From: wcheng@sourceware.org <wcheng@sourceware.org>
To: cluster-devel.redhat.com
Subject: [Cluster-devel] cluster/gfs-kernel/src/gfs daemon.c gfs_ondisk ...
Date: 13 Mar 2007 21:21:22 -0000 [thread overview]
Message-ID: <20070313212122.4022.qmail@sourceware.org> (raw)
CVSROOT: /cvs/cluster
Module name: cluster
Branch: RHEL4
Changes by: wcheng at sourceware.org 2007-03-13 21:21:21
Modified files:
gfs-kernel/src/gfs: daemon.c gfs_ondisk.h incore.h ioctl.c
ops_fstype.c ops_super.c rgrp.c rgrp.h
super.c super.h
Log message:
Bugzilla 220622: backport GFS2 statfs() implementation to GFS1.
There are few compromises made while porting GFS2 approach over, mostly
to avoid on-disk structure changes. GFS2 allocates (number-of-nodes + 1)
physical files into disk during mkfs time but GFS1 only has one extra
space (the unused license file) for this purpose. We deviate from GFS2
by writing the local per-node changes into a memory buffer.
Doing a "df" on a quiet filesystem now results:
dhcp145 (1 cpu HP): old df took 0.875 seconds, new df 0.008 second
dhcp146 (4 cpus DELL): old df took 0.808 seconds, new df 0.006 second.
Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/gfs-kernel/src/gfs/daemon.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.5&r2=1.5.2.1
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/gfs-kernel/src/gfs/gfs_ondisk.h.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.7.2.1&r2=1.7.2.2
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/gfs-kernel/src/gfs/incore.h.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.17.2.8&r2=1.17.2.9
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/gfs-kernel/src/gfs/ioctl.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.7.2.6&r2=1.7.2.7
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/gfs-kernel/src/gfs/ops_fstype.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.13.2.3&r2=1.13.2.4
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/gfs-kernel/src/gfs/ops_super.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.12.2.5&r2=1.12.2.6
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/gfs-kernel/src/gfs/rgrp.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.14.2.3&r2=1.14.2.4
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/gfs-kernel/src/gfs/rgrp.h.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.3&r2=1.3.2.1
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/gfs-kernel/src/gfs/super.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.12.2.7&r2=1.12.2.8
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/gfs-kernel/src/gfs/super.h.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.2&r2=1.2.2.1
--- cluster/gfs-kernel/src/gfs/daemon.c 2005/01/04 10:07:11 1.5
+++ cluster/gfs-kernel/src/gfs/daemon.c 2007/03/13 21:21:21 1.5.2.1
@@ -207,6 +207,19 @@
complete(&sdp->sd_thread_completion);
for (;;) {
+ /* Update statfs file */
+ if (gfs_tune_get(sdp, gt_statfs_fast) &&
+ time_after_eq(jiffies,
+ sdp->sd_statfs_sync_time +
+ gfs_tune_get(sdp, gt_statfs_fast) * HZ)) {
+ error = gfs_statfs_sync(sdp);
+ if (error &&
+ error != -EROFS &&
+ !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
+ printk("GFS: fsid=%s: statfs: error = %d\n",
+ sdp->sd_fsname, error);
+ sdp->sd_statfs_sync_time = jiffies;
+ }
/* Update quota file */
if (time_after_eq(jiffies,
sdp->sd_quota_sync_time +
--- cluster/gfs-kernel/src/gfs/gfs_ondisk.h 2006/11/06 21:01:36 1.7.2.1
+++ cluster/gfs-kernel/src/gfs/gfs_ondisk.h 2007/03/13 21:21:21 1.7.2.2
@@ -616,6 +616,31 @@
uint32_t ea_pad;
};
+/*
+ * Statfs change
+ * Describes an change to the pool of free and allocated
+ * blocks.
