Skip to contentSkip to conversation

smb: fix xfstests generic/035

PATCH Superseded ChenXiaoSong 3 participants 11 messages Aug 4 – Aug 5 1 message not in the archive

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

ChenXiaoSong

author reviewer
#0 lore raw

v2 cover smb: fix xfstests generic/035

From: ChenXiaoSong <[email protected]>

v1->v2:
- Add patch #01 #03.
- Patch #02: introduce helper ksmbd_has_other_nonposix_open().

v1: https://lore.kernel.org/linux-cifs/[email protected]/

ChenXiaoSong (3):
smb/server: rename to ksmbd_has_nonposix_open_child() smb/server: deny overwriting targets with non-POSIX opens smb/client: fix nlink of an overwritten open file

 fs/smb/client/inode.c     | 11 ++++++-----
 fs/smb/server/vfs.c       | 15 ++++++++++++++-
 fs/smb/server/vfs_cache.c | 34 +++++++++++++++++++++++++++++++++-
 fs/smb/server/vfs_cache.h |  3 ++-
 4 files changed, 55 insertions(+), 8 deletions(-)
Signature · 1 line
2.54.0
author reviewer
#1 lore raw

PATCH v2 1/3 smb/server: rename to ksmbd_has_nonposix_open_child()

From: ChenXiaoSong <[email protected]>

The original function name ksmbd_has_open_files() could be confused with the function name introduced in the next patch, and it does not accurately describe what this function does.

---
 fs/smb/server/vfs.c       | 2 +-
 fs/smb/server/vfs_cache.c | 2 +-
 fs/smb/server/vfs_cache.h | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)
fs/smb/server/vfs.c +1 -1
index 28940c7b5f83..16ef8d051b18 100644
@@ -721,7 +721,7 @@ int ksmbd_vfs_rename(struct ksmbd_work *work, struct ksmbd_file *old_fp,
 	if (err)
 		goto out_drop_write;
 
-	if (d_is_dir(old_child) && ksmbd_has_open_files(old_fp)) {
+	if (d_is_dir(old_child) && ksmbd_has_nonposix_open_child(old_fp)) {
 		err = -EACCES;
 		goto out3;
 	}
fs/smb/server/vfs_cache.c +1 -1
index eac9886eb6e3..028bc1b0f652 100644
@@ -1137,7 +1137,7 @@ struct ksmbd_file *ksmbd_lookup_fd_inode(struct dentry *dentry)
 	return NULL;
 }
 
-bool ksmbd_has_open_files(struct ksmbd_file *old_fp)
+bool ksmbd_has_nonposix_open_child(struct ksmbd_file *old_fp)
 {
 	struct dentry *dentry = old_fp->filp->f_path.dentry;
 	struct ksmbd_file *fp;
fs/smb/server/vfs_cache.h +1 -1
index 127ea4987e3f..9dca617bc429 100644
@@ -212,7 +212,7 @@ bool ksmbd_has_stream_without_delete_share(struct ksmbd_file *fp);
 int ksmbd_close_fd_app_instance_id(char *app_instance_id);
 struct ksmbd_file *ksmbd_lookup_fd_cguid(char *cguid);
 struct ksmbd_file *ksmbd_lookup_fd_inode(struct dentry *dentry);
-bool ksmbd_has_open_files(struct ksmbd_file *old_fp);
+bool ksmbd_has_nonposix_open_child(struct ksmbd_file *old_fp);
 unsigned int ksmbd_open_durable_fd(struct ksmbd_file *fp);
 struct ksmbd_file *ksmbd_open_fd(struct ksmbd_work *work, struct file *filp);
 void ksmbd_launch_ksmbd_durable_scavenger(void);
Signature · 1 line
2.54.0
author reviewer
#2 lore raw

PATCH v2 2/3 smb/server: deny overwriting targets with non-POSIX opens

In reply to ChenXiaoSong

From: ChenXiaoSong <[email protected]>

Reproducer:

1. server: systemctl start ksmbd
2. client: mount without posix option

     mount -t cifs //${server_ip}/export /mnt

3. client: touch /mnt/file1 /mnt/file2
4. client: C program: int fd = open("/mnt/file2", O_RDONLY);
5. client: C program: rename("/mnt/file1", "/mnt/file2");
6. client: C program: struct stat stbuf; fstat(fd, &stbuf); stbuf.st_nlink is 1, should be 0

This patch fixes xfstests generic/035 when mounted without posix option.

