mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Ben Hutchings <ben@decadent.org.uk>
To: linux-kernel@vger.kernel.org, stable@vger.kernel.org
Cc: akpm@linux-foundation.org, "Al Viro" <viro@zeniv.linux.org.uk>
Subject: [PATCH 3.2 14/27] deal with deadlock in d_walk()
Date: Mon, 29 Dec 2014 02:11:31 +0100	[thread overview]
Message-ID: <lsq.1419815491.439721348@decadent.org.uk> (raw)
In-Reply-To: <lsq.1419815490.288081487@decadent.org.uk>

3.2.66-rc1 review patch.  If anyone has any objections, please let me know.

------------------

From: Al Viro <viro@zeniv.linux.org.uk>

commit ca5358ef75fc69fee5322a38a340f5739d997c10 upstream.

... by not hitting rename_retry for reasons other than rename having
happened.  In other words, do _not_ restart when finding that
between unlocking the child and locking the parent the former got
into __dentry_kill().  Skip the killed siblings instead...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
[bwh: Backported to 3.2:
 - As we only have try_to_ascend() and not d_walk(), apply this
   change to all callers of try_to_ascend()
 - Adjust context to make __dentry_kill() apply to d_kill()]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
---
 fs/dcache.c | 31 ++++++++++++++++---------------
 1 file changed, 16 insertions(+), 15 deletions(-)

--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -306,9 +306,9 @@ static struct dentry *d_kill(struct dent
 	__releases(parent->d_lock)
 	__releases(dentry->d_inode->i_lock)
 {
-	list_del(&dentry->d_child);
+	__list_del_entry(&dentry->d_child);
 	/*
-	 * Inform try_to_ascend() that we are no longer attached to the
+	 * Inform ascending readers that we are no longer attached to the
 	 * dentry tree
 	 */
 	dentry->d_flags |= DCACHE_DENTRY_KILLED;
@@ -949,34 +949,6 @@ void shrink_dcache_for_umount(struct sup
 	}
 }
 
-/*
- * This tries to ascend one level of parenthood, but
- * we can race with renaming, so we need to re-check
- * the parenthood after dropping the lock and check
- * that the sequence number still matches.
- */
-static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
-{
-	struct dentry *new = old->d_parent;
-
-	rcu_read_lock();
-	spin_unlock(&old->d_lock);
-	spin_lock(&new->d_lock);
-
-	/*
-	 * might go back up the wrong parent if we have had a rename
-	 * or deletion
-	 */
-	if (new != old->d_parent ||
-		 (old->d_flags & DCACHE_DENTRY_KILLED) ||
-		 (!locked && read_seqretry(&rename_lock, seq))) {
-		spin_unlock(&new->d_lock);
-		new = NULL;
-	}
-	rcu_read_unlock();
-	return new;
-}
-
 
 /*
  * Search for at least 1 mount point in the dentry's subdirs.
@@ -1032,17 +1004,32 @@ resume:
 	/*
 	 * All done at this level ... ascend and resume the search.
 	 */
+	rcu_read_lock();
+ascend:
 	if (this_parent != parent) {
 		struct dentry *child = this_parent;
-		this_parent = try_to_ascend(this_parent, locked, seq);
-		if (!this_parent)
+		this_parent = child->d_parent;
+
+		spin_unlock(&child->d_lock);
+		spin_lock(&this_parent->d_lock);
+
+		/* might go back up the wrong parent if we have had a rename */
+		if (!locked && read_seqretry(&rename_lock, seq))
 			goto rename_retry;
 		next = child->d_child.next;
+		while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
+			if (next == &this_parent->d_subdirs)
+				goto ascend;
+			child = list_entry(next, struct dentry, d_child);
+			next = next->next;
+		}
+		rcu_read_unlock();
 		goto resume;
 	}
-	spin_unlock(&this_parent->d_lock);
 	if (!locked && read_seqretry(&rename_lock, seq))
 		goto rename_retry;
+	spin_unlock(&this_parent->d_lock);
+	rcu_read_unlock();
 	if (locked)
 		write_sequnlock(&rename_lock);
 	return 0; /* No mount points found in tree */
@@ -1054,6 +1041,8 @@ positive:
 	return 1;
 
 rename_retry:
+	spin_unlock(&this_parent->d_lock);
+	rcu_read_unlock();
 	if (locked)
 		goto again;
 	locked = 1;
@@ -1139,23 +1128,40 @@ resume:
 	/*
 	 * All done at this level ... ascend and resume the search.
 	 */
+	rcu_read_lock();
+ascend:
 	if (this_parent != parent) {
 		struct dentry *child = this_parent;
-		this_parent = try_to_ascend(this_parent, locked, seq);
-		if (!this_parent)
+		this_parent = child->d_parent;
+
+		spin_unlock(&child->d_lock);
+		spin_lock(&this_parent->d_lock);
+
+		/* might go back up the wrong parent if we have had a rename */
+		if (!locked && read_seqretry(&rename_lock, seq))
 			goto rename_retry;
 		next = child->d_child.next;
+		while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
+			if (next == &this_parent->d_subdirs)
+				goto ascend;
+			child = list_entry(next, struct dentry, d_child);
+			next = next->next;
+		}
+		rcu_read_unlock();
 		goto resume;
 	}
 out:
-	spin_unlock(&this_parent->d_lock);
 	if (!locked && read_seqretry(&rename_lock, seq))
 		goto rename_retry;
+	spin_unlock(&this_parent->d_lock);
+	rcu_read_unlock();
 	if (locked)
 		write_sequnlock(&rename_lock);
 	return found;
 
