From: Peter Zijlstra <peterz@infradead.org>
To: Al Viro <viro@ZenIV.linux.org.uk>,
Linus Torvalds <torvalds@linux-foundation.org>,
Chris Mason <clm@fb.com>
Cc: linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
David Howells <dhowells@redhat.com>,
elena.reshetova@intel.com, ishkamiel@gmail.com,
dwindsor@gmail.com, gregkh@linuxfoundation.org,
peterz@infradead.org
Subject: [RFC][PATCH 06/10] fs: Rework i_count
Date: Fri, 24 Feb 2017 16:43:35 +0100 [thread overview]
Message-ID: <20170224162044.267733351@infradead.org> (raw)
In-Reply-To: <20170224154329.478276481@infradead.org>
[-- Attachment #1: peterz-fs-inode-3d2.patch --]
[-- Type: text/plain, Size: 4012 bytes --]
The inode count: i_count, does not conform to normal reference
counting semantics, its a usage count. That is, objects can and do
live without any references (i_count == 0).
This is because the reference from the inode hash table is not
accounted. This makes that find_inode_fast() can result in 0->1
transitions and similarly, iput() can do 1->0->1 transitions.
This patch changes things to include the inode hash table's reference
and reworks iput() to avoid spurious drops to 0.
It does mean that we now call super_operations::drop_inode() with
non-zero i_count, but I could not find an implementation where this
mattered. Only once we really decide to fully drop the inode and
remove it from the hash will we do the final dec_and_test.
This basically boils down to pushing part of iput() into
atomic_dec_and_lock().
Also, this would allow an RCU based find_inode*().
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
fs/btrfs/inode.c | 2 +-
fs/inode.c | 24 +++++++++++++++++++-----
include/linux/fs.h | 11 ++++++++++-
3 files changed, 30 insertions(+), 7 deletions(-)
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -3172,7 +3172,7 @@ void btrfs_add_delayed_iput(struct inode
struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
struct btrfs_inode *binode = BTRFS_I(inode);
- if (atomic_add_unless(&inode->i_count, -1, 1))
+ if (atomic_add_unless(&inode->i_count, -1, 2))
return;
spin_lock(&fs_info->delayed_iput_lock);
--- a/fs/inode.c
+++ b/fs/inode.c
@@ -135,7 +135,7 @@ int inode_init_always(struct super_block
inode->i_sb = sb;
inode->i_blkbits = sb->s_blocksize_bits;
inode->i_flags = 0;
- atomic_set(&inode->i_count, 1);
+ atomic_set(&inode->i_count, 2); /* hashed and ref */
inode->i_op = &empty_iops;
inode->i_fop = &no_open_fops;
inode->__i_nlink = 1;
@@ -387,7 +387,7 @@ static void init_once(void *foo)
void __iget(struct inode *inode)
{
lockdep_assert_held(&inode->i_lock);
- atomic_inc(&inode->i_count);
+ WARN_ON(!atomic_inc_not_zero(&inode->i_count));
}
/*
@@ -395,7 +395,7 @@ void __iget(struct inode *inode)
*/
void ihold(struct inode *inode)
{
- WARN_ON(atomic_inc_return(&inode->i_count) < 2);
+ WARN_ON(atomic_inc_return(&inode->i_count) < 3);
}
EXPORT_SYMBOL(ihold);
@@ -814,6 +814,8 @@ static struct inode *find_inode_fast(str
{
struct inode *inode = NULL;
+ lockdep_assert_held(&inode_hash_lock);
+
repeat:
hlist_for_each_entry(inode, head, i_hash) {
if (inode->i_ino != ino)
@@ -1488,11 +1490,12 @@ void iput(struct inode *inode)
BUG_ON(inode->i_state & I_CLEAR);
retry:
- if (!atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
+ if (atomic_add_unless(&inode->i_count, -1, 2))
return;
+ spin_lock(&inode->i_lock);
+
if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
- atomic_inc(&inode->i_count);
inode->i_state &= ~I_DIRTY_TIME;
spin_unlock(&inode->i_lock);
trace_writeback_lazytime_iput(inode);
@@ -1500,6 +1503,7 @@ void iput(struct inode *inode)
goto retry;
}
+ atomic_dec(&inode->i_count); /* 2 -> 1 */
WARN_ON(inode->i_state & I_NEW);
/*
@@ -1520,6 +1524,16 @@ void iput(struct inode *inode)
spin_unlock(&inode->i_lock);
return;
}
+
+ /*
+ * If, at this point, only the hashtable has a reference left
+ * continue to take the inode out, otherwise someone got a ref
+ * while we weren't looking.
