CIFS posix acl permission checking
RFC PATCH 1/1 CIFS posix acl permission checking
Hello everyone
Firstly, please note that I'm new to the kernel community, so if I'm doing something wrong, just please politely point it out to me and I'll try to fix it, or at the very least not do it again. Also please bear in mind that English isn't my native language.
Anyway, I recently realised that while the CIFS file system driver in Linux does supports posix acl retrieval and modification using the getfacl and setfacl command line tools, it does not use acl for client side permission checking. On the server side Samba does consult the acl, but that doesn't really help when cifs.ko never even asks the server, due to the users only source of permission being from an acl.
I'm attaching a first attempt at implementing it. I have tested it, but only on a single setup, so I can give no guarantees to its portability. Also please note that this is my first kernel patch, so if I'm doing something wrong, such as not following some coding standard, please enlighten me and I'll gladly fix it.
I only subscribed to the linux-fsdevel list, so please include it in any response you might send.
Best Regards
Jon Severinsson
commit fab1872fcbb8a5cdb85d7e7a88ecf0cad99d1c80
Author: Jon Severinsson <[email protected]> Date: Mon Mar 1 15:49:40 2010 +0100
[CIFS] Adds support for permision checking vs. posix acl.
CIFS already supports getting and setting acls through getfacl and setfacl, but prior to this patch, any acls was ignored when doing permission checking.
fs/cifs/cifsfs.c +10 -1
@@ -284,7 +284,7 @@ static int cifs_permission(struct inode *inode, int mask)on the client (above and beyond ACL on servers) for servers which do not support setting and viewing mode bits, so allowing client to check permissions is useful */ -return generic_permission (inode, mask,NULL ); +return generic_permission (inode, mask, inode->i_op->check_acl); }static struct kmem_cache *cifs_inode_cachep;
@@ -702,6 +702,9 @@ const struct inode_operations cifs_dir_inode_ops = { .getxattr = cifs_getxattr, .listxattr = cifs_listxattr, .removexattr = cifs_removexattr, +#ifdef CONFIG_CIFS_POSIX + .check_acl = cifs_check_acl, +#endif #endif };
@@ -716,6 +719,9 @@ const struct inode_operations cifs_file_inode_ops = { .getxattr = cifs_getxattr, .listxattr = cifs_listxattr, .removexattr = cifs_removexattr, +#ifdef CONFIG_CIFS_POSIX + .check_acl = cifs_check_acl, +#endif #endif };
@@ -732,6 +738,9 @@ const struct inode_operations cifs_symlink_inode_ops = { .getxattr = cifs_getxattr, .listxattr = cifs_listxattr, .removexattr = cifs_removexattr, +#ifdef CONFIG_CIFS_POSIX + .check_acl = cifs_check_acl, +#endif #endif };
fs/cifs/cifsfs.h +4 -0
@@ -102,11 +102,15 @@ extern int cifs_readlink(struct dentry *direntry, char __user *buffer, int buflen);extern int cifs_symlink (struct inode *inode,struct dentry *direntry,const char *symname); + +/* Functions related to extended attributes */ extern int cifs_removexattr (struct dentry *,const char *);extern int cifs_setxattr (struct dentry *,const char *,const void *,size_t ,int );extern ssize_t cifs_getxattr (struct dentry *,const char *,void *,size_t );extern ssize_t cifs_listxattr (struct dentry *,char *,size_t ); +extern int cifs_check_acl (struct inode *inode,int mask); +extern long cifs_ioctl (struct file *filep,unsigned int cmd,unsigned long arg);#ifdef CONFIG_CIFS_EXPERIMENTAL
fs/cifs/xattr.c +71 -0
74 lines not on this page · raw message
Quoted 6 lines · Jon Severinsson
Hello everyone
Firstly, please note that I'm new to the kernel community, so if I'm doing something wrong, just please politely point it out to me and I'll try to fix it, or at the very least not do it again. Also please bear in mind that English isn't my native language.
Okey dokey, here's some minor mistakes you made.
- Generally a good idea to cc the CIFS development list for CIFS patches. I'm sure they'll notice it here, but just to be on the safe side, I've added them to the cc.
- You've included ifdefs around the check_acl entry in inode_operations, *and* inside the definition of cifs_check_acl. You only need to do one or the other, and opinion is divided on which is better.
(further comments inline)
Quoted 98 lines
Anyway, I recently realised that while the CIFS file system driver in Linux does supports posix acl retrieval and modification using the getfacl and setfacl command line tools, it does not use acl for client side permission checking. On the server side Samba does consult the acl, but that doesn't really help when cifs.ko never even asks the server, due to the users only source of permission being from an acl.
