Skip to contentSkip to conversation

ceph: take reference to req->r_parent at point of assignment

PATCH In review Jeff Layton 2 participants 3 messages Jun 18, 2021 – Jun 21, 2021 Ceph client

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ceph: take reference to req->r_parent at point of assignment

Currently, we set the r_parent pointer but then don't take a reference to it until we submit the request. If we end up freeing the req before that point, then we'll do a iput when we shouldn't.

Instead, take the inode reference in the callers, so that it's always safe to call ceph_mdsc_put_request on the req, even before submission.

---
 fs/ceph/dir.c        | 9 +++++++++
 fs/ceph/export.c     | 1 +
 fs/ceph/file.c       | 1 +
 fs/ceph/mds_client.c | 1 -
 4 files changed, 11 insertions(+), 1 deletion(-)

Note that this isn't a problem with the existing code, because we never put the last reference before submission, but with the coming fscrypt patchset, we can end up doing this and this becomes a problem. With ths change, a set r_parent field means a reference *was* taken.

fs/ceph/dir.c +9 -0
index bd508b1aeac2..a656c5c00e65 100644
@@ -788,6 +788,7 @@ static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
 		mask |= CEPH_CAP_XATTR_SHARED;
 	req->r_args.getattr.mask = cpu_to_le32(mask);
 
+	ihold(dir);
 	req->r_parent = dir;
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	err = ceph_mdsc_do_request(mdsc, NULL, req);
@@ -861,6 +862,7 @@ static int ceph_mknod(struct user_namespace *mnt_userns, struct inode *dir,
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
 	req->r_parent = dir;
+	ihold(dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_args.mknod.mode = cpu_to_le32(mode);
 	req->r_args.mknod.rdev = cpu_to_le32(rdev);
@@ -922,6 +924,8 @@ static int ceph_symlink(struct user_namespace *mnt_userns, struct inode *dir,
 		goto out;
 	}
 	req->r_parent = dir;
+	ihold(dir);
+
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
@@ -986,6 +990,7 @@ static int ceph_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
 	req->r_parent = dir;
+	ihold(dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_args.mkdir.mode = cpu_to_le32(mode);
 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
@@ -1030,6 +1035,7 @@ static int ceph_link(struct dentry *old_dentry, struct inode *dir,
 	req->r_num_caps = 2;
 	req->r_old_dentry = dget(old_dentry);
 	req->r_parent = dir;
+	ihold(dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
@@ -1151,6 +1157,7 @@ static int ceph_unlink(struct inode *dir, struct dentry *dentry)
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
 	req->r_parent = dir;
+	ihold(dir);
 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 	req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
@@ -1225,6 +1232,7 @@ static int ceph_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
 	req->r_old_dentry = dget(old_dentry);
 	req->r_old_dentry_dir = old_dir;
 	req->r_parent = new_dir;
+	ihold(new_dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
 	req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
@@ -1721,6 +1729,7 @@ static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
 			req->r_dentry = dget(dentry);
 			req->r_num_caps = 2;
 			req->r_parent = dir;
+			ihold(dir);
 
 			mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
 			if (ceph_security_xattr_wanted(dir))
fs/ceph/export.c +1 -0
index 65540a4429b2..1d65934c1262 100644
@@ -542,6 +542,7 @@ static int ceph_get_name(struct dentry *parent, char *name,
 	ihold(inode);
 	req->r_ino2 = ceph_vino(d_inode(parent));
 	req->r_parent = d_inode(parent);
+	ihold(req->r_parent);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_num_caps = 2;
 	err = ceph_mdsc_do_request(mdsc, NULL, req);
fs/ceph/file.c +1 -0
index d3874c2df4b1..c8fd11cf4510 100644
@@ -706,6 +706,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 		mask |= CEPH_CAP_XATTR_SHARED;
 	req->r_args.open.mask = cpu_to_le32(mask);
 	req->r_parent = dir;
+	ihold(dir);
 
 	if (flags & O_CREAT) {
 		struct ceph_file_layout lo;
fs/ceph/mds_client.c +0 -1
index c03098c58be3..52ae5373437d 100644
@@ -2983,7 +2983,6 @@ int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
 		ceph_take_cap_refs(ci, CEPH_CAP_PIN, false);
 		__ceph_touch_fmode(ci, mdsc, fmode);
 		spin_unlock(&ci->i_ceph_lock);
-		ihold(req->r_parent);
 	}
 	if (req->r_old_dentry_dir)
 		ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
Signature · 1 line
2.31.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 25 lines · Jeff Layton

Currently, we set the r_parent pointer but then don't take a reference to it until we submit the request. If we end up freeing the req before that point, then we'll do a iput when we shouldn't.

