Re: [Ocfs2-devel] [PATCH] ocfs2: fix the application IO timeout when fstrim is running

From: Changwei Ge
Date: Tue Jan 15 2019 - 04:33:57 EST


Hi Gang,

It looks good to me.

On 2019/1/15 16:23, Gang He wrote:
> Hello ChangWei,
>
>>>> On 2019/1/15 at 16:00, in message
> <63ADC13FD55D6546B7DECE290D39E37301278265F3@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
> Changwei Ge <ge.changwei@xxxxxxx> wrote:
>> On 2019/1/15 13:49, Gang He wrote:
>>> Hello Changewei,
>>>
>>>>>> On 2019/1/15 at 11:50, in message
>>> <63ADC13FD55D6546B7DECE290D39E3730127825EFD@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
>>> Changwei Ge <ge.changwei@xxxxxxx> wrote:
>>>> Hi Gang,
>>>>
>>>> Most parts of this patch look sane to me, just a tiny question...
>>>>
>>>> On 2019/1/11 17:01, Gang He wrote:
>>>>> The user reported this problem, the upper application IO was
>>>>> timeout when fstrim was running on this ocfs2 partition. the
>>>>> application monitoring resource agent considered that this
>>>>> application did not work, then this node was fenced by the cluster
>>>>> brain (e.g. pacemaker).
>>>>> The root cause is that fstrim thread always holds main_bm meta-file
>>>>> related locks until all the cluster groups are trimmed.
>>>>> This patch will make fstrim thread release main_bm meta-file
>>>>> related locks when each cluster group is trimmed, this will let
>>>>> the current application IO has a chance to claim the clusters from
>>>>> main_bm meta-file.
>>>>>
>>>>> Signed-off-by: Gang He <ghe@xxxxxxxx>

