On 11/13/2014 11:05 AM, Dongsheng Yang wrote:
On 11/12/2014 10:04 PM, Cyril Scetbon wrote:
Anyone on this ? There is an issue with quotas depending on the write
rate. The more we can write before a sync, the more we can exceed
quotas limits
Hi Cyril, I attempted to reproduce the problem you reported in
Linux.3.17, but failed.
It seems that the issue in this thread was fixed already. Could you
test on Linux.3.17 or later version?
Hi Cyril, after some more investigation, there actually is a problem you
described here.
*reason*
The logic in fallocate to update qgroup->excl here is:
1). btrfs_qgroup_reserve(). It reserves N_bytes.
2). btrfs_prealloc_file_range(). It record a qgroup ref into
trans->qgroup_ref_list.
3). btrfs_qgroup_free(). It clear the N_bytes from reservation.
... ...
commit_transaction().
Similar logic in writing data.
Currently qgroup->excl is updated in qgroup_excl_accounting and it is
called in commit_transation().
It means there probably is a *window* between btrfs_qgroup_free() and
commit_transaction() where
the N_bytes is not accounted into qgroup. In this *window*, more data
request will be accepted
even the total size of these requests exceeds the limit, and qgroup
can't realize it.
There is a quick fix attached on this mail, it delays the reserve_free
operation after updating qgroup->excl.
*NOTE*
It is just a quick fix for your problem here, could you help to test is
on your box? Patch is based on
Linux 3.17 (bfe01a5ba2490f299e1d2d5508cbbbadd897bbe9).
More work about btrfs quota is on the TODO list.
Thanx
Below is my log:
[root@atest-guest linux_btrfs]# uname -a
Linux atest-guest 3.17.0+ #60 SMP Thu Nov 13 06:51:55 EST 2014 x86_64
x86_64 x86_64 GNU/Linux
[root@atest-guest linux_btrfs]# btrfs qgroup show -e /mnt
qgroupid rfer excl max_excl
-------- ---- ---- --------
0/5 16384 16384 0
[root@atest-guest linux_btrfs]# btrfs sub create /mnt/sub
Create subvolume '/mnt/sub'
[root@atest-guest linux_btrfs]# sync
[root@atest-guest linux_btrfs]# btrfs qgroup show -e /mnt
qgroupid rfer excl max_excl
-------- ---- ---- --------
0/5 16384 16384 0
0/257 16384 16384 0
[root@atest-guest linux_btrfs]# btrfs qgroup limit -e 300M /mnt/sub
[root@atest-guest linux_btrfs]# btrfs qgroup show -e /mnt
qgroupid rfer excl max_excl
-------- ---- ---- --------
0/5 16384 16384 0
0/257 16384 16384 314572800
[root@atest-guest linux_btrfs]# for((i=0;i<10;i++));do dd if=/dev/zero
of=/mnt/sub/data$i bs=6M count=10; done
10+0 records in
10+0 records out
62914560 bytes (63 MB) copied, 0.0401096 s, 1.6 GB/s
10+0 records in
10+0 records out
62914560 bytes (63 MB) copied, 0.0431105 s, 1.5 GB/s
10+0 records in
10+0 records out
62914560 bytes (63 MB) copied, 0.0407304 s, 1.5 GB/s
10+0 records in
10+0 records out
62914560 bytes (63 MB) copied, 0.041441 s, 1.5 GB/s
dd: error writing \u2018/mnt/sub/data4\u2019: Disk quota exceeded
10+0 records in
9+0 records out
60817408 bytes (61 MB) copied, 0.0398122 s, 1.5 GB/s
dd: error writing \u2018/mnt/sub/data5\u2019: Disk quota exceeded
1+0 records in
0+0 records out
1703936 bytes (1.7 MB) copied, 0.00318793 s, 534 MB/s
dd: error writing \u2018/mnt/sub/data6\u2019: Disk quota exceeded
1+0 records in
0+0 records out
0 bytes (0 B) copied, 0.00239674 s, 0.0 kB/s
dd: error writing \u2018/mnt/sub/data7\u2019: Disk quota exceeded
1+0 records in
0+0 records out
0 bytes (0 B) copied, 0.002271 s, 0.0 kB/s
dd: error writing \u2018/mnt/sub/data8\u2019: Disk quota exceeded
1+0 records in
0+0 records out
0 bytes (0 B) copied, 0.00232063 s, 0.0 kB/s
dd: error writing \u2018/mnt/sub/data9\u2019: Disk quota exceeded
1+0 records in
0+0 records out
0 bytes (0 B) copied, 0.00185336 s, 0.0 kB/s
[root@atest-guest linux_btrfs]# sync
[root@atest-guest linux_btrfs]# btrfs qgroup show -e /mnt
qgroupid rfer excl max_excl
-------- ---- ---- --------
0/5 16384 16384 0
0/257 314195968 314195968 314572800
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
.