+ */
+
+struct gfs_statfs_change {
+ __be64 sc_total;
+ __be64 sc_free;
+ __be64 sc_dinodes;
+};
+
+/*
+struct gfs_statfs_change_host {
+ __u64 sc_total;
+ __u64 sc_free;
+ __u64 sc_dinodes;
+};
+*/
+struct gfs_statfs_change_host {
+ int64_t sc_total;
+ int64_t sc_free;
+ int64_t sc_dinodes;
+};
+
/* Endian functions */
#define GFS_ENDIAN_BIG
--- cluster/gfs-kernel/src/gfs/incore.h 2007/01/22 07:43:52 1.17.2.8
+++ cluster/gfs-kernel/src/gfs/incore.h 2007/03/13 21:21:21 1.17.2.9
@@ -893,6 +893,7 @@
unsigned int gt_greedy_quantum;
unsigned int gt_greedy_max;
unsigned int gt_rgrp_try_threshold;
+ unsigned int gt_statfs_fast;
};
/*
@@ -963,6 +964,13 @@
struct gfs_tune sd_tune; /* Filesystem tuning structure */
+ /* statfs */
+ struct inode *sd_statfs_inode;
+ spinlock_t sd_statfs_spin;
+ struct gfs_statfs_change_host sd_statfs_master;
+ struct gfs_statfs_change_host sd_statfs_local;
+ unsigned long sd_statfs_sync_time;
+
/* Resource group stuff */
struct gfs_inode *sd_riinode; /* Resource Index (rindex) inode */
--- cluster/gfs-kernel/src/gfs/ioctl.c 2007/01/22 07:43:52 1.7.2.6
+++ cluster/gfs-kernel/src/gfs/ioctl.c 2007/03/13 21:21:21 1.7.2.7
@@ -483,6 +483,7 @@
gfs_printf("greedy_quantum %u\n", gt->gt_greedy_quantum);
gfs_printf("greedy_max %u\n", gt->gt_greedy_max);
gfs_printf("rgrp_try_threshold %u\n", gt->gt_rgrp_try_threshold);
+ gfs_printf("statfs_fast %u\n", gt->gt_statfs_fast);
error = 0;
@@ -513,6 +514,7 @@
struct gfs_tune *gt = &sdp->sd_tune;
char param[ARG_SIZE], value[ARG_SIZE];
unsigned int x;
+ int error;
if (!capable(CAP_SYS_ADMIN))
return -EACCES;
@@ -752,6 +754,15 @@
return -EINVAL;
tune_set(gt_rgrp_try_threshold, x);
+ } else if (strcmp(param, "statfs_fast") == 0) {
+ if (sscanf(value, "%u", &x) != 1)
+ return -EINVAL;
+ error = gfs_statfs_init(sdp, x);
+ if (error)
+ return error;
+ else
+ tune_set(gt_statfs_fast, x);
+
} else
return -EINVAL;
@@ -1496,5 +1507,3 @@
return error;
}
-
-
--- cluster/gfs-kernel/src/gfs/ops_fstype.c 2006/09/15 21:42:05 1.13.2.3
+++ cluster/gfs-kernel/src/gfs/ops_fstype.c 2007/03/13 21:21:21 1.13.2.4
@@ -100,6 +100,8 @@
spin_lock_init(&sdp->sd_rg_recent_lock);
spin_lock_init(&sdp->sd_rg_forward_lock);
+ spin_lock_init(&sdp->sd_statfs_spin);
+
for (x = 0; x < GFS_GL_HASH_SIZE; x++) {
sdp->sd_gl_hash[x].hb_lock = RW_LOCK_UNLOCKED;
INIT_LIST_HEAD(&sdp->sd_gl_hash[x].hb_list);
@@ -472,9 +474,16 @@
/* Read in the license inode */
+ /* Piggy back fast df on this inode location
+ * sb->sb_quota_di.no_formal_ino = jindex_dinode + 2;
+ * sb->sb_quota_di.no_addr = jindex_dinode + 2;
+ * sb->sb_license_di.no_formal_ino = jindex_dinode + 3;
+ * sb->sb_license_di.no_addr = jindex_dinode + 3;
+ */
+
error = gfs_get_linode(sdp);
if (error) {
- printk("GFS: fsid=%s: can't get license file inode: %d\n",
+ printk("GFS: fsid=%s: can't get license/statfs file inode: %d\n",
sdp->sd_fsname, error);
goto fail_qi_free;
}
--- cluster/gfs-kernel/src/gfs/ops_super.c 2006/09/15 21:42:05 1.12.2.5
+++ cluster/gfs-kernel/src/gfs/ops_super.c 2007/03/13 21:21:21 1.12.2.6
@@ -283,6 +283,9 @@
atomic_inc(&sdp->sd_ops_super);
+ if (gfs_tune_get(sdp, gt_statfs_fast))
+ return(gfs_statfs_fast(sdp, (void *)buf));
+
error = gfs_stat_gfs(sdp, &sg, TRUE);
if (error)
return error;