Instead, take the inode reference in the callers, so that it's always safe to call ceph_mdsc_put_request on the req, even before submission.

---
 fs/ceph/dir.c        | 9 +++++++++
 fs/ceph/export.c     | 1 +
 fs/ceph/file.c       | 1 +
 fs/ceph/mds_client.c | 1 -
 4 files changed, 11 insertions(+), 1 deletion(-)

Note that this isn't a problem with the existing code, because we never put the last reference before submission, but with the coming fscrypt patchset, we can end up doing this and this becomes a problem. With ths change, a set r_parent field means a reference *was* taken.

fs/ceph/dir.c +0 -0
index bd508b1aeac2..a656c5c00e65 100644

fs/ceph/dir.c static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry, · line 792

Quoted 5 lines
fs/ceph/dir.c +1 -0
@@ -788,6 +788,7 @@ static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
 		mask |= CEPH_CAP_XATTR_SHARED;
 	req->r_args.getattr.mask = cpu_to_le32(mask);
 
+	ihold(dir);
 	req->r_parent = dir;

Other than my OCD saying that these two statements should be in the reverse order (or maybe all the other should...?), feel free to add my

Cheers,
--
Luís

Quoted 98 lines
fs/ceph/dir.c +8 -0
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	err = ceph_mdsc_do_request(mdsc, NULL, req);
@@ -861,6 +862,7 @@ static int ceph_mknod(struct user_namespace *mnt_userns, struct inode *dir,
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
 	req->r_parent = dir;
+	ihold(dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_args.mknod.mode = cpu_to_le32(mode);
 	req->r_args.mknod.rdev = cpu_to_le32(rdev);
@@ -922,6 +924,8 @@ static int ceph_symlink(struct user_namespace *mnt_userns, struct inode *dir,
 		goto out;
 	}
 	req->r_parent = dir;
+	ihold(dir);
+
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
@@ -986,6 +990,7 @@ static int ceph_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
 	req->r_parent = dir;
+	ihold(dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_args.mkdir.mode = cpu_to_le32(mode);
 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
@@ -1030,6 +1035,7 @@ static int ceph_link(struct dentry *old_dentry, struct inode *dir,
 	req->r_num_caps = 2;
 	req->r_old_dentry = dget(old_dentry);
 	req->r_parent = dir;
+	ihold(dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
@@ -1151,6 +1157,7 @@ static int ceph_unlink(struct inode *dir, struct dentry *dentry)
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
 	req->r_parent = dir;
+	ihold(dir);
 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 	req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
@@ -1225,6 +1232,7 @@ static int ceph_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
 	req->r_old_dentry = dget(old_dentry);
 	req->r_old_dentry_dir = old_dir;
 	req->r_parent = new_dir;
+	ihold(new_dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
 	req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
@@ -1721,6 +1729,7 @@ static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
 			req->r_dentry = dget(dentry);
 			req->r_num_caps = 2;
 			req->r_parent = dir;
+			ihold(dir);
 
 			mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
 			if (ceph_security_xattr_wanted(dir))
fs/ceph/export.c +1 -0
index 65540a4429b2..1d65934c1262 100644
@@ -542,6 +542,7 @@ static int ceph_get_name(struct dentry *parent, char *name,
 	ihold(inode);
 	req->r_ino2 = ceph_vino(d_inode(parent));
 	req->r_parent = d_inode(parent);
+	ihold(req->r_parent);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_num_caps = 2;
 	err = ceph_mdsc_do_request(mdsc, NULL, req);
fs/ceph/file.c +1 -0
index d3874c2df4b1..c8fd11cf4510 100644
@@ -706,6 +706,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 		mask |= CEPH_CAP_XATTR_SHARED;
 	req->r_args.open.mask = cpu_to_le32(mask);
 	req->r_parent = dir;
+	ihold(dir);
 
 	if (flags & O_CREAT) {
 		struct ceph_file_layout lo;
fs/ceph/mds_client.c +0 -1
index c03098c58be3..52ae5373437d 100644
@@ -2983,7 +2983,6 @@ int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
 		ceph_take_cap_refs(ci, CEPH_CAP_PIN, false);
 		__ceph_touch_fmode(ci, mdsc, fmode);
 		spin_unlock(&ci->i_ceph_lock);
-		ihold(req->r_parent);
 	}
 	if (req->r_old_dentry_dir)
 		ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
Signature · 1 line
2.31.1
#2 lore raw
Quoted 29 lines · Luis Henriques
Quoted 28 lines · Jeff Layton

Currently, we set the r_parent pointer but then don't take a reference to it until we submit the request. If we end up freeing the req before that point, then we'll do a iput when we shouldn't.

