linux-fsdevel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Al Viro <viro@ZenIV.linux.org.uk>
To: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org
Subject: Re: [PATCH][RFC] %pd - for printing dentry name
Date: Tue, 2 Feb 2010 17:01:57 +0000	[thread overview]
Message-ID: <20100202170157.GH12882@ZenIV.linux.org.uk> (raw)
In-Reply-To: <m1k4uwfag1.fsf@fess.ebiederm.org>

On Mon, Feb 01, 2010 at 09:55:42PM -0800, Eric W. Biederman wrote:

> We already have rename_lock, which is only take for write in d_move_locked.
> 
> I wonder if there is an instruction sequence that could guarantee that the
> string copy is done atomically from the perspective of another cpu,
> d_iname fits nicely on a single cache line so it should be possible.  
> 
> That is a stronger guarantee than we need.   All we really need is the
> guarantee that a reader will see the string null terminator.  dentries already
> have rcu safe lifetimes.
> 
> Hmm.
> 
> We should be able to do:
>         struct qstr *name;
>         int len;
>         char buf[MAX_LEN + 1];
>         long seq;
> 
>         do {
> 		seq = read_seqbegin(&rename_lock);
>         	rcu_read_lock();
> 	        name = rcu_dereference(dentry->d_name.name);
> 	        len = dentry->d_name.len;
>                 if (read_seqretry(&rename_lock, seq)
>                 	continue;
> 	        if (len > MAX_LEN)
> 	        	len = MAX_LEN;
> 		memcpy(buf, name, len);
> 	        buf[len] = '\0';
> 	        rcu_read_unlock();
> 	} while (read_seqretry(&rename_lock, seq));

Actually, WTF do we bother?  seqlock writer grabs embedded spinlock.
So can we, since *that* is far more narrow than ->d_lock and it's
static, so it's not like somebody outside of dcache.c can decide
to grab.

We could just inline string() into dname_string() and take the entire
thing to fs/dcache.c.  And protect it either with direct locking
of rename_lock.lock or another seq_writelock; contention is not
an issue, since if printk guts are your hotpath you are already
FUBAR.

Something like that (completely untested):

diff --git a/fs/dcache.c b/fs/dcache.c
index 953173a..a4d30bc 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -1703,6 +1703,48 @@ void d_move(struct dentry * dentry, struct dentry * target)
 	spin_unlock(&dcache_lock);
 }
 
+/*
+ * for vsprintf use only.
+ */
+char *dname_string(char *buf, char *end, struct dentry *d,
+		   int precision, int width, int left_align)
+{
+	int len, i;
+	const unsigned char *s;
+
+	write_seqlock(&rename_lock);
+	if (precision > d->d_name.len)
+		precision = d->d_name.len;
+
+	s = d->d_name.name;
+	if ((unsigned long)s < PAGE_SIZE)
+		s = "(null)";
+
+	len = strnlen(s, precision);
+
+	if (!left_align) {
+		while (len < width--) {
+			if (buf < end)
+				*buf = ' ';
+			++buf;
+		}
+	}
+	for (i = 0; i < len; ++i) {
+		if (buf < end)
+			*buf = *s;
+		++buf; ++s;
+	}
+
+	while (len < width--) {
+		if (buf < end)
+			*buf = ' ';
+		++buf;
+	}
+
+	write_sequnlock(&rename_lock);
+	return buf;
+}
+
 /**
  * d_ancestor - search for an ancestor
  * @p1: ancestor dentry
diff --git a/include/linux/dcache.h b/include/linux/dcache.h
index 30b93b2..2dc286a 100644
--- a/include/linux/dcache.h
+++ b/include/linux/dcache.h
@@ -379,5 +379,6 @@ extern struct vfsmount *lookup_mnt(struct path *);
 extern struct dentry *lookup_create(struct nameidata *nd, int is_dir);
 
 extern int sysctl_vfs_cache_pressure;
+extern char *dname_string(char *, char *, struct dentry *, int, int, int);
 
 #endif	/* __LINUX_DCACHE_H */
diff --git a/lib/vsprintf.c b/lib/vsprintf.c
index 3b8aeec..ab0e40a 100644
--- a/lib/vsprintf.c
+++ b/lib/vsprintf.c
@@ -915,6 +915,8 @@ static char *uuid_string(char *buf, char *end, const u8 *addr,
  *             [0][1][2][3]-[4][5]-[6][7]-[8][9]-[10][11][12][13][14][15]
  *           little endian output byte order is:
  *             [3][2][1][0]-[5][4]-[7][6]-[8][9]-[10][11][12][13][14][15]
+ * - 'd' For dentry name.  NOTE: don't use under dentry->d_lock, there
+ *       you can safely use ->d_name.name instead.
  *
  * Note: The difference between 'S' and 'F' is that on ia64 and ppc64
  * function pointers are really function descriptors, which contain a
@@ -958,6 +960,9 @@ static char *pointer(const char *fmt, char *buf, char *end, void *ptr,
 		break;
 	case 'U':
 		return uuid_string(buf, end, ptr, spec, fmt);
+	case 'd':
+		return dname_string(buf, end, ptr, spec.precision,
+				    spec.field_width, spec.flags & LEFT);
 	}
 	spec.flags |= SMALL;
 	if (spec.field_width == -1) {
@@ -1191,6 +1196,7 @@ qualifier:
  *   http://tools.ietf.org/html/draft-ietf-6man-text-addr-representation-00
  * %pU[bBlL] print a UUID/GUID in big or little endian using lower or upper
  *   case.
+ * %pd print dentry name
  * %n is ignored
  *
  * The return value is the number of characters which would

  reply	other threads:[~2010-02-02 17:02 UTC|newest]

Thread overview: 30+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2010-02-01 22:25 [PATCH][RFC] %pd - for printing dentry name Al Viro
2010-02-01 22:34 ` Al Viro
2010-02-01 22:37 ` Linus Torvalds
2010-02-01 23:18   ` Al Viro
2010-02-02  1:06     ` Al Viro
2010-02-02  5:55       ` Eric W. Biederman
2010-02-02 17:01         ` Al Viro [this message]
2010-02-02 18:10           ` Olivier Galibert
2010-02-02 19:19           ` Eric W. Biederman
2010-02-03  3:04             ` Al Viro
2010-02-04  4:53               ` Al Viro
2010-02-02  4:22     ` Linus Torvalds
2010-02-02  5:00       ` Al Viro
2010-02-02  6:36         ` Nick Piggin
2010-02-04  6:02           ` Al Viro
2010-02-04  7:40             ` Nick Piggin
2010-02-02  6:53     ` Paul E. McKenney
2010-02-02  7:09       ` Al Viro
2010-02-02 13:32         ` Matthew Wilcox
2010-02-02 15:56           ` Linus Torvalds
2010-02-02 16:13             ` Matthew Wilcox
2010-02-02 16:43           ` Al Viro
2010-02-03 10:52         ` Paul E. McKenney
2010-02-03  2:49       ` Paul E. McKenney
2010-02-04 15:29         ` Linus Torvalds
2010-02-04 16:02           ` Paul E. McKenney
2010-02-04 17:13             ` Linus Torvalds
2010-02-04 17:36               ` Al Viro
2010-02-07 16:34                 ` Paul E. McKenney
2010-02-01 22:45 ` Joe Perches

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=20100202170157.GH12882@ZenIV.linux.org.uk \
    --to=viro@zeniv.linux.org.uk \
    --cc=ebiederm@xmission.com \
    --cc=linux-fsdevel@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=torvalds@linux-foundation.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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).