I'm attaching a first attempt at implementing it. I have tested it, but only on a single setup, so I can give no guarantees to its portability. Also please note that this is my first kernel patch, so if I'm doing something wrong, such as not following some coding standard, please enlighten me and I'll gladly fix it.
I only subscribed to the linux-fsdevel list, so please include it in any response you might send.
Best Regards
Jon Severinssoncommit
fab1872fcbb8a5cdb85d7e7a88ecf0cad99d1c80Author: Jon Severinsson <[email protected]> Date: Mon Mar 1 15:49:40 2010 +0100[CIFS] Adds support for permision checking vs. posix acl.
CIFS already supports getting and setting acls through getfacl and setfacl, but prior to this patch, any acls was ignored when doing permission checking.
fs/cifs/cifsfs.c +10
-1@@ -284,7 +284,7 @@ static int cifs_permission(struct inode *inode, int mask)on the client (above and beyond ACL on servers) for servers which do not support setting and viewing mode bits, so allowing client to check permissions is useful */ -return generic_permission (inode, mask,NULL ); +return generic_permission (inode, mask, inode->i_op->check_acl); }static struct kmem_cache *cifs_inode_cachep;@@ -702,6 +702,9 @@ const struct inode_operations cifs_dir_inode_ops = { .getxattr = cifs_getxattr, .listxattr = cifs_listxattr, .removexattr = cifs_removexattr, +#ifdef CONFIG_CIFS_POSIX + .check_acl = cifs_check_acl, +#endif #endif };@@ -716,6 +719,9 @@ const struct inode_operations cifs_file_inode_ops = { .getxattr = cifs_getxattr, .listxattr = cifs_listxattr, .removexattr = cifs_removexattr, +#ifdef CONFIG_CIFS_POSIX + .check_acl = cifs_check_acl, +#endif #endif };@@ -732,6 +738,9 @@ const struct inode_operations cifs_symlink_inode_ops = { .getxattr = cifs_getxattr, .listxattr = cifs_listxattr, .removexattr = cifs_removexattr, +#ifdef CONFIG_CIFS_POSIX + .check_acl = cifs_check_acl, +#endif #endif };fs/cifs/cifsfs.h +4
-0@@ -102,11 +102,15 @@ extern int cifs_readlink(struct dentry *direntry, char __user *buffer, int buflen);extern int cifs_symlink (struct inode *inode,struct dentry *direntry,const char *symname); + +/* Functions related to extended attributes */ extern int cifs_removexattr (struct dentry *,const char *);extern int cifs_setxattr (struct dentry *,const char *,const void *,size_t ,int );extern ssize_t cifs_getxattr (struct dentry *,const char *,void *,size_t );extern ssize_t cifs_listxattr (struct dentry *,char *,size_t ); +extern int cifs_check_acl (struct inode *inode,int mask); +extern long cifs_ioctl (struct file *filep,unsigned int cmd,unsigned long arg);#ifdef CONFIG_CIFS_EXPERIMENTAL fs/cifs/xattr.c +0
-0@@ -374,3 +374,74 @@ ssize_t cifs_listxattr(struct dentry *direntry, char *data, size_t buf_size)#endif return rc;
fs/cifs/xattr.c ssize_t cifs_listxattr(struct dentry *direntry, char *data, size_t buf_size) · line 385
Quoted 10 lines
fs/cifs/xattr.c +9
-0} + +int cifs_check_acl (struct inode *inode,int mask) +{ +int rc = -EOPNOTSUPP ; +#ifdef CONFIG_CIFS_XATTR +#ifdef CONFIG_CIFS_POSIX +struct dentry *dentry =NULL ; +size_t buf_size; +void *ea_value =NULL ;
I don't think you need to initialise ea_value, do you?
