[RESEND PATCH] ceph: don't truncate file in atomic_open

[RESEND PATCH] ceph: don't truncate file in atomic_open

[Hu Weiwen]

Clear O_TRUNC from the flags sent in the MDS create request.

`atomic_open' is called before permission check. We should not do any
modification to the file here. The caller will do the truncation
afterward.

Fixes: 124e68e74099 ("ceph: file operations")
Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
---
rebased onto ceph_client repo testing branch

 fs/ceph/file.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index 296fd1c7ece8..289e66e9cbb0 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -745,6 +745,11 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	err = ceph_wait_on_conflict_unlink(dentry);
 	if (err)
 		return err;
+	/*
+	 * Do not truncate the file, since atomic_open is called before the
+	 * permission check. The caller will do the truncation afterward.
+	 */
+	flags &= ~O_TRUNC;
 
 retry:
 	if (flags & O_CREAT) {
@@ -836,9 +841,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 	req->r_new_inode = new_inode;
 	new_inode = NULL;
-	err = ceph_mdsc_do_request(mdsc,
-				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
-				   req);
+	err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
 	if (err == -ENOENT) {
 		dentry = ceph_handle_snapdir(req, dentry);
 		if (IS_ERR(dentry)) {
-- 
2.25.1

Re: [RESEND PATCH] ceph: don't truncate file in atomic_open

[Xiubo Li]

On 7/1/22 10:52 AM, Hu Weiwen wrote:
> Clear O_TRUNC from the flags sent in the MDS create request.
>
> `atomic_open' is called before permission check. We should not do any
> modification to the file here. The caller will do the truncation
> afterward.
>
> Fixes: 124e68e74099 ("ceph: file operations")
> Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
> ---
> rebased onto ceph_client repo testing branch
>
>   fs/ceph/file.c | 9 ++++++---
>   1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/fs/ceph/file.c b/fs/ceph/file.c
> index 296fd1c7ece8..289e66e9cbb0 100644
> --- a/fs/ceph/file.c
> +++ b/fs/ceph/file.c
> @@ -745,6 +745,11 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
>   	err = ceph_wait_on_conflict_unlink(dentry);
>   	if (err)
>   		return err;
> +	/*
> +	 * Do not truncate the file, since atomic_open is called before the
> +	 * permission check. The caller will do the truncation afterward.
> +	 */
> +	flags &= ~O_TRUNC;
>   
>   retry:
>   	if (flags & O_CREAT) {
> @@ -836,9 +841,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
>   	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
>   	req->r_new_inode = new_inode;
>   	new_inode = NULL;
> -	err = ceph_mdsc_do_request(mdsc,
> -				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
> -				   req);
> +	err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
>   	if (err == -ENOENT) {
>   		dentry = ceph_handle_snapdir(req, dentry);
>   		if (IS_ERR(dentry)) {

Merged it and thanks!

-- Xiubo

ceph: don't truncate file in atomic_open

[Xiubo Li]

Hi Maitainers

This patch is a fix in kceph module and should be backported to any 
affected stable old kernels. And the original patch missed tagging 
stable and got merged already months ago:

commit 7cb9994754f8a36ae9e5ec4597c5c4c2d6c03832
Author: Hu Weiwen <sehuww@mail.scut.edu.cn>
Date:   Fri Jul 1 10:52:27 2022 +0800

     ceph: don't truncate file in atomic_open

     Clear O_TRUNC from the flags sent in the MDS create request.

     `atomic_open' is called before permission check. We should not do any
     modification to the file here. The caller will do the truncation
     afterward.

     Fixes: 124e68e74099 ("ceph: file operations")
     Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
     Reviewed-by: Xiubo Li <xiubli@redhat.com>
     Signed-off-by: Ilya Dryomov <idryomov@gmail.com>


Just a single patch.

I am not very sure this is the correct way to do this, if anything else 
I need to do to backport this to old kernels please let me know.

Thanks!