>From 8d203d87385fa56753832864ff462ad8b61b7c8f Mon Sep 17 00:00:00 2001
From: Dongsheng Yang <yangds.fnst@xxxxxxxxxxxxxx>
Date: Thu, 13 Nov 2014 12:56:06 -0500
Subject: [PATCH] btrfs: free quota reserved in delay_ref handler
Signed-off-by: Dongsheng Yang <yangds.fnst@xxxxxxxxxxxxxx>
---
fs/btrfs/extent-tree.c | 2 +-
fs/btrfs/file.c | 2 +-
fs/btrfs/qgroup.c | 23 +++++++++++++++++++++--
fs/btrfs/qgroup.h | 1 +
4 files changed, 24 insertions(+), 4 deletions(-)
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 3efe1c3..7b27ff6 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -5300,7 +5300,7 @@ void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
trace_btrfs_space_reservation(root->fs_info, "delalloc",
btrfs_ino(inode), to_free, 0);
if (root->fs_info->quota_enabled) {
- btrfs_qgroup_free(root, num_bytes +
+ btrfs_qgroup_delayed_free(root, num_bytes +
dropped * root->leafsize);
}
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index ff1cc03..58b5237 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -2609,7 +2609,7 @@ static long btrfs_fallocate(struct file *file, int mode,
out:
mutex_unlock(&inode->i_mutex);
if (root->fs_info->quota_enabled)
- btrfs_qgroup_free(root, alloc_end - alloc_start);
+ btrfs_qgroup_delayed_free(root, alloc_end - alloc_start);
out_reserve_fail:
/* Let go of our reservation. */
btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c
index ded5c60..1329500 100644
--- a/fs/btrfs/qgroup.c
+++ b/fs/btrfs/qgroup.c
@@ -73,6 +73,7 @@ struct btrfs_qgroup {
* reservation tracking
*/
u64 reserved;
+ u64 delayed_release;
/*
* lists
@@ -1409,6 +1410,11 @@ static int qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
qgroup->excl += sign * oper->num_bytes;
qgroup->excl_cmpr += sign * oper->num_bytes;
+ if (qgroup->delayed_release) {
+ qgroup->reserved -= qgroup->delayed_release;
+ qgroup->delayed_release = 0;
+ }
+
qgroup_dirty(fs_info, qgroup);
/* Get all of the parent groups that contain this qgroup */
@@ -2440,7 +2446,7 @@ out:
return ret;
}
-void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
+static void qgroup_free(struct btrfs_root *root, u64 num_bytes, int delay)
{
struct btrfs_root *quota_root;
struct btrfs_qgroup *qgroup;
@@ -2478,7 +2484,10 @@ void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
qg = u64_to_ptr(unode->aux);
- qg->reserved -= num_bytes;
+ if (delay)
+ qg->delayed_release += num_bytes;
+ else
+ qg->reserved -= num_bytes;
list_for_each_entry(glist, &qg->groups, next_group) {
ret = ulist_add(fs_info->qgroup_ulist,
@@ -2493,6 +2502,16 @@ out:
spin_unlock(&fs_info->qgroup_lock);
}
+void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
+{
+ return qgroup_free(root, num_bytes, 0);
+}
+
+void btrfs_qgroup_delayed_free(struct btrfs_root *root, u64 num_bytes)
+{
+ return qgroup_free(root, num_bytes, 1);
+}
+
void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
{
if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
diff --git a/fs/btrfs/qgroup.h b/fs/btrfs/qgroup.h
index 18cc68c..1c50238 100644
--- a/fs/btrfs/qgroup.h
+++ b/fs/btrfs/qgroup.h
@@ -97,6 +97,7 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
struct btrfs_qgroup_inherit *inherit);
int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes);
void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes);
+void btrfs_qgroup_delayed_free(struct btrfs_root *root, u64 num_bytes);
void assert_qgroups_uptodate(struct btrfs_trans_handle *trans);
--
1.8.3.1
