From: Xiubo Li <xiubli@redhat.com>
To: Jeff Layton <jlayton@kernel.org>
Cc: idryomov@gmail.com, pdonnell@redhat.com, ceph-devel@vger.kernel.org
Subject: Re: [PATCH v2] ceph: defer flushing the capsnap if the Fb is used
Date: Sat, 9 Jan 2021 10:08:40 +0800 [thread overview]
Message-ID: <d81773a1-e843-472d-5997-e53cedcfb7b3@redhat.com> (raw)
In-Reply-To: <97ec811665eee254627c46d7999818fd94bab9ed.camel@kernel.org>
On 2021/1/9 2:24, Jeff Layton wrote:
> On Thu, 2021-01-07 at 10:30 +0800, xiubli@redhat.com wrote:
>> From: Xiubo Li <xiubli@redhat.com>
>>
>> If the Fb cap is used it means the client is flushing the dirty
>> data to OSD, just defer flushing the capsnap.
>>
>> URL: https://tracker.ceph.com/issues/48679
>> URL: https://tracker.ceph.com/issues/48640
>> Signed-off-by: Xiubo Li <xiubli@redhat.com>
>> ---
>>
>> V2:
>> - Fix inode reference leak bug
>>
>> fs/ceph/caps.c | 32 +++++++++++++++++++-------------
>> fs/ceph/snap.c | 6 +++---
>> 2 files changed, 22 insertions(+), 16 deletions(-)
>>
>> diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
>> index abbf48fc6230..2f2451d563bd 100644
>> --- a/fs/ceph/caps.c
>> +++ b/fs/ceph/caps.c
>> @@ -3047,6 +3047,7 @@ static void __ceph_put_cap_refs(struct ceph_inode_info *ci, int had,
>> {
>> struct inode *inode = &ci->vfs_inode;
>> int last = 0, put = 0, flushsnaps = 0, wake = 0;
>> + bool check_flushsnaps = false;
>>
>>
>>
>>
>> spin_lock(&ci->i_ceph_lock);
>> if (had & CEPH_CAP_PIN)
>> @@ -3064,25 +3065,15 @@ static void __ceph_put_cap_refs(struct ceph_inode_info *ci, int had,
>> if (--ci->i_wb_ref == 0) {
>> last++;
>> put++;
>> + check_flushsnaps = true;
>> }
>> dout("put_cap_refs %p wb %d -> %d (?)\n",
>> inode, ci->i_wb_ref+1, ci->i_wb_ref);
>> }
>> - if (had & CEPH_CAP_FILE_WR)
>> + if (had & CEPH_CAP_FILE_WR) {
>> if (--ci->i_wr_ref == 0) {
>> last++;
>> - if (__ceph_have_pending_cap_snap(ci)) {
>> - struct ceph_cap_snap *capsnap =
>> - list_last_entry(&ci->i_cap_snaps,
>> - struct ceph_cap_snap,
>> - ci_item);
>> - capsnap->writing = 0;
>> - if (ceph_try_drop_cap_snap(ci, capsnap))
>> - put++;
>> - else if (__ceph_finish_cap_snap(ci, capsnap))
>> - flushsnaps = 1;
>> - wake = 1;
>> - }
>> + check_flushsnaps = true;
>> if (ci->i_wrbuffer_ref_head == 0 &&
>> ci->i_dirty_caps == 0 &&
>> ci->i_flushing_caps == 0) {
>> @@ -3094,6 +3085,21 @@ static void __ceph_put_cap_refs(struct ceph_inode_info *ci, int had,
>> if (!__ceph_is_any_real_caps(ci) && ci->i_snap_realm)
>> drop_inode_snap_realm(ci);
>> }
>> + }
>> + if (check_flushsnaps) {
>> + if (__ceph_have_pending_cap_snap(ci)) {
>> + struct ceph_cap_snap *capsnap =
>> + list_last_entry(&ci->i_cap_snaps,
>> + struct ceph_cap_snap,
>> + ci_item);
>> + capsnap->writing = 0;
>> + if (ceph_try_drop_cap_snap(ci, capsnap))
>> + put++;
>> + else if (__ceph_finish_cap_snap(ci, capsnap))
>> + flushsnaps = 1;
>> + wake = 1;
>> + }
>> + }
>
> Ok, so let's assume you're putting Fb. You increment put and set
> check_flushsnaps to true. Later, you get down to here and call
> ceph_try_drop_cap_snap and it returns true and now you've incremented
> "put" twice.
>
> Is that right? Do Fb caps hold two inode references?
Yeah, one in ceph_take_cap_refs().
Another one is in ceph_queue_cap_snap() and when `used & (Fb | Fw)` is
true, the flush capsnap will be delayed by holding the inode ref, so we
need to put the inode ref here or in __ceph_finish_cap_snap().
> Either way, I think this function needs some better
> documentation/comments, particularly since you're making a significant
> change to how it works.
Okay, I will post the V3 after my back later to add more comments about
this.
Thanks
>
>> spin_unlock(&ci->i_ceph_lock);
>>
>>
>>
>>
>> dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
>> diff --git a/fs/ceph/snap.c b/fs/ceph/snap.c
>> index b611f829cb61..639fb91cc9db 100644
>> --- a/fs/ceph/snap.c
>> +++ b/fs/ceph/snap.c
>> @@ -561,10 +561,10 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci)
>> capsnap->context = old_snapc;
>> list_add_tail(&capsnap->ci_item, &ci->i_cap_snaps);
>>
>>
>>
>>
>> - if (used & CEPH_CAP_FILE_WR) {
>> + if (used & (CEPH_CAP_FILE_WR | CEPH_CAP_FILE_BUFFER)) {
>> dout("queue_cap_snap %p cap_snap %p snapc %p"
>> - " seq %llu used WR, now pending\n", inode,
>> - capsnap, old_snapc, old_snapc->seq);
>> + " seq %llu used WR | BUFFFER, now pending\n",
>> + inode, capsnap, old_snapc, old_snapc->seq);
>> capsnap->writing = 1;
>> } else {
>> /* note mtime, size NOW. */
prev parent reply other threads:[~2021-01-09 2:10 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-01-07 2:30 [PATCH v2] ceph: defer flushing the capsnap if the Fb is used xiubli
2021-01-08 18:24 ` Jeff Layton
2021-01-09 2:08 ` Xiubo Li [this message]
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=d81773a1-e843-472d-5997-e53cedcfb7b3@redhat.com \
--to=xiubli@redhat.com \
--cc=ceph-devel@vger.kernel.org \
--cc=idryomov@gmail.com \
--cc=jlayton@kernel.org \
--cc=pdonnell@redhat.com \
/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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox