zstd_stat_alloc_fail;
kstat_named_t zstd_stat_alloc_fallback;
kstat_named_t zstd_stat_com_alloc_fail;
kstat_named_t zstd_stat_dec_alloc_fail;
kstat_named_t zstd_stat_com_inval;
kstat_named_t zstd_stat_dec_inval;
kstat_named_t zstd_stat_dec_header_inval;
kstat_named_t zstd_stat_com_fail;
kstat_named_t zstd_stat_dec_fail;
kstat_named_t zstd_stat_buffers;
kstat_named_t zstd_stat_size;
} zstd_stats_t;
static zstd_stats_t zstd_stats = {
{ "alloc_fail", KSTAT_DATA_UINT64 },
{ "alloc_fallback", KSTAT_DATA_UINT64 },
{ "compress_alloc_fail", KSTAT_DATA_UINT64 },
{ "decompress_alloc_fail", KSTAT_DATA_UINT64 },
{ "compress_level_invalid", KSTAT_DATA_UINT64 },
{ "decompress_level_invalid", KSTAT_DATA_UINT64 },
{ "decompress_header_invalid", KSTAT_DATA_UINT64 },
{ "compress_failed", KSTAT_DATA_UINT64 },
{ "decompress_failed", KSTAT_DATA_UINT64 },
{ "buffers", KSTAT_DATA_UINT64 },
{ "size", KSTAT_DATA_UINT64 },
};
/* Enums describing the allocator type specified by kmem_type in zstd_kmem */
enum zstd_kmem_type {
ZSTD_KMEM_UNKNOWN = 0,
/* Allocation type using kmem_vmalloc */
ZSTD_KMEM_DEFAULT,
/* Pool based allocation using mempool_alloc */
ZSTD_KMEM_POOL,
/* Reserved fallback memory for decompression only */
ZSTD_KMEM_DCTX,
ZSTD_KMEM_COUNT,
};
/* Structure for pooled memory objects */
struct zstd_pool {
void *mem;
size_t size;
kmutex_t barrier;
hrtime_t timeout;
};
/* Global structure for handling memory allocations */
struct zstd_kmem {
enum zstd_kmem_type kmem_type;
size_t kmem_size;
struct zstd_pool *pool;
};
/* Fallback memory structure used for decompression only if memory runs out */
struct zstd_fallback_mem {
size_t mem_size;
void *mem;
kmutex_t barrier;
};
struct zstd_levelmap {
int16_t zstd_level;
enum zio_zstd_levels level;
};
/*
* ZSTD memory handlers
*
* For decompression we use a different handler which also provides fallback
* memory allocation in case memory runs out.
*
* The ZSTD handlers were split up for the most simplified implementation.
*/
static void *zstd_alloc(void *opaque, size_t size);
static void *zstd_dctx_alloc(void *opaque, size_t size);
static void zstd_free(void *opaque, void *ptr);
/* Compression memory handler */
static const ZSTD_customMem zstd_malloc = {
zstd_alloc,
zstd_free,
NULL,
};
/* Decompression memory handler */
static const ZSTD_customMem zstd_dctx_malloc = {
zstd_dctx_alloc,
zstd_free,
NULL,
};
/* Level map for converting ZFS internal levels to ZSTD levels and vice versa */
static struct zstd_levelmap zstd_levels[] = {
{ZIO_ZSTD_LEVEL_1, ZIO_ZSTD_LEVEL_1},
{ZIO_ZSTD_LEVEL_2, ZIO_ZSTD_LEVEL_2},
{ZIO_ZSTD_LEVEL_3, ZIO_ZSTD_LEVEL_3},
{ZIO_ZSTD_LEVEL_4, ZIO_ZSTD_LEVEL_4},
{ZIO_ZSTD_LEVEL_5, ZIO_ZSTD_LEVEL_5},
{ZIO_ZSTD_LEVEL_6, ZIO_ZSTD_LEVEL_6},
{ZIO_ZSTD_LEVEL_7, ZIO_ZSTD_LEVEL_7},
{ZIO_ZSTD_LEVEL_8, ZIO_ZSTD_LEVEL_8},
{ZIO_ZSTD_LEVEL_9, ZIO_ZSTD_LEVEL_9},
{ZIO_ZSTD_LEVEL_10, ZIO_ZSTD_LEVEL_10},
{ZIO_ZSTD_LEVEL_11, ZIO_ZSTD_LEVEL_11},
