All of lore.kernel.org
 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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.