Reviewed-by: Changwei Ge <ge.changwei@xxxxxxx>

>>>>> ---
>>>>> fs/ocfs2/alloc.c | 159 +++++++++++++++++++++++++----------------
>>>>> fs/ocfs2/dlmglue.c | 5 ++
>>>>> fs/ocfs2/ocfs2.h | 1 +
>>>>> fs/ocfs2/ocfs2_trace.h | 2 +
>>>>> fs/ocfs2/super.c | 2 +
>>>>> 5 files changed, 106 insertions(+), 63 deletions(-)
>>>>>
>>>>> diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
>>>>> index d1cbb27808e2..6f0999015a44 100644
>>>>> --- a/fs/ocfs2/alloc.c
>>>>> +++ b/fs/ocfs2/alloc.c
>>>>> @@ -7532,10 +7532,11 @@ static int ocfs2_trim_group(struct super_block *sb,
>>>>> return count;
>>>>> }
>>>>>
>>>>> -int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range)
>>>>> +static
>>>>> +int ocfs2_trim_mainbm(struct super_block *sb, struct fstrim_range *range)
>>>>> {
>>>>> struct ocfs2_super *osb = OCFS2_SB(sb);
>>>>> - u64 start, len, trimmed, first_group, last_group, group;
>>>>> + u64 start, len, trimmed = 0, first_group, last_group = 0, group = 0;
>>>>> int ret, cnt;
>>>>> u32 first_bit, last_bit, minlen;
>>>>> struct buffer_head *main_bm_bh = NULL;
>>>>> @@ -7543,7 +7544,6 @@ int ocfs2_trim_fs(struct super_block *sb, struct
>>>> fstrim_range *range)
>>>>> struct buffer_head *gd_bh = NULL;
>>>>> struct ocfs2_dinode *main_bm;
>>>>> struct ocfs2_group_desc *gd = NULL;
>>>>> - struct ocfs2_trim_fs_info info, *pinfo = NULL;
>>>>>
>>>>> start = range->start >> osb->s_clustersize_bits;
>>>>> len = range->len >> osb->s_clustersize_bits;
>>>>> @@ -7552,6 +7552,9 @@ int ocfs2_trim_fs(struct super_block *sb, struct
>>>> fstrim_range *range)
>>>>> if (minlen >= osb->bitmap_cpg || range->len < sb->s_blocksize)
>>>>> return -EINVAL;
>>>>>
>>>>> + trace_ocfs2_trim_mainbm(start, len, minlen);
>>>>> +
>>>>> +next_group:
>>>>> main_bm_inode = ocfs2_get_system_file_inode(osb,
>>>>> GLOBAL_BITMAP_SYSTEM_INODE,
>>>>> OCFS2_INVALID_SLOT);
>>>>> @@ -7570,64 +7573,34 @@ int ocfs2_trim_fs(struct super_block *sb, struct
>>>> fstrim_range *range)
>>>>> }
>>>>> main_bm = (struct ocfs2_dinode *)main_bm_bh->b_data;
>>>>>
>>>>> - if (start >= le32_to_cpu(main_bm->i_clusters)) {
>>>>> - ret = -EINVAL;
>>>>> - goto out_unlock;
>>>>> - }
>>>>> -
>>>>> - len = range->len >> osb->s_clustersize_bits;
>>>>> - if (start + len > le32_to_cpu(main_bm->i_clusters))
>>>>> - len = le32_to_cpu(main_bm->i_clusters) - start;
>>>>> -
>>>>> - trace_ocfs2_trim_fs(start, len, minlen);
>>>>> -
>>>>> - ocfs2_trim_fs_lock_res_init(osb);
>>>>> - ret = ocfs2_trim_fs_lock(osb, NULL, 1);
>>>>> - if (ret < 0) {
>>>>> - if (ret != -EAGAIN) {
>>>>> - mlog_errno(ret);
>>>>> - ocfs2_trim_fs_lock_res_uninit(osb);
>>>>> + /*
>>>>> + * Do some check before trim the first group.
>>>>> + */
>>>>> + if (!group) {
>>>>> + if (start >= le32_to_cpu(main_bm->i_clusters)) {
>>>>> + ret = -EINVAL;
>>>>> goto out_unlock;
>>>>> }
>>>>>
>>>>> - mlog(ML_NOTICE, "Wait for trim on device (%s) to "
>>>>> - "finish, which is running from another node.\n",
>>>>> - osb->dev_str);
>>>>> - ret = ocfs2_trim_fs_lock(osb, &info, 0);
>>>>> - if (ret < 0) {
>>>>> - mlog_errno(ret);
>>>>> - ocfs2_trim_fs_lock_res_uninit(osb);
>>>>> - goto out_unlock;
>>>>> - }
>>>>> + if (start + len > le32_to_cpu(main_bm->i_clusters))
>>>>> + len = le32_to_cpu(main_bm->i_clusters) - start;
>>>>>
>>>>> - if (info.tf_valid && info.tf_success &&
>>>>> - info.tf_start == start && info.tf_len == len &&
>>>>> - info.tf_minlen == minlen) {
>>>>> - /* Avoid sending duplicated trim to a shared device */
>>>>> - mlog(ML_NOTICE, "The same trim on device (%s) was "
>>>>> - "just done from node (%u), return.\n",
>>>>> - osb->dev_str, info.tf_nodenum);
>>>>> - range->len = info.tf_trimlen;
>>>>> - goto out_trimunlock;
>>>>> - }
>>>>> + /*
>>>>> + * Determine first and last group to examine based on
>>>>> + * start and len
>>>>> + */
>>>>> + first_group = ocfs2_which_cluster_group(main_bm_inode, start);
>>>>> + if (first_group == osb->first_cluster_group_blkno)
>>>>> + first_bit = start;
>>>>> + else
>>>>> + first_bit = start - ocfs2_blocks_to_clusters(sb,
>>>>> + first_group);
>>>>> + last_group = ocfs2_which_cluster_group(main_bm_inode,
>>>>> + start + len - 1);
>>>>> + group = first_group;
>>>>> }
>>>>>
>>>>> - info.tf_nodenum = osb->node_num;
>>>>> - info.tf_start = start;
>>>>> - info.tf_len = len;
>>>>> - info.tf_minlen = minlen;
>>>>> -
>>>>> - /* Determine first and last group to examine based on start and len */
>>>>> - first_group = ocfs2_which_cluster_group(main_bm_inode, start);
>>>>> - if (first_group == osb->first_cluster_group_blkno)
>>>>> - first_bit = start;
>>>>> - else
>>>>> - first_bit = start - ocfs2_blocks_to_clusters(sb, first_group);
>>>>> - last_group = ocfs2_which_cluster_group(main_bm_inode, start + len - 1);
>>>>> - last_bit = osb->bitmap_cpg;
>>>>> -
>>>>> - trimmed = 0;