+ */
+ if (atomic_cmpxchg(&inode->i_count, 1, 0) != 1) {
+ spin_unlock(&inode->i_lock);
+ return;
+ }
if (!drop) {
inode->i_state |= I_WILL_FREE;
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -2713,7 +2713,16 @@ extern unsigned int get_next_ino(void);
static inline int i_count(struct inode *inode)
{
- return atomic_read(&inode->i_count);
+ int i_count = atomic_read(&inode->i_count);
+
+ /*
+ * In order to preserve the 'old' usage-count semantics, remove the
+ * reference that the hash-table has.
+ */
+ if (i_count)
+ i_count--;
+
+ return i_count;
}
extern void __iget(struct inode * inode);
next prev parent reply other threads:[~2017-02-24 18:32 UTC|newest]
Thread overview: 26+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-02-24 15:43 [RFC][PATCH 00/10] On inode::i_count and the usage vs reference count issue Peter Zijlstra
2017-02-24 15:43 ` [RFC][PATCH 01/10] fs: Use lockdep_assert_held() instead of comments Peter Zijlstra
2017-02-24 15:43 ` [RFC][PATCH 02/10] fs: Avoid looking at i_count without i_lock held Peter Zijlstra
[not found] ` <CA+55aFxLw8FXf61rsGYDjA1tS=joDeaF7OSgaepLWwcz4zt=dg@mail.gmail.com>
2017-02-24 17:06 ` Peter Zijlstra
2017-02-24 15:43 ` [RFC][PATCH 03/10] fs: Introduce i_count() Peter Zijlstra
2017-02-24 15:43 ` [RFC][PATCH 04/10] fs: Restructure iput() Peter Zijlstra
2017-02-24 15:43 ` [RFC][PATCH 05/10] fs: Remove iput_final() Peter Zijlstra
2017-02-24 15:43 ` Peter Zijlstra [this message]
2017-02-24 20:49 ` [RFC][PATCH 06/10] fs: Rework i_count Al Viro
2017-02-24 15:43 ` [RFC][PATCH 07/10] orangefs: Use RCU for destroy_inode Peter Zijlstra
2017-02-24 20:52 ` Al Viro
2017-02-24 23:00 ` Mike Marshall
2017-02-25 20:31 ` Mike Marshall
2017-02-27 0:34 ` Mike Marshall
2017-02-27 1:20 ` Linus Torvalds
2017-02-27 8:44 ` David Howells
2017-02-27 14:44 ` Mike Marshall
2017-02-24 15:43 ` [RFC][PATCH 08/10] fs: Do RCU versions for find_inode() Peter Zijlstra
2017-02-24 15:43 ` [RFC][PATCH 09/10] locking/refcount: Provide refcount_dec_unless() Peter Zijlstra
2017-02-27 9:28 ` Reshetova, Elena
2017-02-24 15:43 ` [RFC][PATCH 10/10] fs: Convert i_count over to refcount_t Peter Zijlstra
2017-02-24 16:43 ` [RFC][PATCH 00/10] On inode::i_count and the usage vs reference count issue Christoph Hellwig
2017-02-24 17:07 ` Peter Zijlstra
2017-02-24 20:59 ` David Windsor
[not found] ` <CA+55aFy1bNbsX_3T-s_EUwTP-r_SmJJMvB3=-2nffehFVP=EdQ@mail.gmail.com>
[not found] ` <CA+55aFz0DbAGZ8gc+s35nm1N5frXjK_NOh7QzuSfZeJbjsT6Sg@mail.gmail.com>
[not found] ` <CA+55aFyR8wkHps5_AqUqzx8MDMNxRZZ7+MYH9g=ZCUi=4Oey8w@mail.gmail.com>
2017-02-24 19:24 ` Fwd: " Linus Torvalds
2017-02-24 20:42 ` Al Viro
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=20170224162044.267733351@infradead.org \
--to=peterz@infradead.org \
--cc=clm@fb.com \
--cc=dhowells@redhat.com \
--cc=dwindsor@gmail.com \
--cc=elena.reshetova@intel.com \
--cc=gregkh@linuxfoundation.org \
--cc=ishkamiel@gmail.com \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=torvalds@linux-foundation.org \
--cc=viro@ZenIV.linux.org.uk \
/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
Powered by JetHome