{ZIO_ZSTD_LEVEL_12, ZIO_ZSTD_LEVEL_12},
{ZIO_ZSTD_LEVEL_13, ZIO_ZSTD_LEVEL_13},
{ZIO_ZSTD_LEVEL_14, ZIO_ZSTD_LEVEL_14},
{ZIO_ZSTD_LEVEL_15, ZIO_ZSTD_LEVEL_15},
{ZIO_ZSTD_LEVEL_16, ZIO_ZSTD_LEVEL_16},
{ZIO_ZSTD_LEVEL_17, ZIO_ZSTD_LEVEL_17},
{ZIO_ZSTD_LEVEL_18, ZIO_ZSTD_LEVEL_18},
{ZIO_ZSTD_LEVEL_19, ZIO_ZSTD_LEVEL_19},
{-1, ZIO_ZSTD_LEVEL_FAST_1},
{-2, ZIO_ZSTD_LEVEL_FAST_2},
{-3, ZIO_ZSTD_LEVEL_FAST_3},
{-4, ZIO_ZSTD_LEVEL_FAST_4},
{-5, ZIO_ZSTD_LEVEL_FAST_5},
{-6, ZIO_ZSTD_LEVEL_FAST_6},
{-7, ZIO_ZSTD_LEVEL_FAST_7},
{-8, ZIO_ZSTD_LEVEL_FAST_8},
{-9, ZIO_ZSTD_LEVEL_FAST_9},
{-10, ZIO_ZSTD_LEVEL_FAST_10},
{-20, ZIO_ZSTD_LEVEL_FAST_20},
{-30, ZIO_ZSTD_LEVEL_FAST_30},
{-40, ZIO_ZSTD_LEVEL_FAST_40},
{-50, ZIO_ZSTD_LEVEL_FAST_50},
{-60, ZIO_ZSTD_LEVEL_FAST_60},
{-70, ZIO_ZSTD_LEVEL_FAST_70},
{-80, ZIO_ZSTD_LEVEL_FAST_80},
{-90, ZIO_ZSTD_LEVEL_FAST_90},
{-100, ZIO_ZSTD_LEVEL_FAST_100},
{-500, ZIO_ZSTD_LEVEL_FAST_500},
{-1000, ZIO_ZSTD_LEVEL_FAST_1000},
};
/*
* This variable represents the maximum count of the pool based on the number
* of CPUs plus some buffer. We default to cpu count * 4, see init_zstd.
*/
static int pool_count = 16;
#define ZSTD_POOL_MAX pool_count
#define ZSTD_POOL_TIMEOUT 60 * 2
static struct zstd_fallback_mem zstd_dctx_fallback;
static struct zstd_pool *zstd_mempool_cctx;
static struct zstd_pool *zstd_mempool_dctx;
+/*
+ * The library zstd code expects these if ADDRESS_SANITIZER gets defined,
+ * and while ASAN does this, KASAN defines that and does not. So to avoid
+ * changing the external code, we do this.
+ */
+#if defined(__has_feature)
+#if __has_feature(address_sanitizer)
+#define ADDRESS_SANITIZER 1
+#endif
+#elif defined(__SANITIZE_ADDRESS__)
+#define ADDRESS_SANITIZER 1
+#endif
+#if defined(_KERNEL) && defined(ADDRESS_SANITIZER)
+void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
+void __asan_poison_memory_region(void const volatile *addr, size_t size);
+void __asan_unpoison_memory_region(void const volatile *addr, size_t size) {};
+void __asan_poison_memory_region(void const volatile *addr, size_t size) {};
+#endif
+
static void
zstd_mempool_reap(struct zstd_pool *zstd_mempool)
{
struct zstd_pool *pool;
if (!zstd_mempool || !ZSTDSTAT(zstd_stat_buffers)) {
return;
}
/* free obsolete slots */
for (int i = 0; i < ZSTD_POOL_MAX; i++) {
pool = &zstd_mempool[i];
if (pool->mem && mutex_tryenter(&pool->barrier)) {
/* Free memory if unused object older than 2 minutes */
if (pool->mem && gethrestime_sec() > pool->timeout) {
vmem_free(pool->mem, pool->size);
ZSTDSTAT_SUB(zstd_stat_buffers, 1);
ZSTDSTAT_SUB(zstd_stat_size, pool->size);
pool->mem = NULL;
pool->size = 0;
pool->timeout = 0;
}
mutex_exit(&pool->barrier);
}
}
}
/*
* Try to get a cached allocated buffer from memory pool or allocate a new one
* if necessary. If a object is older than 2 minutes and does not fit the
* requested size, it will be released and a new cached entry will be allocated.
* If other pooled objects are detected without being used for 2 minutes, they
* will be released, too.
*
* The concept is that high frequency memory allocations of bigger objects are
* expensive. So if a lot of work is going on, allocations will be kept for a
* while and can be reused in that time frame.
*
* The scheduled release will be updated every time a object is reused.
*/
static void *
zstd_mempool_alloc(struct zstd_pool *zstd_mempool, size_t size)
{
struct zstd_pool *pool;
struct zstd_kmem *mem = NULL;
if (!zstd_mempool) {
return (NULL);
}
/* Seek for preallocated memory slot and free obsolete slots */
for (int i = 0; i < ZSTD_POOL_MAX; i++) {
pool = &zstd_mempool[i];
/*
* This lock is simply a marker for a pool object being in use.
* If it's already hold, it will be skipped.
*
* We need to create it before checking it to avoid race
* conditions caused by running in a threaded context.
*
* The lock is later released by zstd_mempool_free.
*/
if (mutex_tryenter(&pool->barrier)) {
/*
* Check if objects fits the size, if so we take it and
* update the timestamp.
*/
if (pool->mem && size <= pool->size) {
pool->timeout = gethrestime_sec() +
ZSTD_POOL_TIMEOUT;
mem = pool->mem;
return (mem);
}
mutex_exit(&pool->barrier);
}
}
/*
* If no preallocated slot was found, try to fill in a new one.
*
* We run a similar algorithm twice here to avoid pool fragmentation.
* The first one may generate holes in the list if objects get released.
* We always make sure that these holes get filled instead of adding new
* allocations constantly at the end.
*/
for (int i = 0; i < ZSTD_POOL_MAX; i++) {
pool = &zstd_mempool[i];
if (mutex_tryenter(&pool->barrier)) {
/* Object is free, try to allocate new one */
if (!pool->mem) {
mem = vmem_alloc(size, KM_SLEEP);
if (mem) {
ZSTDSTAT_ADD(zstd_stat_buffers, 1);
ZSTDSTAT_ADD(zstd_stat_size, size);
pool->mem = mem;
pool->size = size;
/* Keep track for later release */
mem->pool = pool;
mem->kmem_type = ZSTD_KMEM_POOL;
mem->kmem_size = size;
}
}
if (size <= pool->size) {
/* Update timestamp */
pool->timeout = gethrestime_sec() +
ZSTD_POOL_TIMEOUT;
return (pool->mem);
}
mutex_exit(&pool->barrier);
}
}
/*
* If the pool is full or the allocation failed, try lazy allocation
* instead.
*/
if (!mem) {
mem = vmem_alloc(size, KM_NOSLEEP);
if (mem) {
mem->pool = NULL;
mem->kmem_type = ZSTD_KMEM_DEFAULT;
mem->kmem_size = size;
}
}
return (mem);
}
/* Mark object as released by releasing the barrier mutex */
static void
zstd_mempool_free(struct zstd_kmem *z)
{
mutex_exit(&z->pool->barrier);
}
/* Convert ZFS internal enum to ZSTD level */
static int
zstd_enum_to_level(enum zio_zstd_levels level, int16_t *zstd_level)
{
if (level > 0 && level <= ZIO_ZSTD_LEVEL_19) {
*zstd_level = zstd_levels[level - 1].zstd_level;
return (0);
}
if (level >= ZIO_ZSTD_LEVEL_FAST_1 &&
level <= ZIO_ZSTD_LEVEL_FAST_1000) {
*zstd_level = zstd_levels[level - ZIO_ZSTD_LEVEL_FAST_1
+ ZIO_ZSTD_LEVEL_19].zstd_level;
return (0);
}
/* Invalid/unknown zfs compression enum - this should never happen. */
return (1);
}
/* Compress block using zstd */
size_t
zfs_zstd_compress(void *s_start, void *d_start, size_t s_len, size_t d_len,
int level)
{
size_t c_len;
int16_t zstd_level;
zfs_zstdhdr_t *hdr;
ZSTD_CCtx *cctx;
hdr = (zfs_zstdhdr_t *)d_start;
/* Skip compression if the specified level is invalid */
if (zstd_enum_to_level(level, &zstd_level)) {
ZSTDSTAT_BUMP(zstd_stat_com_inval);
return (s_len);
}
ASSERT3U(d_len, >=, sizeof (*hdr));
ASSERT3U(d_len, <=, s_len);
ASSERT3U(zstd_level, !=, 0);
cctx = ZSTD_createCCtx_advanced(zstd_malloc);
/*
* Out of kernel memory, gently fall through - this will disable
* compression in zio_compress_data
*/
if (!cctx) {
ZSTDSTAT_BUMP(zstd_stat_com_alloc_fail);
return (s_len);
}
/* Set the compression level */
ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, zstd_level);
/* Use the "magicless" zstd header which saves us 4 header bytes */
ZSTD_CCtx_setParameter(cctx, ZSTD_c_format, ZSTD_f_zstd1_magicless);
/*
* Disable redundant checksum calculation and content size storage since
* this is already done by ZFS itself.
