udf: fix uid/gid options and add uid/gid=ignore and forget options
PATCH 1/1 udf: fix uid/gid options and add uid/gid=ignore and forget options
This patch fixes a bug in udf where it would write uid/gid = 0 to the disk for files owned by the id given with the uid=/gid= mount options. It also adds 4 new mount options: uid/gid=forget and uid/gid=ignore. Without any options the id in core and on disk always match. Giving uid/gid=nnn specifies a default ID to be used in core when the on disk ID is -1. uid/gid=ignore forces the in core ID to allways be used no matter what the on disk ID is. uid/gid=forget forces the on disk ID to always be written out as -1.
The use of these options allows you to override ownerships on a disk or disable ownwership information from being written, allowing the media to be used portably between different computers and possibly different users without permissions issues that would require root to correct.
---
fs/udf/inode.c | 14 ++++++++++---- fs/udf/super.c | 18 +++++++++++++++++- fs/udf/udf_sb.h | 4 ++++ 3 files changed, 31 insertions(+), 5 deletions(-)
2428c90c6e52a317808896b6cd639199388f7ddd
fs/udf/inode.c +0 -0
@@ -1045,10 +1045,12 @@ static void udf_fill_inode(struct inode
} inode->i_uid = le32_to_cpu(fe->uid);
- if ( inode->i_uid == -1 ) inode->i_uid = UDF_SB(inode->i_sb)->s_uid;
+ if ( inode->i_uid == -1 || UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_UID_IGNORE) )
+ inode->i_uid = UDF_SB(inode->i_sb)->s_uid;
inode->i_gid = le32_to_cpu(fe->gid);
- if ( inode->i_gid == -1 ) inode->i_gid = UDF_SB(inode->i_sb)->s_gid;
+ if ( inode->i_gid == -1 || UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_GID_IGNORE) )
+ inode->i_gid = UDF_SB(inode->i_sb)->s_gid;
inode->i_nlink = le16_to_cpu(fe->fileLinkCount); if (!inode->i_nlink) @@ -1335,10 +1337,14 @@ udf_update_inode(struct inode *inode, in return err; }
- if (inode->i_uid != UDF_SB(inode->i_sb)->s_uid)
+ if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_UID_FORGET))
+ fe->uid = cpu_to_le32(-1);
+ else if (inode->i_uid != UDF_SB(inode->i_sb)->s_uid)
fe->uid = cpu_to_le32(inode->i_uid);
- if (inode->i_gid != UDF_SB(inode->i_sb)->s_gid)
+ if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_GID_FORGET))
+ fe->gid = cpu_to_le32(-1);
+ else if (inode->i_gid != UDF_SB(inode->i_sb)->s_gid)
fe->gid = cpu_to_le32(inode->i_gid); udfperms = ((inode->i_mode & S_IRWXO) ) |
fs/udf/super.c +0 -0
@@ -269,7 +269,7 @@ enum {
Opt_gid, Opt_uid, Opt_umask, Opt_session, Opt_lastblock, Opt_anchor, Opt_volume, Opt_partition, Opt_fileset, Opt_rootdir, Opt_utf8, Opt_iocharset,
- Opt_err
+ Opt_err, Opt_uforget, Opt_uignore, Opt_gforget, Opt_gignore
};
static match_table_t tokens = {
@@ -282,6 +282,10 @@ static match_table_t tokens = {
{Opt_adinicb, "adinicb"},
{Opt_shortad, "shortad"},
{Opt_longad, "longad"},+ {Opt_uforget, "uid=forget"},
+ {Opt_uignore, "uid=ignore"},
+ {Opt_gforget, "gid=forget"},
+ {Opt_gignore, "gid=ignore"},
{Opt_gid, "gid=%u"},
{Opt_uid, "uid=%u"},
{Opt_umask, "umask=%o"},
@@ -414,6 +418,18 @@ udf_parse_options(char *options, struct
uopt->flags |= (1 << UDF_FLAG_NLS_MAP);
break;#endif
+ case Opt_uignore:
+ uopt->flags |= (1 << UDF_FLAG_UID_IGNORE);
+ break;
+ case Opt_uforget:
+ uopt->flags |= (1 << UDF_FLAG_UID_FORGET);
+ break;
+ case Opt_gignore:
+ uopt->flags |= (1 << UDF_FLAG_GID_IGNORE);
+ break;
+ case Opt_gforget:
+ uopt->flags |= (1 << UDF_FLAG_GID_FORGET);
+ break; default: printk(KERN_ERR "udf: bad mount option \"%s\" " "or missing value\n", p);
fs/udf/udf_sb.h +0 -0
@@ -20,6 +20,10 @@
#define UDF_FLAG_VARCONV 8
#define UDF_FLAG_NLS_MAP 9
#define UDF_FLAG_UTF8 10
+#define UDF_FLAG_UID_FORGET 11 /* save -1 for uid to disk */ +#define UDF_FLAG_UID_IGNORE 12 /* use sb uid instead of on disk uid */ +#define UDF_FLAG_GID_FORGET 13 +#define UDF_FLAG_GID_IGNORE 14
#define UDF_PART_FLAG_UNALLOC_BITMAP 0x0001
#define UDF_PART_FLAG_UNALLOC_TABLE 0x0002
Signature · 1 line
1.1.3
Quoted 2 lines · Phillip Susi
This patch fixes a bug in udf where it would write uid/gid = 0 to the disk for files owned by the id given with the uid=/gid= mount options.