- Xiubo



On 01/07/2022 10:52, Hu Weiwen wrote:
> Clear O_TRUNC from the flags sent in the MDS create request.
>
> `atomic_open' is called before permission check. We should not do any
> modification to the file here. The caller will do the truncation
> afterward.
>
> Fixes: 124e68e74099 ("ceph: file operations")
> Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
> ---
> rebased onto ceph_client repo testing branch
>
>   fs/ceph/file.c | 9 ++++++---
>   1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/fs/ceph/file.c b/fs/ceph/file.c
> index 296fd1c7ece8..289e66e9cbb0 100644
> --- a/fs/ceph/file.c
> +++ b/fs/ceph/file.c
> @@ -745,6 +745,11 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
>   	err = ceph_wait_on_conflict_unlink(dentry);
>   	if (err)
>   		return err;
> +	/*
> +	 * Do not truncate the file, since atomic_open is called before the
> +	 * permission check. The caller will do the truncation afterward.
> +	 */
> +	flags &= ~O_TRUNC;
>   
>   retry:
>   	if (flags & O_CREAT) {
> @@ -836,9 +841,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
>   	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
>   	req->r_new_inode = new_inode;
>   	new_inode = NULL;
> -	err = ceph_mdsc_do_request(mdsc,
> -				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
> -				   req);
> +	err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
>   	if (err == -ENOENT) {
>   		dentry = ceph_handle_snapdir(req, dentry);
>   		if (IS_ERR(dentry)) {

Re: ceph: don't truncate file in atomic_open

[Xiubo Li]

[-- Attachment #1: Type: text/plain, Size: 3240 bytes --]

Hi Maintainers:

This is one very important bug fixing patch needed to be backported for 
ceph.

I have attached the patches based one the latest branches[1] of 4.9.y, 
4.14.y, 4.19.y, 5.4.y, 5.10.y, 5.15.y and 5.19.y.

This patch is simple and the conflict is trivial.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git

Thanks!

- Xiubo



On 12/10/2022 09:51, Xiubo Li wrote:
> Hi Maitainers
>
> This patch is a fix in kceph module and should be backported to any 
> affected stable old kernels. And the original patch missed tagging 
> stable and got merged already months ago:
>
> commit 7cb9994754f8a36ae9e5ec4597c5c4c2d6c03832
> Author: Hu Weiwen <sehuww@mail.scut.edu.cn>
> Date:   Fri Jul 1 10:52:27 2022 +0800
>
>     ceph: don't truncate file in atomic_open
>
>     Clear O_TRUNC from the flags sent in the MDS create request.
>
>     `atomic_open' is called before permission check. We should not do any
>     modification to the file here. The caller will do the truncation
>     afterward.
>
>     Fixes: 124e68e74099 ("ceph: file operations")
>     Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
>     Reviewed-by: Xiubo Li <xiubli@redhat.com>
>     Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
>
>
> Just a single patch.
>
> I am not very sure this is the correct way to do this, if anything 
> else I need to do to backport this to old kernels please let me know.
>
> Thanks!
>
> - Xiubo
>
>
>
> On 01/07/2022 10:52, Hu Weiwen wrote:
>> Clear O_TRUNC from the flags sent in the MDS create request.
>>
>> `atomic_open' is called before permission check. We should not do any
>> modification to the file here. The caller will do the truncation
>> afterward.
>>
>> Fixes: 124e68e74099 ("ceph: file operations")
>> Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
>> ---
>> rebased onto ceph_client repo testing branch
>>
>>   fs/ceph/file.c | 9 ++++++---
>>   1 file changed, 6 insertions(+), 3 deletions(-)
>>
>> diff --git a/fs/ceph/file.c b/fs/ceph/file.c
>> index 296fd1c7ece8..289e66e9cbb0 100644
>> --- a/fs/ceph/file.c
>> +++ b/fs/ceph/file.c
>> @@ -745,6 +745,11 @@ int ceph_atomic_open(struct inode *dir, struct 
>> dentry *dentry,
>>       err = ceph_wait_on_conflict_unlink(dentry);
>>       if (err)
>>           return err;
>> +    /*
>> +     * Do not truncate the file, since atomic_open is called before the
>> +     * permission check. The caller will do the truncation afterward.
>> +     */
>> +    flags &= ~O_TRUNC;
>>     retry:
>>       if (flags & O_CREAT) {
>> @@ -836,9 +841,7 @@ int ceph_atomic_open(struct inode *dir, struct 
>> dentry *dentry,
>>       set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
>>       req->r_new_inode = new_inode;
>>       new_inode = NULL;
>> -    err = ceph_mdsc_do_request(mdsc,
>> -                   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
>> -                   req);
>> +    err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, 
>> req);
>>       if (err == -ENOENT) {
>>           dentry = ceph_handle_snapdir(req, dentry);
>>           if (IS_ERR(dentry)) {

[-- Attachment #2: 4.9.patch --]
[-- Type: text/x-patch, Size: 1824 bytes --]

From 54677a489cce48b6477042f7e843c9c45af475a3 Mon Sep 17 00:00:00 2001
From: Xiubo Li <xiubli@redhat.com>
Date: Thu, 13 Oct 2022 10:33:07 +0800
Subject: [PATCH] ceph: don't truncate file in atomic_open

[ Upstream commit 7cb9994754f8a36ae9e5ec4597c5c4c2d6c03832]

Clear O_TRUNC from the flags sent in the MDS create request.

