Skip to contentSkip to conversation

CIFS should honour umask

PATCH In review Matt Keenan 2 participants 4 messages Jun 6, 2007 – Jun 7, 2007 1 message not in the archive VFS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 CIFS should honour umask

This patch makes CIFS honour a process' umask like other filesystems. Of course the server is still free to munge the permissions if it wants to; but the client will send the "right" permissions to begin with.

A few caveats;

1) It only applies to filesystems that have CAP_UNIX (aka support unix extensions)
2) It applies the correct mode to the follow up CIFSSMBUnixSetPerms() after remote creation (I can write a new patch if you want with the "right" mode at actual creation time; however the "right" perms will still need to be given to the follow up CIFSSMBUnixSetPerms() anyway).
3) It will probably work best with Samba 3.0.25a or newer (ie with this patch applied
http://lists.samba.org/archive/linux-cifs-client/2007-January/001697.html)
4) It has been compiled, and tested on 2.6.22-rc4 / Samba 3.0.25a (Ubuntu Dapper with a few custom backports), and with a bit of testing seems to work just fine. (it also incidentally side steps bugs in thunderbird and openoffice (the apps don't check the permissions on files they create, they assume they will open() the way that have asked them to be created xref open(O_WRONLY|O_CREAT) => valid fd then mmap(fd,PROT_READ) => EFAULT).

I am going to give this patch a more thorough test tomorrow with ltp. Comments, corrections, et al are welcome.

Matt

Signature · 2 lines
Matt Keenan
OpCode Solutions

--

fs/cifs/dir.c +6 -2
diff -urN linux-2.6.22-rc4/fs/cifs/dir.c linux-2.6.22-rc4.cifs-umask-fix/fs/cifs/dir.c
@@ -206,7 +206,11 @@
 		/* If Open reported that we actually created a file
 		then we now have to set the mode if possible */
 		if ((cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
-			(oplock & CIFS_CREATE_ACTION))
+			(oplock & CIFS_CREATE_ACTION)) {
+			/* respect umask settings like other filesystems;
+			 * if the server wants to munge the bits let it, but the client
+			 * should "Do The Right Thing" (tm) */
+			mode &= ~current->fs->umask;
 			if(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SET_UID) {
 				CIFSSMBUnixSetPerms(xid, pTcon, full_path, mode,
 					(__u64)current->fsuid,
@@ -224,7 +228,7 @@
 					cifs_sb->mnt_cifs_flags & 
 						CIFS_MOUNT_MAP_SPECIAL_CHR);
 			}
-		else {
+		} else {
 			/* BB implement mode setting via Windows security descriptors */
 			/* eg CIFSSMBWinSetPerms(xid,pTcon,full_path,mode,-1,-1,local_nls);*/
 			/* could set r/o dos attribute if mode & 0222 == 0 */
fs/cifs/inode.c +6 -1
diff -urN linux-2.6.22-rc4/fs/cifs/inode.c linux-2.6.22-rc4.cifs-umask-fix/fs/cifs/inode.c
@@ -987,6 +987,11 @@
 		if ((direntry->d_inode) && (direntry->d_inode->i_nlink < 2))
 				direntry->d_inode->i_nlink = 2; 
 		if (cifs_sb->tcon->ses->capabilities & CAP_UNIX)
+		{
+			/* respect umask settings like other filesystems;
+			 * if the server wants to munge the bits let it, but the client
+			 * should "Do The Right Thing" (tm) */
+			mode &= ~current->fs->umask;
 			if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SET_UID) {
 				CIFSSMBUnixSetPerms(xid, pTcon, full_path,
 						    mode,
@@ -1004,7 +1009,7 @@
 						    cifs_sb->mnt_cifs_flags & 
 						    CIFS_MOUNT_MAP_SPECIAL_CHR);
 			}
-		else {
+		} else {
 			/* BB to be implemented via Windows secrty descriptors
 			   eg CIFSSMBWinSetPerms(xid, pTcon, full_path, mode,
 						 -1, -1, local_nls); */
#1 lore raw

Thanks - it looks almost right but you missed mknod case and your patch had some whitespace/formatting problems.