I don't think it does. You can still get zero uid/gid written on disk if you specify uid/gid mount options when not specifying uid=forget. I am not against new features but lets fix the bug first, okay? Adding new mount options to work around buggy code is not the right solution.
The unconditional memset() is clearly a regression caused by Ben Fennema's commit "[PATCH] UDF sync with CVS" in the old-2.6-bkcvs.git repository and should be fixed. Care to try out the included untested patch to see if it fixes the problem? (Sorry for the attachment, I am replying via web mail.)
Pekka
[PATCH] udf: fix uid/gid changing to zero
This patch fixes uid/gid changing to zero when uid/gid mount option is used for UDF. The problem appears when the overriding uid/gid matches inode. The fix is simple, instead of unconditionally zeroing buffer, we now only do it for new inodes.
The regression appeared in commit "[PATCH] UDF sync with CVS" and was first reported by Phillip Susi.
---
fs/udf/ialloc.c | 1 + fs/udf/inode.c | 9 +++++++-- fs/udf/udf_i.h | 1 + include/linux/udf_fs_i.h | 3 ++- 4 files changed, 11 insertions(+), 3 deletions(-)
fs/udf/ialloc.c +1 -0
@@ -146,6 +146,7 @@ struct inode * udf_new_inode (struct inoUDF_I_ALLOCTYPE (inode) =ICBTAG_FLAG_AD_LONG ; inode->i_mtime = inode->i_atime = inode->i_ctime =UDF_I_CRTIME (inode) =current_fs_time (inode->i_sb); +UDF_I_NEW_INODE (inode) =1 ;insert_inode_hash (inode);mark_inode_dirty (inode);up (&sbi->s_alloc_sem);
fs/udf/inode.c +7 -2
@@ -1003,6 +1003,8 @@ static void udf_fill_inode(struct inodelong convtime_usec;int offset; +UDF_I_NEW_INODE (inode) =0 ; + fe = (struct fileEntry *)bh->b_data; efe = (struct extendedFileEntry *)bh->b_data;
@@ -1307,11 +1309,14 @@ udf_update_inode(struct inode *inode, inreturn -EIO ; } -memset (bh->b_data,0x00 , inode->i_sb->s_blocksize); - fe = (struct fileEntry *)bh->b_data; efe = (struct extendedFileEntry *)bh->b_data; +if (UDF_I_NEW_INODE (inode) ==1 ) { +memset (bh->b_data,0x00 , inode->i_sb->s_blocksize); +UDF_I_NEW_INODE (inode) =0 ; + } +if (le16_to_cpu (fe->descTag.tagIdent) ==TAG_IDENT_USE ) { struct unallocSpaceEntry *use =
fs/udf/udf_i.h +1 -0
@@ -16,6 +16,7 @@ static inline struct udf_inode_info *UDF#define UDF_I_EFE (X ) ( UDF_I(X)->i_efe )#define UDF_I_USE (X ) ( UDF_I(X)->i_use )#define UDF_I_STRAT4096 (X ) ( UDF_I(X)->i_strat4096 ) +#define UDF_I_NEW_INODE (X ) ( UDF_I(X)->i_new_inode )#define UDF_I_NEXT_ALLOC_BLOCK (X ) ( UDF_I(X)->i_next_alloc_block )#define UDF_I_NEXT_ALLOC_GOAL (X ) ( UDF_I(X)->i_next_alloc_goal )#define UDF_I_CRTIME (X ) ( UDF_I(X)->i_crtime )
include/linux/udf_fs_i.h +2 -1
@@ -51,7 +51,8 @@ struct udf_inode_info unsigned i_efe : 1; unsigned i_use : 1; unsigned i_strat4096 : 1; - unsigned reserved : 26; + unsigned i_new_inode : 1; + unsigned reserved : 25;union {short_ad *i_sad;
Quoted 9 lines · Pekka Enberg
Quoted 2 lines · Phillip Susi
This patch fixes a bug in udf where it would write uid/gid = 0 to the disk for files owned by the id given with the uid=/gid= mount options.