--- cluster/gfs-kernel/src/gfs/rgrp.c 2005/10/24 15:53:03 1.14.2.3
+++ cluster/gfs-kernel/src/gfs/rgrp.c 2007/03/13 21:21:21 1.14.2.4
@@ -18,6 +18,7 @@
#include <asm/semaphore.h>
#include <linux/completion.h>
#include <linux/buffer_head.h>
+#include <linux/types.h>
#include "gfs.h"
#include "bits.h"
@@ -1657,6 +1658,9 @@
al->al_alloced_data++;
gfs_trans_add_quota(sdp, +1, ip->i_di.di_uid, ip->i_di.di_gid);
+
+ /* total=0, free=-1, dinodes=0 */
+ gfs_statfs_modify(sdp, 0, -1, 0);
}
/**
@@ -1711,6 +1715,9 @@
gfs_trans_add_quota(sdp, +1, ip->i_di.di_uid, ip->i_di.di_gid);
+ /* total=0, free=-1, dinode=0 */
+ gfs_statfs_modify(sdp, 0, -1, 0);
+
return 0;
}
@@ -1726,6 +1733,7 @@
int
gfs_dialloc(struct gfs_inode *dip, uint64_t *block)
{
+ struct gfs_sbd *sdp = dip->i_sbd;
struct gfs_alloc *al = dip->i_alloc;
struct gfs_rgrpd *rgd = al->al_rgd;
uint32_t goal, blk;
@@ -1765,6 +1773,9 @@
al->al_alloced_di++;
al->al_alloced_meta++;
+ /* total=0, free=-1, dinodes=1 */
+ gfs_statfs_modify(sdp, 0, -1, +1);
+
return error;
}
@@ -1797,6 +1808,8 @@
gfs_trans_add_quota(sdp, -(int64_t)blen,
ip->i_di.di_uid,
ip->i_di.di_gid);
+ /* total=0, free=+blen, dinodes=0 */
+ gfs_statfs_modify(sdp, 0, blen, 0);
}
/**
@@ -1831,6 +1844,9 @@
gfs_trans_add_bh(rgd->rd_gl, rgd->rd_bh[0]);
gfs_rgrp_out(&rgd->rd_rg, rgd->rd_bh[0]->b_data);
+ /* total=0, free=blen, dinode=0 */
+ gfs_statfs_modify(sdp, 0, blen, 0);
+
gfs_trans_add_quota(sdp, -(int64_t)blen,
ip->i_di.di_uid,
ip->i_di.di_gid);
@@ -1865,6 +1881,9 @@
gfs_trans_add_bh(rgd->rd_gl, rgd->rd_bh[0]);
gfs_rgrp_out(&rgd->rd_rg, rgd->rd_bh[0]->b_data);
+
+ /* total=0, free=1, dinodes=-1 */
+ gfs_statfs_modify(rgd->rd_sbd, 0, +1, -1);
}
/**
--- cluster/gfs-kernel/src/gfs/rgrp.h 2005/01/04 10:07:12 1.3
+++ cluster/gfs-kernel/src/gfs/rgrp.h 2007/03/13 21:21:21 1.3.2.1
@@ -57,6 +57,10 @@
void gfs_difree_uninit(struct gfs_rgrpd *rgd, uint64_t addr);
void gfs_difree(struct gfs_rgrpd *rgd, struct gfs_inode *ip);
+extern void gfs_statfs_modify(struct gfs_sbd *sdp,
+ int64_t total,
+ int64_t free,
+ int64_t dinodes);
/*
* gfs_rgrp_list
*
--- cluster/gfs-kernel/src/gfs/super.c 2007/02/06 22:08:36 1.12.2.7
+++ cluster/gfs-kernel/src/gfs/super.c 2007/03/13 21:21:21 1.12.2.8
@@ -17,6 +17,8 @@
#include <linux/spinlock.h>
#include <asm/semaphore.h>
#include <linux/completion.h>
+#include <linux/statfs.h>
+#include <linux/types.h>
#include <linux/buffer_head.h>
#include <linux/vmalloc.h>
@@ -33,6 +35,7 @@
#include "rgrp.h"
#include "super.h"
#include "unlinked.h"
+#include "trans.h"
/**
* gfs_tune_init - Fill a gfs_tune structure with default values
@@ -85,6 +88,7 @@
gt->gt_greedy_quantum = HZ / 40;
gt->gt_greedy_max = HZ / 4;
gt->gt_rgrp_try_threshold = 100;
+ gt->gt_statfs_fast = 0;
}
/**
@@ -682,6 +686,7 @@
if (error)
return error;
+ /* iopen obtained in via gfs_glock_get(..gfs_iopen_glops) */
error = gfs_inode_get(i_gh.gh_gl, &sdp->sd_sb.sb_license_di,
CREATE, &sdp->sd_linode);
@@ -770,6 +775,7 @@
!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
return error;
+ gfs_statfs_sync(sdp);
gfs_log_flush(sdp);
gfs_quota_sync(sdp);
gfs_quota_scan(sdp);
@@ -1083,3 +1089,199 @@
up(&sdp->sd_freeze_lock);
}
+
+/*
+ * 03/02/07 wcheng at redhat.com
+ * Fast statfs implementation - mostly based on GFS2 implementation.