Could you try the following and make sure it works for you? If so will merge.

fs/cifs/dir.c +4 -2
index f085db9..8e86aac 100644
@@ -208,7 +208,8 @@ cifs_create(struct inode *inode, struct
               /* If Open reported that we actually created a file
               then we now have to set the mode if possible */
               if ((cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
-                       (oplock & CIFS_CREATE_ACTION))
+                       (oplock & CIFS_CREATE_ACTION)) {
+                       mode &= ~current->fs->umask;
                       if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SET_UID) {
                               CIFSSMBUnixSetPerms(xid, pTcon, full_path, mode,
                                       (__u64)current->fsuid,
@@ -226,7 +227,7 @@ cifs_create(struct inode *inode, struct
                                       cifs_sb->mnt_cifs_flags &
                                               CIFS_MOUNT_MAP_SPECIAL_CHR);
                       }
-               else {
+               } else {
                       /* BB implement mode setting via Windows security
                          descriptors e.g. */
                       /* CIFSSMBWinSetPerms(xid,pTcon,path,mode,-1,-1,nls);*/
@@ -336,6 +337,7 @@ int cifs_mknod(struct inode *inode, stru
       if (full_path == NULL)
               rc = -ENOMEM;
       else if (pTcon->ses->capabilities & CAP_UNIX) {
+               mode &= ~current->fs->umask;
               if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SET_UID) {
                       rc = CIFSSMBUnixSetPerms(xid, pTcon, full_path,
                               mode, (__u64)current->fsuid,
fs/cifs/inode.c +3 -2
index 3e87dad..f0ff12b 100644
@@ -986,7 +986,8 @@ mkdir_get_info:
                 * failed to get it from the server or was set bogus */
               if ((direntry->d_inode) && (direntry->d_inode->i_nlink < 2))
                               direntry->d_inode->i_nlink = 2;
-               if (cifs_sb->tcon->ses->capabilities & CAP_UNIX)
+               if (cifs_sb->tcon->ses->capabilities & CAP_UNIX) {
+                       mode &= ~current->fs->umask;
                       if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SET_UID) {
                               CIFSSMBUnixSetPerms(xid, pTcon, full_path,
                                                   mode,
@@ -1004,7 +1005,7 @@ mkdir_get_info:
                                                   cifs_sb->mnt_cifs_flags &
                                                   CIFS_MOUNT_MAP_SPECIAL_CHR);
                       }
-               else {
+               } else {
                       /* BB to be implemented via Windows secrty descriptors
                          eg CIFSSMBWinSetPerms(xid, pTcon, full_path, mode,
                                                -1, -1, local_nls); */
Quoted 89 lines · Matt Keenan

This patch makes CIFS honour a process' umask like other filesystems. Of course the server is still free to munge the permissions if it wants to; but the client will send the "right" permissions to begin with.

A few caveats;

1) It only applies to filesystems that have CAP_UNIX (aka support unix extensions)
2) It applies the correct mode to the follow up CIFSSMBUnixSetPerms() after remote creation (I can write a new patch if you want with the "right" mode at actual creation time; however the "right" perms will still need to be given to the follow up CIFSSMBUnixSetPerms() anyway).
3) It will probably work best with Samba 3.0.25a or newer (ie with this patch applied
http://lists.samba.org/archive/linux-cifs-client/2007-January/001697.html)
4) It has been compiled, and tested on 2.6.22-rc4 / Samba 3.0.25a (Ubuntu Dapper with a few custom backports), and with a bit of testing seems to work just fine. (it also incidentally side steps bugs in thunderbird and openoffice (the apps don't check the permissions on files they create, they assume they will open() the way that have asked them to be created xref open(O_WRONLY|O_CREAT) => valid fd then mmap(fd,PROT_READ) => EFAULT).

I am going to give this patch a more thorough test tomorrow with ltp. Comments, corrections, et al are welcome.