`atomic_open' is called before permission check. We should not do any
modification to the file here. The caller will do the truncation
afterward.

Fixes: 124e68e74099 ("ceph: file operations")
Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
[Xiubo: fixed a trivial conflict for 4.9 backport]
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
 fs/ceph/file.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index e818344a052c..e6bf3cd8fae0 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -354,6 +354,11 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	err = ceph_init_dentry(dentry);
 	if (err < 0)
 		return err;
+	/*
+	 * Do not truncate the file, since atomic_open is called before the
+	 * permission check. The caller will do the truncation afterward.
+	 */
+	flags &= ~O_TRUNC;
 
 	if (flags & O_CREAT) {
 		err = ceph_pre_init_acls(dir, &mode, &acls);
@@ -384,9 +389,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
        req->r_args.open.mask = cpu_to_le32(mask);
 
 	req->r_locked_dir = dir;           /* caller holds dir->i_mutex */
-	err = ceph_mdsc_do_request(mdsc,
-				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
-				   req);
+	err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
 	err = ceph_handle_snapdir(req, dentry, err);
 	if (err)
 		goto out_req;
-- 
2.31.1


[-- Attachment #3: 4.14.patch --]
[-- Type: text/x-patch, Size: 1802 bytes --]

From e435a506cd5846839c3981316752d13a15662d94 Mon Sep 17 00:00:00 2001
From: Xiubo Li <xiubli@redhat.com>
Date: Thu, 13 Oct 2022 10:58:51 +0800
Subject: [PATCH] ceph: don't truncate file in atomic_open

[ Upstream commit 7cb9994754f8a36ae9e5ec4597c5c4c2d6c03832]

Clear O_TRUNC from the flags sent in the MDS create request.

`atomic_open' is called before permission check. We should not do any
modification to the file here. The caller will do the truncation
afterward.

Fixes: 124e68e74099 ("ceph: file operations")
Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
[Xiubo: fixed a trivial conflict for 4.14 backport]
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
 fs/ceph/file.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index 1f873034f469..ddfa6ce3a0fb 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -381,6 +381,12 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	if (dentry->d_name.len > NAME_MAX)
 		return -ENAMETOOLONG;
 
+	/*
+	 * Do not truncate the file, since atomic_open is called before the
+	 * permission check. The caller will do the truncation afterward.
+	 */
+	flags &= ~O_TRUNC;
+
 	if (flags & O_CREAT) {
 		err = ceph_pre_init_acls(dir, &mode, &acls);
 		if (err < 0)
@@ -411,9 +417,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 
 	req->r_parent = dir;
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
-	err = ceph_mdsc_do_request(mdsc,
-				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
-				   req);
+	err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
 	err = ceph_handle_snapdir(req, dentry, err);
 	if (err)
 		goto out_req;
-- 
2.31.1


[-- Attachment #4: 4.19.patch --]
[-- Type: text/x-patch, Size: 1804 bytes --]

From 7846c7a8083c6355dbf0713811f2e405791b315b Mon Sep 17 00:00:00 2001
From: Xiubo Li <xiubli@redhat.com>
Date: Thu, 13 Oct 2022 11:04:52 +0800
Subject: [PATCH] ceph: don't truncate file in atomic_open

[ Upstream commit 7cb9994754f8a36ae9e5ec4597c5c4c2d6c03832]

Clear O_TRUNC from the flags sent in the MDS create request.

`atomic_open' is called before permission check. We should not do any
modification to the file here. The caller will do the truncation
afterward.