---
 fs/smb/server/vfs.c       | 13 +++++++++++++
 fs/smb/server/vfs_cache.c | 32 ++++++++++++++++++++++++++++++++
 fs/smb/server/vfs_cache.h |  1 +
 3 files changed, 46 insertions(+)
fs/smb/server/vfs.c +13 -0
index 16ef8d051b18..9102ab47a3aa 100644
@@ -726,6 +726,19 @@ int ksmbd_vfs_rename(struct ksmbd_work *work, struct ksmbd_file *old_fp,
 		goto out3;
 	}
 
+	/*
+	 * See MS-FSA 2.1.5.15.12.
+	 * An overwrite rename must fail with STATUS_ACCESS_DENIED if the
+	 * existing target still has a non-POSIX open.
+	 */
+	if (!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)) &&
+	    d_inode(rd.new_dentry) &&
+	    d_inode(rd.new_dentry) != d_inode(old_child) &&
+	    ksmbd_has_other_nonposix_open(rd.new_dentry)) {
+		err = -EACCES;
+		goto out3;
+	}
+
 	err = ksmbd_vfs_check_rename_share(work, old_path);
 	if (err)
 		goto out3;
fs/smb/server/vfs_cache.c +32 -0
index 028bc1b0f652..94b2b5e48d3d 100644
@@ -1137,6 +1137,38 @@ struct ksmbd_file *ksmbd_lookup_fd_inode(struct dentry *dentry)
 	return NULL;
 }
 
+bool ksmbd_has_other_nonposix_open(struct dentry *dentry)
+{
+	struct ksmbd_file *fp;
+	struct ksmbd_inode *ci;
+	struct inode *inode = d_inode(dentry);
+	bool ret = false;
+
+	if (!inode)
+		return false;
+
+	ci = ksmbd_inode_lookup_lock(dentry);
+	if (!ci)
+		return false;
+
+	down_read(&ci->m_lock);
+	list_for_each_entry(fp, &ci->m_fp_list, node) {
+		if (READ_ONCE(fp->f_state) != FP_INITED)
+			continue;
+		if (inode != file_inode(fp->filp))
+			continue;
+		if (fp->is_posix_ctxt)
+			continue;
+
+		ret = true;
+		break;
+	}
+	up_read(&ci->m_lock);
+	ksmbd_inode_put(ci);
+
+	return ret;
+}
+
 bool ksmbd_has_nonposix_open_child(struct ksmbd_file *old_fp)
 {
 	struct dentry *dentry = old_fp->filp->f_path.dentry;
fs/smb/server/vfs_cache.h +1 -0
index 9dca617bc429..5cac022b540b 100644
@@ -212,6 +212,7 @@ bool ksmbd_has_stream_without_delete_share(struct ksmbd_file *fp);
 int ksmbd_close_fd_app_instance_id(char *app_instance_id);
 struct ksmbd_file *ksmbd_lookup_fd_cguid(char *cguid);
 struct ksmbd_file *ksmbd_lookup_fd_inode(struct dentry *dentry);
+bool ksmbd_has_other_nonposix_open(struct dentry *dentry);
 bool ksmbd_has_nonposix_open_child(struct ksmbd_file *old_fp);
 unsigned int ksmbd_open_durable_fd(struct ksmbd_file *fp);
 struct ksmbd_file *ksmbd_open_fd(struct ksmbd_work *work, struct file *filp);
Signature · 1 line
2.54.0
maintainer
#3 lore raw

Re: [PATCH v2 2/3] smb/server: deny overwriting targets with non-POSIX opens

fs/smb/server/vfs_cache.c struct ksmbd_file *ksmbd_lookup_fd_inode(struct dentry *dentry) · line 1150

Quoted 11 lines
fs/smb/server/vfs_cache.c +11 -0
+bool ksmbd_has_other_nonposix_open(struct dentry *dentry)
+{
+       struct ksmbd_file *fp;
+       struct ksmbd_inode *ci;
+       struct inode *inode = d_inode(dentry);
+       bool ret = false;
+
+       if (!inode)
+               return false;
+
+       ci = ksmbd_inode_lookup_lock(dentry);

The new check is limited to the target dentry's ksmbd_inode->m_fp_list. Therefore, it does not cover a struct ksmbd_file opened through another hardlink dentry, even though file_inode(fp->filp) == d_inode(dentry). For example, if file2.link is a hardlink to file2 and file2.link is opened with a non-POSIX context, an overwrite rename of another file to file2 can miss that open handle.
This appears to be an existing limitation rather than a regression introduced by this patch. However, since this patch is intended to reject overwrites when the target has a non-POSIX ksmbd_file...