Matt

--
Matt Keenan
OpCode Solutions

--
diff -urN linux-2.6.22-rc4/fs/cifs/dir.c linux-2.6.22-rc4.cifs-umask-fix/fs/cifs/dir.c

fs/cifs/dir.c 2007-06-06 19:27:00.000000000 +0100 renamed from fs/cifs/dir.c 2007-06-06 08:34:03.000000000 +0100 +6 -2
@@ -206,7 +206,11 @@
                /* If Open reported that we actually created a file
                then we now have to set the mode if possible */
                if ((cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
-                       (oplock & CIFS_CREATE_ACTION))
+                       (oplock & CIFS_CREATE_ACTION)) {
+                       /* respect umask settings like other filesystems;
+                        * if the server wants to munge the bits let it, but the client
+                        * should "Do The Right Thing" (tm) */
+                       mode &= ~current->fs->umask;
                        if(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SET_UID) {
                                CIFSSMBUnixSetPerms(xid, pTcon, full_path, mode,
                                        (__u64)current->fsuid,
@@ -224,7 +228,7 @@
                                        cifs_sb->mnt_cifs_flags &
                                                CIFS_MOUNT_MAP_SPECIAL_CHR);
                        }
-               else {
+               } else {
                        /* BB implement mode setting via Windows security descriptors */
                        /* eg CIFSSMBWinSetPerms(xid,pTcon,full_path,mode,-1,-1,local_nls);*/
                        /* could set r/o dos attribute if mode & 0222 == 0 */

diff -urN linux-2.6.22-rc4/fs/cifs/inode.c linux-2.6.22-rc4.cifs-umask-fix/fs/cifs/inode.c

fs/cifs/inode.c 2007-06-06 19:26:26.000000000 +0100 renamed from fs/cifs/inode.c 2007-06-06 08:34:03.000000000 +0100 +6 -1
@@ -987,6 +987,11 @@
                if ((direntry->d_inode) && (direntry->d_inode->i_nlink < 2))
                                direntry->d_inode->i_nlink = 2;
                if (cifs_sb->tcon->ses->capabilities & CAP_UNIX)
+               {
+                       /* respect umask settings like other filesystems;
+                        * if the server wants to munge the bits let it, but the client
+                        * should "Do The Right Thing" (tm) */
+                       mode &= ~current->fs->umask;
                        if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SET_UID) {
                                CIFSSMBUnixSetPerms(xid, pTcon, full_path,
                                                    mode,
@@ -1004,7 +1009,7 @@
                                                    cifs_sb->mnt_cifs_flags &
                                                    CIFS_MOUNT_MAP_SPECIAL_CHR);
                        }
-               else {
+               } else {
                        /* BB to be implemented via Windows secrty descriptors
                           eg CIFSSMBWinSetPerms(xid, pTcon, full_path, mode,
                                                 -1, -1, local_nls); */
Signature · 3 lines
Thanks,

Steve
#3 lore raw
In reply to a message not in the archive

For the non-unix case (e.g. Windows servers) the mode will be taken from the default specified on the mount. I am not sure if we also should add code to also honor umask in that case.

I am not sure how common it is to change umask to different values in different processes which would access the same mount.

Thoughts?

Quoted 6 lines
Quoted 5 lines · Steve French

Thanks - it looks almost right but you missed mknod case and your patch had some whitespace/formatting problems.

Could you try the following and make sure it works for you? If so will merge.

#4 lore raw
Quoted 4 lines · Steve French

For the non-unix case (e.g. Windows servers) the mode will be taken from the default specified on the mount. I am not sure if we also should add code to also honor umask in that case.

I don't think it would be necessary to add code for the windows case, we should just rely on windows semantics (with acl if supplied) for pure windows mounts.

Quoted 3 lines

I am not sure how common it is to change umask to different values in different processes which would access the same mount.

I think if we are going to use umask we should use it the way it is used in Unix, i.e. tagged to the process. I realise that in some circumstances that having a mount wide umask would be handy, however if we are going to apply unix style semantics we should stick to the unix style process bound umask; i.e. the method of least suprise.

Matt

Signature · 2 lines
Matt Keenan
OpCode Solutions