>>>>> - for (group = first_group; group <= last_group;) {
>>>>> + do {
>>>>> if (first_bit + len >= osb->bitmap_cpg)
>>>>> last_bit = osb->bitmap_cpg;
>>>>> else
>>>>> @@ -7659,21 +7632,81 @@ int ocfs2_trim_fs(struct super_block *sb, struct
>>>> fstrim_range *range)
>>>>> group = ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
>>>>> else
>>>>> group += ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
>>>>> - }
>>>>> - range->len = trimmed * sb->s_blocksize;
>>>>> + } while (0);
>>>>>
>>>>> - info.tf_trimlen = range->len;
>>>>> - info.tf_success = (ret ? 0 : 1);
>>>>> - pinfo = &info;
>>>>> -out_trimunlock:
>>>>> - ocfs2_trim_fs_unlock(osb, pinfo);
>>>>> - ocfs2_trim_fs_lock_res_uninit(osb);
>>>>> out_unlock:
>>>>> ocfs2_inode_unlock(main_bm_inode, 0);
>>>>> brelse(main_bm_bh);
>>>>> + main_bm_bh = NULL;
>>>>> out_mutex:
>>>>> inode_unlock(main_bm_inode);
>>>>> iput(main_bm_inode);
>>>>> +
>>>>> + /*
>>>>> + * If all the groups trim are not done or failed, but we should release
>>>>> + * main_bm related locks for avoiding the current IO starve, then go to
>>>>> + * trim the next group
>>>>> + */
>>>>> + if (ret >= 0 && group <= last_group)
>>>>> + goto next_group;
>>>>> out:
>>>>> + range->len = trimmed * sb->s_blocksize;
>>>>> + return ret;
>>>>> +}
>>>>> +
>>>>> +int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range)
>>>>> +{
>>>>> + int ret;
>>>>> + struct ocfs2_super *osb = OCFS2_SB(sb);
>>>>> + struct ocfs2_trim_fs_info info, *pinfo = NULL;
>>>>> +
>>>>> + ocfs2_trim_fs_lock_res_init(osb);
>>>>> +
>>>>> + trace_ocfs2_trim_fs(range->start, range->len, range->minlen);
>>>>> +
>>>>> + ret = ocfs2_trim_fs_lock(osb, NULL, 1);
>>>>> + if (ret < 0) {
>>>>> + if (ret != -EAGAIN) {
>>>>> + mlog_errno(ret);
>>>>> + ocfs2_trim_fs_lock_res_uninit(osb);
>>>>> + return ret;
>>>>> + }
>>>>> +
>>>>> + mlog(ML_NOTICE, "Wait for trim on device (%s) to "
>>>>> + "finish, which is running from another node.\n",
>>>>> + osb->dev_str);
>>>>> + ret = ocfs2_trim_fs_lock(osb, &info, 0);
>>>>> + if (ret < 0) {
>>>>> + mlog_errno(ret);
>>>>> + ocfs2_trim_fs_lock_res_uninit(osb);
>>>>> + return ret;
>>>>> + }
>>>>> +
>>>>> + if (info.tf_valid && info.tf_success &&
>>>>> + info.tf_start == range->start &&
>>>>> + info.tf_len == range->len &&
>>>>> + info.tf_minlen == range->minlen) {
>>>>> + /* Avoid sending duplicated trim to a shared device */
>>>>> + mlog(ML_NOTICE, "The same trim on device (%s) was "
>>>>> + "just done from node (%u), return.\n",
>>>>> + osb->dev_str, info.tf_nodenum);
>>>>> + range->len = info.tf_trimlen;
>>>>> + goto out;
>>>>> + }
>>>>> + }
>>>>> +
>>>>> + info.tf_nodenum = osb->node_num;
>>>>> + info.tf_start = range->start;
>>>>> + info.tf_len = range->len;
>>>>> + info.tf_minlen = range->minlen;
>>>>> +
>>>>> + ret = ocfs2_trim_mainbm(sb, range);
>>>>> +
>>>>> + info.tf_trimlen = range->len;
>>>>> + info.tf_success = (ret < 0 ? 0 : 1);
>>>>> + pinfo = &info;
>>>>> +out:
>>>>> + ocfs2_trim_fs_unlock(osb, pinfo);
>>>>> + ocfs2_trim_fs_lock_res_uninit(osb);
>>>>> return ret;
>>>>> }
>>>>> diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
>>>>> index 7c835824247e..af405586c5b1 100644
>>>>> --- a/fs/ocfs2/dlmglue.c
>>>>> +++ b/fs/ocfs2/dlmglue.c
>>>>> @@ -686,6 +686,9 @@ void ocfs2_trim_fs_lock_res_init(struct ocfs2_super *osb)
>>>>> {
>>>>> struct ocfs2_lock_res *lockres = &osb->osb_trim_fs_lockres;
>>>>>
>>>>> + /* Only one trimfs thread are allowed to work at the same time. */
>>>>> + mutex_lock(&osb->obs_trim_fs_mutex);
>>>>> +
>>>>
>>>> Cluster lock of fstrim have a trylock behavior, will it be better if we
>>>> trylock here?
>>> Here, I prefer to just serialize fstrim threads on the local node to
>> simplify the code logic,
>>> maybe the user want to do like that, although this behavior is not
>> recommended.
>>> You know, on one node, ideally, the user should call fstrim command once
>> regularly.
>>> If he calls fstrim command more times in a very short time,
>>> the code will not make each fstrim command return to failure, just do the
>> fstrim task one by one.
>>
>> I have a thought having nothing to do with your patch.
>> Do you think it's possible for us to implement *discard on unlink file* like
>> ext4 does.
>> So the application doesn't have to invoke fstrim periodically.
> Yes, we can do some investigation for adding a mount option "discard" to support discard a file when it is deleted.
> This can be considered as another feature, since its discard occasion and granularity is different with fstrim.
> The scheduled fstrim command can considered as a traditional file system level trim.
> If the file system supports the on-demand discard when some blocks were released, that's better.
> Of course. these two features can coexist.