*/
ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 0);
ZSTD_CCtx_setParameter(cctx, ZSTD_c_contentSizeFlag, 0);
c_len = ZSTD_compress2(cctx,
hdr->data,
d_len - sizeof (*hdr),
s_start, s_len);
ZSTD_freeCCtx(cctx);
/* Error in the compression routine, disable compression. */
if (ZSTD_isError(c_len)) {
/*
* If we are aborting the compression because the saves are
* too small, that is not a failure. Everything else is a
* failure, so increment the compression failure counter.
*/
if (ZSTD_getErrorCode(c_len) != ZSTD_error_dstSize_tooSmall) {
ZSTDSTAT_BUMP(zstd_stat_com_fail);
}
return (s_len);
}
/*
* Encode the compressed buffer size at the start. We'll need this in
* decompression to counter the effects of padding which might be added
* to the compressed buffer and which, if unhandled, would confuse the
* hell out of our decompression function.
*/
hdr->c_len = BE_32(c_len);
/*
* Check version for overflow.
* The limit of 24 bits must not be exceeded. This allows a maximum
* version 1677.72.15 which we don't expect to be ever reached.
*/
ASSERT3U(ZSTD_VERSION_NUMBER, <=, 0xFFFFFF);
/*
* Encode the compression level as well. We may need to know the
* original compression level if compressed_arc is disabled, to match
* the compression settings to write this block to the L2ARC.
*
* Encode the actual level, so if the enum changes in the future, we
* will be compatible.
*
* The upper 24 bits store the ZSTD version to be able to provide
* future compatibility, since new versions might enhance the
* compression algorithm in a way, where the compressed data will
* change.
*
* As soon as such incompatibility occurs, handling code needs to be
* added, differentiating between the versions.
*/
hdr->version = ZSTD_VERSION_NUMBER;
hdr->level = level;
hdr->raw_version_level = BE_32(hdr->raw_version_level);
return (c_len + sizeof (*hdr));
}
/* Decompress block using zstd and return its stored level */
int
zfs_zstd_decompress_level(void *s_start, void *d_start, size_t s_len,
size_t d_len, uint8_t *level)
{
ZSTD_DCtx *dctx;
size_t result;
int16_t zstd_level;
uint32_t c_len;
const zfs_zstdhdr_t *hdr;
zfs_zstdhdr_t hdr_copy;
hdr = (const zfs_zstdhdr_t *)s_start;
c_len = BE_32(hdr->c_len);
/*
* Make a copy instead of directly converting the header, since we must
* not modify the original data that may be used again later.
*/
hdr_copy.raw_version_level = BE_32(hdr->raw_version_level);
/*
* NOTE: We ignore the ZSTD version for now. As soon as any
* incompatibility occurs, it has to be handled accordingly.
* The version can be accessed via `hdr_copy.version`.
*/
/*
* Convert and check the level
* An invalid level is a strong indicator for data corruption! In such
* case return an error so the upper layers can try to fix it.