 rename_retry:
+	spin_unlock(&this_parent->d_lock);
+	rcu_read_unlock();
 	if (found)
 		return found;
 	if (locked)
@@ -2914,26 +2920,43 @@ resume:
 		}
 		spin_unlock(&dentry->d_lock);
 	}
+	rcu_read_lock();
+ascend:
 	if (this_parent != root) {
 		struct dentry *child = this_parent;
 		if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
 			this_parent->d_flags |= DCACHE_GENOCIDE;
 			this_parent->d_count--;
 		}
-		this_parent = try_to_ascend(this_parent, locked, seq);
-		if (!this_parent)
+		this_parent = child->d_parent;
+
+		spin_unlock(&child->d_lock);
+		spin_lock(&this_parent->d_lock);
+
+		/* might go back up the wrong parent if we have had a rename */
+		if (!locked && read_seqretry(&rename_lock, seq))
 			goto rename_retry;
 		next = child->d_child.next;
+		while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
+			if (next == &this_parent->d_subdirs)
+				goto ascend;
+			child = list_entry(next, struct dentry, d_child);
+			next = next->next;
+		}
+		rcu_read_unlock();
 		goto resume;
 	}
-	spin_unlock(&this_parent->d_lock);
 	if (!locked && read_seqretry(&rename_lock, seq))
 		goto rename_retry;
+	spin_unlock(&this_parent->d_lock);
+	rcu_read_unlock();
 	if (locked)
 		write_sequnlock(&rename_lock);
 	return;
 
 rename_retry:
+	spin_unlock(&this_parent->d_lock);
+	rcu_read_unlock();
 	if (locked)
 		goto again;
 	locked = 1;


  parent reply	other threads:[~2014-12-29  1:16 UTC|newest]

Thread overview: 33+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-12-29  1:11 [PATCH 3.2 00/27] 3.2.66-rc1 review Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 25/27] drivers/net: macvtap and tun depend on INET Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 24/27] ipv4: dst_entry leak in ip_send_unicast_reply() Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 03/27] sata_fsl: fix error handling of irq_of_parse_and_map Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 12/27] x86, kvm: Clear paravirt_enabled on KVM guests for espfix32's benefit Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 11/27] [media] ttusb-dec: buffer overflow in ioctl Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 02/27] AHCI: Add DeviceIDs for Sunrise Point-LP SATA controller Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 16/27] s390,time: revert direct ktime path for s390 clockevent device Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 10/27] x86/tls: Validate TLS entries to protect espfix Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 21/27] ipv4: fix nexthop attlen check in fib_nh_match Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 15/27] ext4: make orphan functions be no-op in no-journal mode Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 20/27] net: sctp: fix memory leak in auth key management Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 13/27] move d_rcu from overlapping d_child to overlapping d_alias Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 08/27] net: sctp: fix NULL pointer dereference in af->from_addr_param on malformed packet Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 22/27] tcp: md5: remove spinlock usage in fast path Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 06/27] i2c: davinci: generate STP always when NACK is received Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 05/27] ahci: disable MSI on SAMSUNG 0xa800 SSD Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 09/27] KVM: x86: Don't report guest userspace emulation error to userspace Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 17/27] drm: fix DRM_IOCTL_MODE_GETFB handle-leak Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 19/27] drivers/net, ipv6: Select IPv6 fragment idents for virtio UFO packets Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 23/27] tcp: md5: do not use alloc_percpu() Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 07/27] udf: Avoid infinite loop when processing indirect ICBs Ben Hutchings
2014-12-29  1:11 ` Ben Hutchings [this message]
2014-12-29  1:11 ` [PATCH 3.2 04/27] mm: fix swapoff hang after page migration and fork Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 01/27] drm/i915: Unlock panel even when LVDS is disabled Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 27/27] x86: kvm: use alternatives for VMCALL vs. VMMCALL if kernel text is read-only Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 18/27] crypto: ghash-clmulni-intel - use C implementation for setkey() Ben Hutchings
2014-12-29  1:11 ` [PATCH 3.2 26/27] net: sctp: use MAX_HEADER for headroom reserve in output path Ben Hutchings
2014-12-29  1:14 ` [PATCH 3.2 00/27] 3.2.66-rc1 review Ben Hutchings
2014-12-29  9:39 ` Guenter Roeck
2014-12-29 11:28   ` Ben Hutchings
2014-12-30  0:26 ` Satoru Takeuchi
2014-12-30  1:56   ` Ben Hutchings

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=lsq.1419815491.439721348@decadent.org.uk \
    --to=ben@decadent.org.uk \
    --cc=akpm@linux-foundation.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=stable@vger.kernel.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

all inboxes | Powered by JetHome®