Fixes: 124e68e74099 ("ceph: file operations")
Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
[Xiubo: fixed a trivial conflict for 4.19 backport]
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
 fs/ceph/file.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index 4ce2752c8b71..95d7906fb9ea 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -446,6 +446,12 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	if (dentry->d_name.len > NAME_MAX)
 		return -ENAMETOOLONG;
 
+	/*
+	 * Do not truncate the file, since atomic_open is called before the
+	 * permission check. The caller will do the truncation afterward.
+	 */
+	flags &= ~O_TRUNC;
+
 	if (flags & O_CREAT) {
 		if (ceph_quota_is_max_files_exceeded(dir))
 			return -EDQUOT;
@@ -478,9 +484,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 
 	req->r_parent = dir;
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
-	err = ceph_mdsc_do_request(mdsc,
-				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
-				   req);
+	err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
 	err = ceph_handle_snapdir(req, dentry, err);
 	if (err)
 		goto out_req;
-- 
2.31.1


[-- Attachment #5: 5.4.patch --]
[-- Type: text/x-patch, Size: 1803 bytes --]

From ebce1c991e59c57684782c9e506bc474918ee882 Mon Sep 17 00:00:00 2001
From: Xiubo Li <xiubli@redhat.com>
Date: Thu, 13 Oct 2022 11:08:15 +0800
Subject: [PATCH] ceph: don't truncate file in atomic_open

[ Upstream commit 7cb9994754f8a36ae9e5ec4597c5c4c2d6c03832]

Clear O_TRUNC from the flags sent in the MDS create request.

`atomic_open' is called before permission check. We should not do any
modification to the file here. The caller will do the truncation
afterward.

Fixes: 124e68e74099 ("ceph: file operations")
Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
[Xiubo: fixed a trivial conflict for 5.4 backport]
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
 fs/ceph/file.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index aa1eac6d89f2..83122fc5f813 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -452,6 +452,12 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	if (dentry->d_name.len > NAME_MAX)
 		return -ENAMETOOLONG;
 
+	/*
+	 * Do not truncate the file, since atomic_open is called before the
+	 * permission check. The caller will do the truncation afterward.
+	 */
+	flags &= ~O_TRUNC;
+
 	if (flags & O_CREAT) {
 		if (ceph_quota_is_max_files_exceeded(dir))
 			return -EDQUOT;
@@ -490,9 +496,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 
 	req->r_parent = dir;
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
-	err = ceph_mdsc_do_request(mdsc,
-				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
-				   req);
+	err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
 	err = ceph_handle_snapdir(req, dentry, err);
 	if (err)
 		goto out_req;
-- 
2.31.1


[-- Attachment #6: 5.10.patch --]
[-- Type: text/x-patch, Size: 1785 bytes --]

From db49cff7e45b4884657e9932f069ce9ad7d63978 Mon Sep 17 00:00:00 2001
From: Xiubo Li <xiubli@redhat.com>
Date: Thu, 13 Oct 2022 10:51:19 +0800
Subject: [PATCH] ceph: don't truncate file in atomic_open

[ Upstream commit 7cb9994754f8a36ae9e5ec4597c5c4c2d6c03832]

Clear O_TRUNC from the flags sent in the MDS create request.

`atomic_open' is called before permission check. We should not do any
modification to the file here. The caller will do the truncation
afterward.

Fixes: 124e68e74099 ("ceph: file operations")
Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
[Xiubo: fixed a trivial conflict for 5.10 backport]
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
 fs/ceph/file.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index 93d986856f1c..943655e36a79 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -703,6 +703,12 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	if (dentry->d_name.len > NAME_MAX)
 		return -ENAMETOOLONG;
 
+	/*
+	 * Do not truncate the file, since atomic_open is called before the
+	 * permission check. The caller will do the truncation afterward.
+	 */
+	flags &= ~O_TRUNC;
+
 	if (flags & O_CREAT) {
 		if (ceph_quota_is_max_files_exceeded(dir))
 			return -EDQUOT;
@@ -769,9 +775,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	}
 
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
-	err = ceph_mdsc_do_request(mdsc,
-				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
-				   req);
+	err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
 	err = ceph_handle_snapdir(req, dentry, err);
 	if (err)
 		goto out_req;
-- 
2.31.1


[-- Attachment #7: 5.15.patch --]
[-- Type: text/x-patch, Size: 1805 bytes --]

From 3524a8ab33573375a14a3d5608501598ad26e007 Mon Sep 17 00:00:00 2001
From: Xiubo Li <xiubli@redhat.com>
Date: Thu, 13 Oct 2022 11:12:23 +0800
Subject: [PATCH] ceph: don't truncate file in atomic_open

[ Upstream commit 7cb9994754f8a36ae9e5ec4597c5c4c2d6c03832]

Clear O_TRUNC from the flags sent in the MDS create request.

`atomic_open' is called before permission check. We should not do any
modification to the file here. The caller will do the truncation
afterward.