*/
if (zstd_enum_to_level(hdr_copy.level, &zstd_level)) {
ZSTDSTAT_BUMP(zstd_stat_dec_inval);
return (1);
}
ASSERT3U(d_len, >=, s_len);
ASSERT3U(hdr_copy.level, !=, ZIO_COMPLEVEL_INHERIT);
/* Invalid compressed buffer size encoded at start */
if (c_len + sizeof (*hdr) > s_len) {
ZSTDSTAT_BUMP(zstd_stat_dec_header_inval);
return (1);
}
dctx = ZSTD_createDCtx_advanced(zstd_dctx_malloc);
if (!dctx) {
ZSTDSTAT_BUMP(zstd_stat_dec_alloc_fail);
return (1);
}
/* Set header type to "magicless" */
ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, ZSTD_f_zstd1_magicless);
/* Decompress the data and release the context */
result = ZSTD_decompressDCtx(dctx, d_start, d_len, hdr->data, c_len);
ZSTD_freeDCtx(dctx);
/*
* Returns 0 on success (decompression function returned non-negative)
* and non-zero on failure (decompression function returned negative.
*/
if (ZSTD_isError(result)) {
ZSTDSTAT_BUMP(zstd_stat_dec_fail);
return (1);
}
if (level) {
*level = hdr_copy.level;
}
return (0);
}
/* Decompress datablock using zstd */
int
zfs_zstd_decompress(void *s_start, void *d_start, size_t s_len, size_t d_len,
int level __maybe_unused)
{
return (zfs_zstd_decompress_level(s_start, d_start, s_len, d_len,
NULL));
}
/* Allocator for zstd compression context using mempool_allocator */
static void *
zstd_alloc(void *opaque __maybe_unused, size_t size)
{
size_t nbytes = sizeof (struct zstd_kmem) + size;
struct zstd_kmem *z = NULL;
z = (struct zstd_kmem *)zstd_mempool_alloc(zstd_mempool_cctx, nbytes);
if (!z) {
ZSTDSTAT_BUMP(zstd_stat_alloc_fail);
return (NULL);
}
return ((void*)z + (sizeof (struct zstd_kmem)));
}
/*
* Allocator for zstd decompression context using mempool_allocator with
* fallback to reserved memory if allocation fails
*/
static void *
zstd_dctx_alloc(void *opaque __maybe_unused, size_t size)
{
size_t nbytes = sizeof (struct zstd_kmem) + size;
struct zstd_kmem *z = NULL;
enum zstd_kmem_type type = ZSTD_KMEM_DEFAULT;
z = (struct zstd_kmem *)zstd_mempool_alloc(zstd_mempool_dctx, nbytes);
if (!z) {
/* Try harder, decompression shall not fail */
z = vmem_alloc(nbytes, KM_SLEEP);
if (z) {
z->pool = NULL;
}
ZSTDSTAT_BUMP(zstd_stat_alloc_fail);
} else {
return ((void*)z + (sizeof (struct zstd_kmem)));
}
/* Fallback if everything fails */
if (!z) {
/*
* Barrier since we only can handle it in a single thread. All
* other following threads need to wait here until decompression
* is completed. zstd_free will release this barrier later.
*/
mutex_enter(&zstd_dctx_fallback.barrier);
z = zstd_dctx_fallback.mem;
type = ZSTD_KMEM_DCTX;
ZSTDSTAT_BUMP(zstd_stat_alloc_fallback);
}
/* Allocation should always be successful */
if (!z) {
return (NULL);
}
z->kmem_type = type;
z->kmem_size = nbytes;
return ((void*)z + (sizeof (struct zstd_kmem)));
}
/* Free allocated memory by its specific type */
static void
zstd_free(void *opaque __maybe_unused, void *ptr)
{
struct zstd_kmem *z = (ptr - sizeof (struct zstd_kmem));
enum zstd_kmem_type type;
ASSERT3U(z->kmem_type, <, ZSTD_KMEM_COUNT);
ASSERT3U(z->kmem_type, >, ZSTD_KMEM_UNKNOWN);
type = z->kmem_type;
switch (type) {
case ZSTD_KMEM_DEFAULT:
vmem_free(z, z->kmem_size);
break;
case ZSTD_KMEM_POOL:
zstd_mempool_free(z);
break;
case ZSTD_KMEM_DCTX:
mutex_exit(&zstd_dctx_fallback.barrier);
break;
default:
break;
}
}
/* Allocate fallback memory to ensure safe decompression */
static void __init
create_fallback_mem(struct zstd_fallback_mem *mem, size_t size)
{
mem->mem_size = size;
mem->mem = vmem_zalloc(mem->mem_size, KM_SLEEP);
mutex_init(&mem->barrier, NULL, MUTEX_DEFAULT, NULL);
}
/* Initialize memory pool barrier mutexes */
static void __init
zstd_mempool_init(void)
{
zstd_mempool_cctx = (struct zstd_pool *)
kmem_zalloc(ZSTD_POOL_MAX * sizeof (struct zstd_pool), KM_SLEEP);
zstd_mempool_dctx = (struct zstd_pool *)
kmem_zalloc(ZSTD_POOL_MAX * sizeof (struct zstd_pool), KM_SLEEP);
for (int i = 0; i < ZSTD_POOL_MAX; i++) {
mutex_init(&zstd_mempool_cctx[i].barrier, NULL,
MUTEX_DEFAULT, NULL);
mutex_init(&zstd_mempool_dctx[i].barrier, NULL,
MUTEX_DEFAULT, NULL);
}
}
/* Initialize zstd-related memory handling */
static int __init
zstd_meminit(void)
{
zstd_mempool_init();
/*
* Estimate the size of the fallback decompression context.