I don't think it does. You can still get zero uid/gid written on disk if you specify uid/gid mount options when not specifying uid=forget. I am not against new features but lets fix the bug first, okay? Adding new mount options to work around buggy code is not the right solution.
My bad, I meant to remove those ifs, not just prepend them with an else. Try this one:
Subject: [PATCH] udf: fix uid/gid options and add uid/gid=ignore and forget options
This patch fixes a bug in udf where it would write uid/gid = 0 to the disk for files owned by the id given with the uid=/gid= mount options. It also adds 4 new mount options: uid/gid=forget and uid/gid=ignore. Without any options the id in core and on disk always match. Giving uid/gid=nnn specifies a default ID to be used in core when the on disk ID is -1. uid/gid=ignore forces the in core ID to allways be used no matter what the on disk ID is. uid/gid=forget forces the on disk ID to always be written out as -1.
The use of these options allows you to override ownerships on a disk or disable ownwership information from being written, allowing the media to be used portably between different computers and possibly different users without permissions issues that would require root to correct.
---
fs/udf/inode.c | 18 +++++++++++------- fs/udf/super.c | 18 +++++++++++++++++- fs/udf/udf_sb.h | 4 ++++ 3 files changed, 32 insertions(+), 8 deletions(-)
c8393f6e4fe6159fd916f3c68091e76bbfdc5fd8
fs/udf/inode.c +0 -0
@@ -1045,10 +1045,12 @@ static void udf_fill_inode(struct inode
} inode->i_uid = le32_to_cpu(fe->uid);
- if ( inode->i_uid == -1 ) inode->i_uid = UDF_SB(inode->i_sb)->s_uid;
+ if ( inode->i_uid == -1 || UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_UID_IGNORE) )
+ inode->i_uid = UDF_SB(inode->i_sb)->s_uid;
inode->i_gid = le32_to_cpu(fe->gid);
- if ( inode->i_gid == -1 ) inode->i_gid = UDF_SB(inode->i_sb)->s_gid;
+ if ( inode->i_gid == -1 || UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_GID_IGNORE) )
+ inode->i_gid = UDF_SB(inode->i_sb)->s_gid;
inode->i_nlink = le16_to_cpu(fe->fileLinkCount); if (!inode->i_nlink) @@ -1335,11 +1337,13 @@ udf_update_inode(struct inode *inode, in return err; }
- if (inode->i_uid != UDF_SB(inode->i_sb)->s_uid)
- fe->uid = cpu_to_le32(inode->i_uid);
-
- if (inode->i_gid != UDF_SB(inode->i_sb)->s_gid)
- fe->gid = cpu_to_le32(inode->i_gid);
+ if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_UID_FORGET))
+ fe->uid = cpu_to_le32(-1);
+ else fe->uid = cpu_to_le32(inode->i_uid);
+
+ if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_GID_FORGET))
+ fe->gid = cpu_to_le32(-1);
+ else fe->gid = cpu_to_le32(inode->i_gid);
udfperms = ((inode->i_mode & S_IRWXO) ) | ((inode->i_mode & S_IRWXG) << 2) |
fs/udf/super.c +0 -0
@@ -269,7 +269,7 @@ enum {
Opt_gid, Opt_uid, Opt_umask, Opt_session, Opt_lastblock, Opt_anchor, Opt_volume, Opt_partition, Opt_fileset, Opt_rootdir, Opt_utf8, Opt_iocharset,
- Opt_err
+ Opt_err, Opt_uforget, Opt_uignore, Opt_gforget, Opt_gignore
};
static match_table_t tokens = {
@@ -282,6 +282,10 @@ static match_table_t tokens = {
{Opt_adinicb, "adinicb"},
{Opt_shortad, "shortad"},
{Opt_longad, "longad"},+ {Opt_uforget, "uid=forget"},
+ {Opt_uignore, "uid=ignore"},
+ {Opt_gforget, "gid=forget"},