Instead, take the inode reference in the callers, so that it's always safe to call ceph_mdsc_put_request on the req, even before submission.

---
 fs/ceph/dir.c        | 9 +++++++++
 fs/ceph/export.c     | 1 +
 fs/ceph/file.c       | 1 +
 fs/ceph/mds_client.c | 1 -
 4 files changed, 11 insertions(+), 1 deletion(-)

Note that this isn't a problem with the existing code, because we never put the last reference before submission, but with the coming fscrypt patchset, we can end up doing this and this becomes a problem. With ths change, a set r_parent field means a reference *was* taken.

fs/ceph/dir.c +0 -0
index bd508b1aeac2..a656c5c00e65 100644
@@ -788,6 +788,7 @@ static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
 		mask |= CEPH_CAP_XATTR_SHARED;
 	req->r_args.getattr.mask = cpu_to_le32(mask);
 
Quoted 8 lines

fs/ceph/dir.c static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry, · line 792

Quoted 2 lines
fs/ceph/dir.c +1 -0
+	ihold(dir);
 	req->r_parent = dir;

Other than my OCD saying that these two statements should be in the reverse order (or maybe all the other should...?), feel free to add my

Thanks. I sort of wish ihold() returned the inode pointer to make this sort of thing more clear, but changing it now would probably be painful.

Quoted 99 lines
Quoted 98 lines
fs/ceph/dir.c +8 -0
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	err = ceph_mdsc_do_request(mdsc, NULL, req);
@@ -861,6 +862,7 @@ static int ceph_mknod(struct user_namespace *mnt_userns, struct inode *dir,
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
 	req->r_parent = dir;
+	ihold(dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_args.mknod.mode = cpu_to_le32(mode);
 	req->r_args.mknod.rdev = cpu_to_le32(rdev);
@@ -922,6 +924,8 @@ static int ceph_symlink(struct user_namespace *mnt_userns, struct inode *dir,
 		goto out;
 	}
 	req->r_parent = dir;
+	ihold(dir);
+
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
@@ -986,6 +990,7 @@ static int ceph_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
 	req->r_parent = dir;
+	ihold(dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_args.mkdir.mode = cpu_to_le32(mode);
 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
@@ -1030,6 +1035,7 @@ static int ceph_link(struct dentry *old_dentry, struct inode *dir,
 	req->r_num_caps = 2;
 	req->r_old_dentry = dget(old_dentry);
 	req->r_parent = dir;
+	ihold(dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
@@ -1151,6 +1157,7 @@ static int ceph_unlink(struct inode *dir, struct dentry *dentry)
 	req->r_dentry = dget(dentry);
 	req->r_num_caps = 2;
 	req->r_parent = dir;
+	ihold(dir);
 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 	req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
@@ -1225,6 +1232,7 @@ static int ceph_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
 	req->r_old_dentry = dget(old_dentry);
 	req->r_old_dentry_dir = old_dir;
 	req->r_parent = new_dir;
+	ihold(new_dir);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
 	req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
@@ -1721,6 +1729,7 @@ static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
 			req->r_dentry = dget(dentry);
 			req->r_num_caps = 2;
 			req->r_parent = dir;
+			ihold(dir);
 
 			mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
 			if (ceph_security_xattr_wanted(dir))
fs/ceph/export.c +1 -0
index 65540a4429b2..1d65934c1262 100644
@@ -542,6 +542,7 @@ static int ceph_get_name(struct dentry *parent, char *name,
 	ihold(inode);
 	req->r_ino2 = ceph_vino(d_inode(parent));
 	req->r_parent = d_inode(parent);
+	ihold(req->r_parent);
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_num_caps = 2;
 	err = ceph_mdsc_do_request(mdsc, NULL, req);
fs/ceph/file.c +1 -0
index d3874c2df4b1..c8fd11cf4510 100644
@@ -706,6 +706,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 		mask |= CEPH_CAP_XATTR_SHARED;
 	req->r_args.open.mask = cpu_to_le32(mask);
 	req->r_parent = dir;
+	ihold(dir);
 
 	if (flags & O_CREAT) {
 		struct ceph_file_layout lo;
fs/ceph/mds_client.c +0 -1
index c03098c58be3..52ae5373437d 100644
@@ -2983,7 +2983,6 @@ int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
 		ceph_take_cap_refs(ci, CEPH_CAP_PIN, false);
 		__ceph_touch_fmode(ci, mdsc, fmode);
 		spin_unlock(&ci->i_ceph_lock);
-		ihold(req->r_parent);
 	}
 	if (req->r_old_dentry_dir)
 		ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
Signature · 1 line
2.31.1
Signature · 1 line
Jeff Layton <[email protected]>