Quoted 4 lines
fs/cifs/xattr.c +4
-0+ssize_t ea_size; +struct posix_acl *acl =NULL ; + +/* CIFS gets acl from server by path, and thus needs a dentry rather than
fs/cifs/xattr.c ssize_t cifs_listxattr(struct dentry *direntry, char *data, size_t buf_size) · line 399
Quoted 10 lines
fs/cifs/xattr.c +10
-0+an inode. Note that the path of each dentry will point to the same inode +on the backing fs at the server, so their acls will be the same, and it +doesn't matter which one we pick, so just pick the fist. */ +if (list_empty (&inode->i_dentry)) + dentry = list_first_entry(&inode->i_dentry, struct dentry, d_alias); + +/* If we didn't get an dentry for the inode, something went terribly wrong. +All we can do at this point is to return an error. */ +if (!dentry ||IS_ERR (dentry)) +return PTR_ERR (dentry);
I don't think dentry can possibly be IS_ERR here, and returning PTR_ERR(NULL) doesn't do what you think it does. Also, you've got the sense of list_empty backwards. I think what you meant to do here was:
if (!list_empty(&inode->i_dentry)) dentry = list_first_entry(&inode->i_dentry, struct dentry, d_alias); else return -EINVAL;
Quoted 47 lines
fs/cifs/xattr.c +47
-0+/* Try to fit the extended attribute corresponding to the posix acl in 4k +memory. 4k was chosen because it always fits in a single page, and is +the maximum on a default ext2/3/4 backing fs. */ + buf_size =4096 ; + ea_value =kmalloc (buf_size,GFP_KERNEL ); +if (!ea_value) { + rc = -EAGAIN ; + goto check_acl_exit; + } + ea_size =cifs_getxattr (dentry,POSIX_ACL_XATTR_ACCESS , ea_value, buf_size); + +/* If 4k wasn't enough, try 64k, the maximum on any current backing fs. */ +if (ea_size == -ERANGE ) { +kfree (ea_value); + buf_size =65536 ; + ea_value =kmalloc (buf_size,GFP_KERNEL ); +if (!ea_value) { + rc = -EAGAIN ; + goto check_acl_exit; + } + ea_size =cifs_getxattr (dentry,POSIX_ACL_XATTR_ACCESS , ea_value, buf_size); + } + +/* If we didn't get any extended attribute, set the error code and exit */ +if (ea_size <=0 ) { + rc = -EAGAIN ; +if (ea_size == -EOPNOTSUPP || ea_size == -EIO || ea_size == -ENOTDIR || ea_size == -ENOENT || ea_size == -EFAULT || ea_size == -EACCES ) + rc = ea_size; + goto check_acl_exit; + } + +/* Set the appropriate return value. Adapted from ext4. */ + acl =posix_acl_from_xattr (ea_value, ea_size); +if (IS_ERR (acl)) + rc =PTR_ERR (acl); +else if (acl) + rc =posix_acl_permission (inode, acl, mask); +else + rc = -EAGAIN ; + +check_acl_exit: +posix_acl_release (acl); +kfree (ea_value); +#endif /* CONFIG_CIFS_POSIX */ +#endif /* CONFIG_CIFS_XATTR */ +return rc; +}
Signature · 4 lines
Matthew Wilcox Intel Open Source Technology Centre "Bill, look, we understand that you're interested in selling us this operating system, but compare it to ours. We can't possibly take such a retrograde step."
Quoted 7 lines · Matthew Wilcox
Quoted 4 lines · Jon Severinsson
Firstly, please note that I'm new to the kernel community, so if I'm doing something wrong, just please politely point it out to me and I'll try to fix it, or at the very least not do it again. Also please bear in mind that English isn't my native language.
Okey dokey, here's some minor mistakes you made.
Thank you.
Quoted 3 lines
- Generally a good idea to cc the CIFS development list for CIFS patches. I'm sure they'll notice it here, but just to be on the safe side, I've added them to the cc.
Sure thing. I used linux-fsdevel as I didn't find anything more specific on vger.kernel.org, didn't think to look at lists.samba.org.
Quoted 3 lines
- You've included ifdefs around the
check_aclentry ininode_operations, *and* inside the definition ofcifs_check_acl. You only need to do one or the other, and opinion is divided on which is better.
While I did recognize the redundancy, I decided to follow the same convention the other functions in xattr.c did, and include ifdefs at both locations.
I also considered the possible reasons for the existing functions to do both, and and came up with two reasons. The first simply being the paradigm of defensive programming, always check before doing a call, but never assume that the check has been done before being called. The second one is that of performance. The ifdefs has to be in cifs_check_acl to protect against other callers (while this patch doesn't introduce any, it doesn't prevent further patches from adding them), and not including the ifdefs in inode_operations would mean a completely useless function call on every permission check when a feature was turned off at compile time. The second one is a micro-optimization I
don't really care fore, but defensive programming I do respect.
With this in mind, what do you recommend, double protection, breaking convention or changing the existing code?
Quoted 11 lines
Quoted 11 lines
fs/cifs/xattr.c +5
-0@@ -374,3 +374,74 @@ ssize_t cifs_listxattr(struct dentry *direntry, char *data, size_t buf_size) #endifreturn rc; } + +int cifs_check_acl (struct inode *inode,int mask) +{ +int rc = -EOPNOTSUPP ; +#ifdef CONFIG_CIFS_XATTR
Quoted 8 lines
fs/cifs/xattr.c ssize_t cifs_listxattr(struct dentry *direntry, char · line 387
Quoted 6 lines
fs/cifs/xattr.c +6
-0+#ifdef CONFIG_CIFS_POSIX +struct dentry *dentry =NULL ; +size_t buf_size; +void *ea_value =NULL ; +ssize_t ea_size; +struct posix_acl *acl =NULL ;I don't think you need to initialise
ea_value, do you?