Quoted 17 lines
fs/smb/server/vfs_cache.c +17 -0
+       if (!ci)
+               return false;
+
+       down_read(&ci->m_lock);
+       list_for_each_entry(fp, &ci->m_fp_list, node) {
+               if (READ_ONCE(fp->f_state) != FP_INITED)
+                       continue;
+               if (inode != file_inode(fp->filp))
+                       continue;
+               if (fp->is_posix_ctxt)
+                       continue;
+
+               ret = true;
+               break;
+       }
+       up_read(&ci->m_lock);
+       ksmbd_inode_put(ci);
author reviewer
#4 lore raw

Re: [PATCH v2 2/3] smb/server: deny overwriting targets with non-POSIX opens

Got it, I will add it to my todo list, and submit a separate patch to fix this pre-existing issue.

Quoted 2 lines · Namjae Jeon

This appears to be an existing limitation rather than a regression introduced by this patch

Signature · 3 lines
ChenXiaoSong <[email protected]>
Chinese Homepage: https://chenxiaosong.com
English Homepage: https://chenxiaosong.com/en

A message that is not in the archive (sent off-list or not yet received): [email protected]

maintainer
#6 lore raw

Re: [PATCH v2 2/3] smb/server: deny overwriting targets with non-POSIX opens

Quoted 4 lines · ChenXiaoSong

Hi Namjae,

What do you think about iterating over global_ft.idr instead?

Looks ok. I will apply after running my tests tonight. Thanks!

Quoted 27 lines
```
ksmbd_has_other_nonposix_open() {
        read_lock(&global_ft.lock);
        idr_for_each_entry(global_ft.idr, fp, id) {
        }
        read_unlock(&global_ft.lock);
}
```
Quoted 10 lines · Namjae Jeon

The new check is limited to the target dentry's ksmbd_inode->m_fp_list. Therefore, it does not cover a struct ksmbd_file opened through another hardlink dentry, even though file_inode(fp->filp) == d_inode(dentry). For example, if file2.link is a hardlink to file2 and file2.link is opened with a non-POSIX context, an overwrite rename of another file to file2 can miss that open handle.
This appears to be an existing limitation rather than a regression introduced by this patch. However, since this patch is intended to reject overwrites when the target has a non-POSIX ksmbd_file...

--
ChenXiaoSong <[email protected]>
Chinese Homepage: https://chenxiaosong.com
English Homepage: https://chenxiaosong.com/en

author reviewer
#8 lore raw

PATCH v2 3/3 smb/client: fix nlink of an overwritten open file

In reply to ChenXiaoSong

From: ChenXiaoSong <[email protected]>

Reproducer:

1. server: systemctl start ksmbd
2. client: mount with posix option

     mount -t cifs -o posix //${server_ip}/export /mnt

3. client: touch /mnt/file1 /mnt/file2
4. client: C program: int fd = open("/mnt/file2", O_RDONLY);
5. client: C program: rename("/mnt/file1", "/mnt/file2");
6. client: C program: struct stat stbuf; fstat(fd, &stbuf); stbuf.st_nlink is 1, should be 0

This patch fixes xfstests generic/035 when mounted with posix option.

---
 fs/smb/client/inode.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
fs/smb/client/inode.c +6 -5
index 0afff761aab9..0b6273ce01a8 100644
@@ -2648,11 +2648,8 @@ cifs_rename2(struct mnt_idmap *idmap, struct inode *source_dir,
 	if (d_really_is_positive(target_dentry)) {
 		if (!rc) {
 			struct inode *inode = d_inode(target_dentry);
-			/*
-			 * Samba and ksmbd servers allow renaming a target
-			 * directory that is open, so make sure to update
-			 * ->i_nlink and then mark it as delete pending.
-			 */
+
+			/* Update the target link count after rename. */
 			if (S_ISDIR(inode->i_mode)) {
 				drop_cached_dir_by_name(xid, tcon, to_name, cifs_sb);
 				spin_lock(&inode->i_lock);
@@ -2663,6 +2660,10 @@ cifs_rename2(struct mnt_idmap *idmap, struct inode *source_dir,
 				CIFS_I(inode)->time = 0; /* force reval */
 				inode_set_ctime_current(inode);
 				inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
+			} else {
+				cifs_mark_open_handles_for_deleted_file(inode, to_name);
+				cifs_drop_nlink(inode);
+				inode_set_ctime_current(inode);
 			}
 		} else if (rc == -EACCES || rc == -EEXIST) {
 			/*
Signature · 1 line
2.54.0
#9 lore raw