+ */
+
+void gfs_statfs_change_in(struct gfs_statfs_change_host *sc, const void *buf)
+{
+ const struct gfs_statfs_change *str = buf;
+
+ sc->sc_total = be64_to_cpu(str->sc_total);
+ sc->sc_free = be64_to_cpu(str->sc_free);
+ sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
+}
+
+void gfs_statfs_change_out(const struct gfs_statfs_change_host *sc, void *buf)
+{
+ struct gfs_statfs_change *str = buf;
+
+ str->sc_total = cpu_to_be64(sc->sc_total);
+ str->sc_free = cpu_to_be64(sc->sc_free);
+ str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
+}
+
+int gfs_statfs_init(struct gfs_sbd *sdp, int flag)
+{
+ int error;
+
+ /* if flag == 0, do we want to turn this off ? */
+ if (!flag)
+ return 0;
+
+ error = gfs_statfs_start(sdp);
+ if (error)
+ printk("GFS: fsid=%s: can't initialize statfs subsystem: %d\n",
+ sdp->sd_fsname, error);
+
+ return error;
+}
+int gfs_statfs_start(struct gfs_sbd *sdp)
+{
+ struct gfs_stat_gfs sg;
+ struct gfs_inode *m_ip;
+ struct gfs_statfs_change_host *m_sc = &sdp->sd_statfs_master;
+ struct gfs_statfs_change_host *l_sc = &sdp->sd_statfs_local;
+ struct buffer_head *m_bh;
+ struct gfs_holder gh;
+ int error;
+
+ printk("GFS: fsid=%s: fast statfs start time = %lu\n",
+ sdp->sd_fsname, get_seconds());
+
+ /* created via gfs_get_linode() in fill_super(). */
+ /* gfs_inode_glops */
+ m_ip = sdp->sd_linode;
+
+ /* get real statistics */
+ error = gfs_stat_gfs(sdp, &sg, TRUE);
+ if (error)
+ return error;
+
+ /* make sure the page is refreshed via glock flushing */
+ error = gfs_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
+ &gh);
+ if (error)
+ goto gfs_statfs_start_out;
+
+ error = gfs_get_inode_buffer(m_ip, &m_bh);
+ if (error)
+ goto gfs_statfs_start_unlock;
+
+ error = gfs_trans_begin(sdp, 1, 0);
+ if (error)
+ goto gfs_statfs_start_bh;
+
+ spin_lock(&sdp->sd_statfs_spin);
+ m_sc->sc_total = sg.sg_total_blocks;
+ m_sc->sc_free = sg.sg_free + sg.sg_free_dinode + sg.sg_free_meta;
+ m_sc->sc_dinodes = sg.sg_used_dinode;
+ memset(l_sc, 0, sizeof(struct gfs_statfs_change_host));
+ spin_unlock(&sdp->sd_statfs_spin);
+
+ gfs_trans_add_bh(m_ip->i_gl, m_bh);
+ gfs_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs_dinode));
+
+ gfs_trans_end(sdp);
+
+gfs_statfs_start_bh:
+ brelse(m_bh);
+
+gfs_statfs_start_unlock:
+ gfs_glock_dq_uninit(&gh);
+
+gfs_statfs_start_out:
+ return 0;
+}
+
+void gfs_statfs_modify(struct gfs_sbd *sdp,
+ int64_t total,
+ int64_t free,
+ int64_t dinodes)
+{
+ struct gfs_statfs_change_host *l_sc = &sdp->sd_statfs_local;
+
+ spin_lock(&sdp->sd_statfs_spin);
+ l_sc->sc_total += total;
+ l_sc->sc_free += free;
+ l_sc->sc_dinodes += dinodes;
+ spin_unlock(&sdp->sd_statfs_spin);
+}
+
+int gfs_statfs_sync(struct gfs_sbd *sdp)
+{
+ struct gfs_inode *m_ip = sdp->sd_linode;
+ struct gfs_statfs_change_host *m_sc = &sdp->sd_statfs_master;
+ struct gfs_statfs_change_host *l_sc = &sdp->sd_statfs_local;
+ struct gfs_holder gh;
+ struct buffer_head *m_bh;