* The expected size on x64 with current ZSTD should be about 160 KB.
*/
create_fallback_mem(&zstd_dctx_fallback,
P2ROUNDUP(ZSTD_estimateDCtxSize() + sizeof (struct zstd_kmem),
PAGESIZE));
return (0);
}
/* Release object from pool and free memory */
static void __exit
release_pool(struct zstd_pool *pool)
{
mutex_destroy(&pool->barrier);
vmem_free(pool->mem, pool->size);
pool->mem = NULL;
pool->size = 0;
}
/* Release memory pool objects */
static void __exit
zstd_mempool_deinit(void)
{
for (int i = 0; i < ZSTD_POOL_MAX; i++) {
release_pool(&zstd_mempool_cctx[i]);
release_pool(&zstd_mempool_dctx[i]);
}
kmem_free(zstd_mempool_dctx, ZSTD_POOL_MAX * sizeof (struct zstd_pool));
kmem_free(zstd_mempool_cctx, ZSTD_POOL_MAX * sizeof (struct zstd_pool));
zstd_mempool_dctx = NULL;
zstd_mempool_cctx = NULL;
}
/* release unused memory from pool */
void
zfs_zstd_cache_reap_now(void)
{
/*
* Short-circuit if there are no buffers to begin with.
*/
if (ZSTDSTAT(zstd_stat_buffers) == 0)
return;
/*
* calling alloc with zero size seeks
* and releases old unused objects
*/
zstd_mempool_reap(zstd_mempool_cctx);
zstd_mempool_reap(zstd_mempool_dctx);
}
extern int __init
zstd_init(void)
{
/* Set pool size by using maximum sane thread count * 4 */
pool_count = (boot_ncpus * 4);
zstd_meminit();
/* Initialize kstat */
zstd_ksp = kstat_create("zfs", 0, "zstd", "misc",
KSTAT_TYPE_NAMED, sizeof (zstd_stats) / sizeof (kstat_named_t),
KSTAT_FLAG_VIRTUAL);
if (zstd_ksp != NULL) {
zstd_ksp->ks_data = &zstd_stats;
kstat_install(zstd_ksp);
}
return (0);
}
extern void __exit
zstd_fini(void)
{
/* Deinitialize kstat */
if (zstd_ksp != NULL) {
kstat_delete(zstd_ksp);
zstd_ksp = NULL;
}
/* Release fallback memory */
vmem_free(zstd_dctx_fallback.mem, zstd_dctx_fallback.mem_size);
mutex_destroy(&zstd_dctx_fallback.barrier);
/* Deinit memory pool */
zstd_mempool_deinit();
}
#if defined(_KERNEL)
module_init(zstd_init);
module_exit(zstd_fini);
ZFS_MODULE_DESCRIPTION("ZSTD Compression for ZFS");
ZFS_MODULE_LICENSE("Dual BSD/GPL");
ZFS_MODULE_VERSION(ZSTD_VERSION_STRING);
EXPORT_SYMBOL(zfs_zstd_compress);
EXPORT_SYMBOL(zfs_zstd_decompress_level);
EXPORT_SYMBOL(zfs_zstd_decompress);
EXPORT_SYMBOL(zfs_zstd_cache_reap_now);
#endif