Good to hear this. :-)

>
> Thanks
> Gang
>
>>
>> Thanks,
>> Changwei
>>
>>>
>>> Thanks
>>> Gang
>>>
>>>>
>>>> Thanks,
>>>> Changwei
>>>>
>>>>> ocfs2_lock_res_init_once(lockres);
>>>>> ocfs2_build_lock_name(OCFS2_LOCK_TYPE_TRIM_FS, 0, 0, lockres->l_name);
>>>>> ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_TRIM_FS,
>>>>> @@ -698,6 +701,8 @@ void ocfs2_trim_fs_lock_res_uninit(struct ocfs2_super
>>>> *osb)
>>>>>
>>>>> ocfs2_simple_drop_lockres(osb, lockres);
>>>>> ocfs2_lock_res_free(lockres);
>>>>> +
>>>>> + mutex_unlock(&osb->obs_trim_fs_mutex);
>>>>> }
>>>>>
>>>>> static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res,
>>>>> diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
>>>>> index 4f86ac0027b5..1f029fbe8b8d 100644
>>>>> --- a/fs/ocfs2/ocfs2.h
>>>>> +++ b/fs/ocfs2/ocfs2.h
>>>>> @@ -407,6 +407,7 @@ struct ocfs2_super
>>>>> struct ocfs2_lock_res osb_rename_lockres;
>>>>> struct ocfs2_lock_res osb_nfs_sync_lockres;
>>>>> struct ocfs2_lock_res osb_trim_fs_lockres;
>>>>> + struct mutex obs_trim_fs_mutex;
>>>>> struct ocfs2_dlm_debug *osb_dlm_debug;
>>>>>
>>>>> struct dentry *osb_debug_root;
>>>>> diff --git a/fs/ocfs2/ocfs2_trace.h b/fs/ocfs2/ocfs2_trace.h
>>>>> index 2ee76a90ba8f..dc4bce1649c1 100644
>>>>> --- a/fs/ocfs2/ocfs2_trace.h
>>>>> +++ b/fs/ocfs2/ocfs2_trace.h
>>>>> @@ -712,6 +712,8 @@ TRACE_EVENT(ocfs2_trim_extent,
>>>>>
>>>>> DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(ocfs2_trim_group);
>>>>>
>>>>> +DEFINE_OCFS2_ULL_ULL_ULL_EVENT(ocfs2_trim_mainbm);
>>>>> +
>>>>> DEFINE_OCFS2_ULL_ULL_ULL_EVENT(ocfs2_trim_fs);
>>>>>
>>>>> /* End of trace events for fs/ocfs2/alloc.c. */
>>>>> diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
>>>>> index 3415e0b09398..96ae7cedd487 100644
>>>>> --- a/fs/ocfs2/super.c
>>>>> +++ b/fs/ocfs2/super.c
>>>>> @@ -1847,6 +1847,8 @@ static int ocfs2_mount_volume(struct super_block *sb)
>>>>> if (ocfs2_is_hard_readonly(osb))
>>>>> goto leave;
>>>>>
>>>>> + mutex_init(&osb->obs_trim_fs_mutex);
>>>>> +
>>>>> status = ocfs2_dlm_init(osb);
>>>>> if (status < 0) {
>>>>> mlog_errno(status);
>>>>>
>>>
>