+ int error;
+
+ error = gfs_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
+ &gh);
+ if (error)
+ return error;
+
+ error = gfs_get_inode_buffer(m_ip, &m_bh);
+ if (error)
+ goto gfs_statfs_sync_out;
+
+ /* if no change, simply return */
+ spin_lock(&sdp->sd_statfs_spin);
+ gfs_statfs_change_in(m_sc, m_bh->b_data +
+ sizeof(struct gfs_dinode));
+ if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
+ spin_unlock(&sdp->sd_statfs_spin);
+ goto out_bh;
+ }
+ spin_unlock(&sdp->sd_statfs_spin);
+
+ error = gfs_trans_begin(sdp, 1, 0);
+ if (error)
+ goto out_bh;
+
+ spin_lock(&sdp->sd_statfs_spin);
+ m_sc->sc_total += l_sc->sc_total;
+ m_sc->sc_free += l_sc->sc_free;
+ m_sc->sc_dinodes += l_sc->sc_dinodes;
+ memset(l_sc, 0, sizeof(struct gfs_statfs_change_host));
+ spin_unlock(&sdp->sd_statfs_spin);
+
+ gfs_trans_add_bh(m_ip->i_gl, m_bh);
+ gfs_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs_dinode));
+
+ gfs_trans_end(sdp);
+
+out_bh:
+ brelse(m_bh);
+
+gfs_statfs_sync_out:
+ gfs_glock_dq_uninit(&gh);
+ return error;
+}
+
+int gfs_statfs_fast(struct gfs_sbd *sdp, void *b)
+{
+ struct kstatfs *buf = (struct kstatfs *)b;
+ struct gfs_statfs_change_host sc, *m_sc = &sdp->sd_statfs_master;
+ struct gfs_statfs_change_host *l_sc = &sdp->sd_statfs_local;
+
+ spin_lock(&sdp->sd_statfs_spin);
+
+ sc.sc_total = m_sc->sc_total + l_sc->sc_total;
+ sc.sc_free = m_sc->sc_free + l_sc->sc_free;
+ sc.sc_dinodes = m_sc->sc_dinodes + l_sc->sc_dinodes;
+ spin_unlock(&sdp->sd_statfs_spin);
+
+ if (sc.sc_free < 0)
+ sc.sc_free = 0;
+ if (sc.sc_free > sc.sc_total)
+ sc.sc_free = sc.sc_total;
+ if (sc.sc_dinodes < 0)
+ sc.sc_dinodes = 0;
+
+ /* fill in the statistics */
+ memset(buf, 0, sizeof(struct kstatfs));
+
+ buf->f_type = GFS_MAGIC; buf->f_bsize = sdp->sd_sb.sb_bsize;
+ buf->f_blocks = sc.sc_total;
+ buf->f_bfree = sc.sc_free;
+ buf->f_bavail = sc.sc_free;
+ buf->f_files = sc.sc_dinodes + sc.sc_free;
+ buf->f_ffree = sc.sc_free;
+ buf->f_namelen = GFS_FNAMESIZE;
+
+ return 0;
+}
--- cluster/gfs-kernel/src/gfs/super.h 2005/01/04 10:07:12 1.2
+++ cluster/gfs-kernel/src/gfs/super.h 2007/03/13 21:21:21 1.2.2.1
@@ -42,6 +42,10 @@
int gfs_make_fs_rw(struct gfs_sbd *sdp);
int gfs_make_fs_ro(struct gfs_sbd *sdp);
+int gfs_statfs_init(struct gfs_sbd *sdp, int flag);
+int gfs_statfs_sync(struct gfs_sbd *sdp);
+int gfs_statfs_fast(struct gfs_sbd *sdp, void *buf);
+
struct gfs_stat_gfs {
uint64_t sg_total_blocks;
uint64_t sg_free;
next reply other threads:[~2007-03-13 21:21 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2007-03-13 21:21 wcheng [this message]
-- strict thread matches above, loose matches on Subject: below --
2007-06-19 14:47 [Cluster-devel] cluster/gfs-kernel/src/gfs daemon.c gfs_ondisk wcheng
2007-06-19 21:26 wcheng
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20070313212122.4022.qmail@sourceware.org \
--to=wcheng@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.