Fixes: 124e68e74099 ("ceph: file operations")
Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
[Xiubo: fixed a trivial conflict for 5.15 backport]
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
 fs/ceph/file.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index e34d52df4a13..53bffda3c76c 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -703,6 +703,12 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	if (dentry->d_name.len > NAME_MAX)
 		return -ENAMETOOLONG;
 
+	/*
+	 * Do not truncate the file, since atomic_open is called before the
+	 * permission check. The caller will do the truncation afterward.
+	 */
+	flags &= ~O_TRUNC;
+
 	if (flags & O_CREAT) {
 		if (ceph_quota_is_max_files_exceeded(dir))
 			return -EDQUOT;
@@ -770,9 +776,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	}
 
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
-	err = ceph_mdsc_do_request(mdsc,
-				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
-				   req);
+	err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
 	if (err == -ENOENT) {
 		dentry = ceph_handle_snapdir(req, dentry);
 		if (IS_ERR(dentry)) {
-- 
2.31.1


[-- Attachment #8: 5.19.patch --]
[-- Type: text/x-patch, Size: 1805 bytes --]

From ea2af51ef3c67bbf34855cd4d5c95437600047e5 Mon Sep 17 00:00:00 2001
From: Xiubo Li <xiubli@redhat.com>
Date: Thu, 13 Oct 2022 11:15:18 +0800
Subject: [PATCH] ceph: don't truncate file in atomic_open

[ Upstream commit 7cb9994754f8a36ae9e5ec4597c5c4c2d6c03832]

Clear O_TRUNC from the flags sent in the MDS create request.

`atomic_open' is called before permission check. We should not do any
modification to the file here. The caller will do the truncation
afterward.

Fixes: 124e68e74099 ("ceph: file operations")
Signed-off-by: Hu Weiwen <sehuww@mail.scut.edu.cn>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
[Xiubo: fixed a trivial conflict for 5.19 backport]
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
 fs/ceph/file.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index da59e836a06e..a43e40138a3b 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -740,6 +740,12 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	if (dentry->d_name.len > NAME_MAX)
 		return -ENAMETOOLONG;
 
+	/*
+	 * Do not truncate the file, since atomic_open is called before the
+	 * permission check. The caller will do the truncation afterward.
+	 */
+	flags &= ~O_TRUNC;
+
 	if (flags & O_CREAT) {
 		if (ceph_quota_is_max_files_exceeded(dir))
 			return -EDQUOT;
@@ -807,9 +813,7 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
 	}
 
 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
-	err = ceph_mdsc_do_request(mdsc,
-				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
-				   req);
+	err = ceph_mdsc_do_request(mdsc, (flags & O_CREAT) ? dir : NULL, req);
 	if (err == -ENOENT) {
 		dentry = ceph_handle_snapdir(req, dentry);
 		if (IS_ERR(dentry)) {
-- 
2.31.1

Re: ceph: don't truncate file in atomic_open

[Greg KH]

On Wed, Oct 12, 2022 at 09:51:43AM +0800, Xiubo Li wrote:
> Hi Maitainers
> 
> This patch is a fix in kceph module and should be backported to any affected
> stable old kernels. And the original patch missed tagging stable and got
> merged already months ago:

Does not apply to the trees, sorry.  Please provide a working backport
if you wish to see it applied.

thanks,

greg k-h

Re: ceph: don't truncate file in atomic_open

[Greg KH]

On Thu, Oct 13, 2022 at 11:30:06AM +0800, Xiubo Li wrote:
> Hi Maintainers:
> 
> This is one very important bug fixing patch needed to be backported for
> ceph.
> 
> I have attached the patches based one the latest branches[1] of 4.9.y,
> 4.14.y, 4.19.y, 5.4.y, 5.10.y, 5.15.y and 5.19.y.
> 
> This patch is simple and the conflict is trivial.
> 
> [1] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git

Ah, missed this, thanks, now queued up.

greg k-h

Re: ceph: don't truncate file in atomic_open

[Xiubo Li]

Hi Greg,

On 13/10/2022 17:12, Greg KH wrote:
> On Wed, Oct 12, 2022 at 09:51:43AM +0800, Xiubo Li wrote:
>> Hi Maitainers
>>
>> This patch is a fix in kceph module and should be backported to any affected
>> stable old kernels. And the original patch missed tagging stable and got
>> merged already months ago:
> Does not apply to the trees, sorry.  Please provide a working backport
> if you wish to see it applied.

Already sent it out and I saw your have applied to them.

Thanks very much!

- Xiubo


> thanks,
>
> greg k-h
>