Re: [PATCH v2 3/3] smb/client: fix nlink of an overwritten open file

Quoted 1 line

This patch fixes xfstests generic/035 when mounted with posix option.

generic/035 passes to Samba with posix mount option without the patch. Any idea why this patch was needed for ksmbd but not for Samba

Quoted 53 lines · ChenXiaoSong

From: ChenXiaoSong <[email protected]>

Reproducer:

1. server: systemctl start ksmbd
2. client: mount with posix option

     mount -t cifs -o posix //${server_ip}/export /mnt

3. client: touch /mnt/file1 /mnt/file2
4. client: C program: int fd = open("/mnt/file2", O_RDONLY);
5. client: C program: rename("/mnt/file1", "/mnt/file2");
6. client: C program: struct stat stbuf; fstat(fd, &stbuf); stbuf.st_nlink is 1, should be 0

This patch fixes xfstests generic/035 when mounted with posix option.

---
 fs/smb/client/inode.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
fs/smb/client/inode.c +6 -6
index 0afff761aab9..0b6273ce01a8 100644
@@ -2648,11 +2648,8 @@ cifs_rename2(struct mnt_idmap *idmap, struct inode *source_dir,
        if (d_really_is_positive(target_dentry)) {
                if (!rc) {
                        struct inode *inode = d_inode(target_dentry);
-                       /*
-                        * Samba and ksmbd servers allow renaming a target
-                        * directory that is open, so make sure to update
-                        * ->i_nlink and then mark it as delete pending.
-                        */
+
+                       /* Update the target link count after rename. */
                        if (S_ISDIR(inode->i_mode)) {
                                drop_cached_dir_by_name(xid, tcon, to_name, cifs_sb);
                                spin_lock(&inode->i_lock);
@@ -2663,6 +2660,10 @@ cifs_rename2(struct mnt_idmap *idmap, struct inode *source_dir,
                                CIFS_I(inode)->time = 0; /* force reval */
                                inode_set_ctime_current(inode);
                                inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
+                       } else {
+                               cifs_mark_open_handles_for_deleted_file(inode, to_name);
+                               cifs_drop_nlink(inode);
+                               inode_set_ctime_current(inode);
                        }
                } else if (rc == -EACCES || rc == -EEXIST) {
                        /*
--

2.54.0

Signature · 3 lines
Thanks,

Steve
author reviewer
#10 lore raw

Re: [PATCH v2 3/3] smb/client: fix nlink of an overwritten open file

Samba has a bug, please check the email: https://lore.kernel.org/linux-cifs/[email protected]/

Quoted 2 lines · Steve French

generic/035 passes to Samba with posix mount option without the patch. Any idea why this patch was needed for ksmbd but not for Samba

Signature · 3 lines
ChenXiaoSong <[email protected]>
Chinese Homepage: https://chenxiaosong.com
English Homepage: https://chenxiaosong.com/en
author reviewer
#11 lore raw

Re: [PATCH v2 3/3] smb/client: fix nlink of an overwritten open file

As Ralph said,

- if there are one or more non-POSIX opens on the destination, MUST fail with STATUS_ACCESS_DENIED
- if there are only POSIX opens, the operation can proceed

Samba should allow the rename to succeed so that the silly-rename is not triggered.

We should fix this bug in Samba so that the Linux client should call cifs_drop_nlink(). Since silly-rename will not be triggered, __cifs_unlink() will not be called either.

Quoted 6 lines · ChenXiaoSong

Samba has a bug, please check the email: https://lore.kernel.org/linux- cifs/[email protected]/

Quoted 2 lines · Steve French

generic/035 passes to Samba with posix mount option without the patch. Any idea why this patch was needed for ksmbd but not for Samba

Signature · 3 lines
ChenXiaoSong <[email protected]>
Chinese Homepage: https://chenxiaosong.com
English Homepage: https://chenxiaosong.com/en