Skip to contentSkip to conversation

ovl: suppress negative dentry in lookup

RFC PATCH Superseded Chengguang Xu 3 participants 3 messages May 7, 2020 – May 11, 2020 Overlay filesystem

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 ovl: suppress negative dentry in lookup

When a file is only in a lower layer, after lookup a negative dentry will be generated in the upper layer or even worse many negetive dentries will be generated in upper/lower layers. These negative dentries will be useless after construction of overlayfs' own dentry and may keep in the memory long time even after unmount of overlayfs instance. This patch tries to kill unnecessary negative dentry during lookup.

---
 fs/overlayfs/namei.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
fs/overlayfs/namei.c +10 -1
index 723d17744758..cf0ec4d7bcec 100644
@@ -200,7 +200,7 @@ static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
 	int err;
 	bool last_element = !post[0];
 
-	this = lookup_positive_unlocked(name, base, namelen);
+	this = lookup_one_len_unlocked(name, base, namelen);
 	if (IS_ERR(this)) {
 		err = PTR_ERR(this);
 		this = NULL;
@@ -209,6 +209,15 @@ static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
 		goto out_err;
 	}
 
+	/* Borrow the check from lookup_positive_unlocked */
+	if (d_flags_negative(smp_load_acquire(&this->d_flags))) {
+		d_drop(this);
+		dput(this);
+		this = NULL;
+		err = -ENOENT;
+		goto out;
+	}
+
 	if (ovl_dentry_weird(this)) {
 		/* Don't support traversing automounts and other weirdness */
 		err = -EREMOTE;
Signature · 1 line
2.20.1
maintainer
#1 lore raw
Quoted 2 lines · Chengguang Xu

When a file is only in a lower layer, after lookup a negative

Or in no layer at all...

Quoted 28 lines

dentry will be generated in the upper layer or even worse many negetive dentries will be generated in upper/lower layers. These negative dentries will be useless after construction of overlayfs' own dentry and may keep in the memory long time even after unmount of overlayfs instance. This patch tries to kill unnecessary negative dentry during lookup.

---
 fs/overlayfs/namei.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
fs/overlayfs/namei.c +1 -1
index 723d17744758..cf0ec4d7bcec 100644
@@ -200,7 +200,7 @@ static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
        int err;
        bool last_element = !post[0];
 
-       this = lookup_positive_unlocked(name, base, namelen);
+       this = lookup_one_len_unlocked(name, base, namelen);
        if (IS_ERR(this)) {
                err = PTR_ERR(this);
                this = NULL;
@@ -209,6 +209,15 @@ static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
                goto out_err;
        }

fs/overlayfs/namei.c static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d, · line 220

Quoted 10 lines
fs/overlayfs/namei.c +9 -0
 
+       /* Borrow the check from lookup_positive_unlocked */
+       if (d_flags_negative(smp_load_acquire(&this->d_flags))) {
+               d_drop(this);
+               dput(this);
+               this = NULL;
+               err = -ENOENT;
+               goto out;
+       }
+

This is a nice improvement, but my feeling is that this low level code belongs in a vfs helper with well documented semantics.

Thanks,
Amir.

maintainer
#2 lore raw
Quoted 36 lines · Amir Goldstein
Quoted 2 lines · Chengguang Xu

When a file is only in a lower layer, after lookup a negative

Or in no layer at all...

Quoted 29 lines

dentry will be generated in the upper layer or even worse many negetive dentries will be generated in upper/lower layers. These negative dentries will be useless after construction of overlayfs' own dentry and may keep in the memory long time even after unmount of overlayfs instance. This patch tries to kill unnecessary negative dentry during lookup.

---
 fs/overlayfs/namei.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
fs/overlayfs/namei.c +1 -1
index 723d17744758..cf0ec4d7bcec 100644
@@ -200,7 +200,7 @@ static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
        int err;
        bool last_element = !post[0];
 
-       this = lookup_positive_unlocked(name, base, namelen);
+       this = lookup_one_len_unlocked(name, base, namelen);
        if (IS_ERR(this)) {
                err = PTR_ERR(this);
                this = NULL;
@@ -209,6 +209,15 @@ static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
                goto out_err;
        }
 
Quoted 12 lines

fs/overlayfs/namei.c static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d, · line 220

Quoted 9 lines
fs/overlayfs/namei.c +9 -0
+       /* Borrow the check from lookup_positive_unlocked */
+       if (d_flags_negative(smp_load_acquire(&this->d_flags))) {
+               d_drop(this);
+               dput(this);
+               this = NULL;
+               err = -ENOENT;
+               goto out;
+       }
+

This is a nice improvement, but my feeling is that this low level code belongs in a vfs helper with well documented semantics.

I agree. Using d_drop() with the parent dir unlocked is not a good idea.

We need a new helper that does all of lookup_positive_unlocked() but with a conditional d_drop() after calling __lookup_slow(). In fact not dropping already cached negatives is probably a good idea, so doing it only in the slowpath should be the right thing.

Thanks,
Miklos