While currently correct, I find it a good idea to immediately null any pointer that is freed in the exit section. Otherwise it is very easy to forget to do that the day someone adds a goto prior to the first assignment, and not nulling then can have unintended consequences in unrelated code. That being said, if you say that the kernel community frowns upon that kind of defensive programming I will definitely remove it.
With your suggested improvement below, there is no need to NULL dentry though, so I'll remove that.
Quoted 22 lines
Quoted 14 lines
fs/cifs/xattr.c +1
-0+/* CIFS gets acl from server by path, and thus needs a dentry rather than
fs/cifs/xattr.c +1
-0+ an inode. Note that the path of each dentry will point to thesame inode
fs/cifs/xattr.c +1
-0+ on the backing fs at the server, so their acls will bethe same, and it
fs/cifs/xattr.c +1
-0+ doesn't matter which one we pick, so just pick thefist. */
fs/cifs/xattr.c +2
-0+if (list_empty (&inode->i_dentry)) + dentry = list_first_entry(&inode->i_dentry, struct dentry,
d_alias);fs/cifs/xattr.c +2
-0+ +/* If we didn't get an dentry for the inode, something went terribly wrong.
fs/cifs/xattr.c
Quoted 3 lines
fs/cifs/xattr.c +3
-0+All we can do at this point is to return an error. */ +if (!dentry ||IS_ERR (dentry)) +return PTR_ERR (dentry);I don't think dentry can possibly be
IS_ERRhere, and returningPTR_ERR(NULL)doesn't do what you think it does. Also, you've got the
sense oflist_emptybackwards. I think what you meant to do here was:
Quoted 5 lines
if (!list_empty(&inode->i_dentry)) dentry = list_first_entry(&inode->i_dentry, struct dentry, d_alias); else return -EINVAL;
Sure, this makes more sense. Originally I used d_find_alias(inode), but later thought it was better to use the information in the inode struct directly. I never really understood how list.h lists was supposed to work, but stopped when cut-and-paste got it working for me. Probably a bad idea...
Attaching an updated patch, mostly for the convenience of anyone only following linux-cifs-client.
And thanks again for your input.
Best Regards
Jon Severinsson
commit fa0b9415cda17b31966542101bc4ceb0c97c87cb
Author: Jon Severinsson <[email protected]> Date: Mon Mar 1 19:24:30 2010 +0100
[CIFS] Adds support for permision checking vs. posix acl.
CIFS already supports getting and setting acls through getfacl and setfacl, but prior to this patch, any acls was ignored when doing permission checking.
fs/cifs/cifsfs.c +10 -1
@@ -284,7 +284,7 @@ static int cifs_permission(struct inode *inode, int mask)on the client (above and beyond ACL on servers) for servers which do not support setting and viewing mode bits, so allowing client to check permissions is useful */ -return generic_permission (inode, mask,NULL ); +return generic_permission (inode, mask, inode->i_op->check_acl); }static struct kmem_cache *cifs_inode_cachep;
@@ -702,6 +702,9 @@ const struct inode_operations cifs_dir_inode_ops = { .getxattr = cifs_getxattr, .listxattr = cifs_listxattr, .removexattr = cifs_removexattr, +#ifdef CONFIG_CIFS_POSIX + .check_acl = cifs_check_acl, +#endif #endif };
@@ -716,6 +719,9 @@ const struct inode_operations cifs_file_inode_ops = { .getxattr = cifs_getxattr, .listxattr = cifs_listxattr, .removexattr = cifs_removexattr, +#ifdef CONFIG_CIFS_POSIX + .check_acl = cifs_check_acl, +#endif #endif };
@@ -732,6 +738,9 @@ const struct inode_operations cifs_symlink_inode_ops = { .getxattr = cifs_getxattr, .listxattr = cifs_listxattr, .removexattr = cifs_removexattr, +#ifdef CONFIG_CIFS_POSIX + .check_acl = cifs_check_acl, +#endif #endif };
fs/cifs/cifsfs.h +4 -0
@@ -102,11 +102,15 @@ extern int cifs_readlink(struct dentry *direntry, char __user *buffer, int buflen);extern int cifs_symlink (struct inode *inode,struct dentry *direntry,const char *symname); + +/* Functions related to extended attributes */ extern int cifs_removexattr (struct dentry *,const char *);extern int cifs_setxattr (struct dentry *,const char *,const void *,size_t ,int );extern ssize_t cifs_getxattr (struct dentry *,const char *,void *,size_t );extern ssize_t cifs_listxattr (struct dentry *,char *,size_t ); +extern int cifs_check_acl (struct inode *inode,int mask); +extern long cifs_ioctl (struct file *filep,unsigned int cmd,unsigned long arg);#ifdef CONFIG_CIFS_EXPERIMENTAL
fs/cifs/xattr.c +68 -0
71 lines not on this page · raw message