+ {Opt_gignore, "gid=ignore"},
{Opt_gid, "gid=%u"},
{Opt_uid, "uid=%u"},
{Opt_umask, "umask=%o"},
@@ -414,6 +418,18 @@ udf_parse_options(char *options, struct
uopt->flags |= (1 << UDF_FLAG_NLS_MAP);
break;#endif
+ case Opt_uignore:
+ uopt->flags |= (1 << UDF_FLAG_UID_IGNORE);
+ break;
+ case Opt_uforget:
+ uopt->flags |= (1 << UDF_FLAG_UID_FORGET);
+ break;
+ case Opt_gignore:
+ uopt->flags |= (1 << UDF_FLAG_GID_IGNORE);
+ break;
+ case Opt_gforget:
+ uopt->flags |= (1 << UDF_FLAG_GID_FORGET);
+ break; default: printk(KERN_ERR "udf: bad mount option \"%s\" " "or missing value\n", p);
fs/udf/udf_sb.h +0 -0
@@ -20,6 +20,10 @@
#define UDF_FLAG_VARCONV 8
#define UDF_FLAG_NLS_MAP 9
#define UDF_FLAG_UTF8 10
+#define UDF_FLAG_UID_FORGET 11 /* save -1 for uid to disk */ +#define UDF_FLAG_UID_IGNORE 12 /* use sb uid instead of on disk uid */ +#define UDF_FLAG_GID_FORGET 13 +#define UDF_FLAG_GID_IGNORE 14
#define UDF_PART_FLAG_UNALLOC_BITMAP 0x0001
#define UDF_PART_FLAG_UNALLOC_TABLE 0x0002
Signature · 1 line
1.1.3
Hi Phillip,
Quoted 9 lines · Phillip Susi
My bad, I meant to remove those ifs, not just prepend them with an else. Try this one:
c8393f6e4fe6159fd916f3c68091e76bbfdc5fd8fs/udf/inode.c +0
-0
fs/udf/inode.c static void udf_fill_inode(struct inode } · line 1049
Quoted 6 lines
fs/udf/inode.c +2
-1@@ -1045,10 +1045,12 @@ static void udf_fill_inode(struct inode } inode->i_uid =le32_to_cpu (fe->uid); -if ( inode->i_uid ==-1 ) inode->i_uid =UDF_SB (inode->i_sb)->s_uid; +if ( inode->i_uid ==-1 ||UDF_QUERY_FLAG (inode->i_sb,UDF_FLAG_UID_IGNORE ) ) + inode->i_uid =UDF_SB (inode->i_sb)->s_uid;
Formatting.
fs/udf/inode.c static void udf_fill_inode(struct inode } · line 1053
Quoted 5 lines
fs/udf/inode.c +2
-1inode->i_gid =le32_to_cpu (fe->gid); -if ( inode->i_gid ==-1 ) inode->i_gid =UDF_SB (inode->i_sb)->s_gid; +if ( inode->i_gid ==-1 ||UDF_QUERY_FLAG (inode->i_sb,UDF_FLAG_GID_IGNORE ) ) + inode->i_gid =UDF_SB (inode->i_sb)->s_gid;
Same here.
Quoted 4 lines
fs/udf/inode.c +0
-0inode->i_nlink =le16_to_cpu (fe->fileLinkCount);if (!inode->i_nlink)@@ -1335,11 +1337,13 @@ udf_update_inode(struct inode *inode, inreturn err;
fs/udf/inode.c udf_update_inode(struct inode *inode, in · line 1342
Quoted 10 lines
fs/udf/inode.c +3
-5} -if (inode->i_uid !=UDF_SB (inode->i_sb)->s_uid) - fe->uid =cpu_to_le32 (inode->i_uid); - -if (inode->i_gid !=UDF_SB (inode->i_sb)->s_gid) - fe->gid =cpu_to_le32 (inode->i_gid); +if (UDF_QUERY_FLAG (inode->i_sb,UDF_FLAG_UID_FORGET )) + fe->uid =cpu_to_le32 (-1 ); +else fe->uid =cpu_to_le32 (inode->i_uid);
This is better, but if id was -1 on disk, we're overriding it unless the forget mount option was specified. Do we want that? I think my patch is a better fix (if it works anyway) and yours should be on top of that. Did you have the chance to test it?
Also, formatting is, wrong, just make it
if (forget) fe->uid = ...; else fe->uid = ...;
fs/udf/inode.c udf_update_inode(struct inode *inode, in · line 1348
Quoted 6 lines
fs/udf/inode.c +3
-0+if (UDF_QUERY_FLAG (inode->i_sb,UDF_FLAG_GID_FORGET )) + fe->gid =cpu_to_le32 (-1 ); +else fe->gid =cpu_to_le32 (inode->i_gid); udfperms = ((inode->i_mode &S_IRWXO ) ) | ((inode->i_mode & S_IRWXG) << 2) |
Same here.
Please document the new mount options in Documentation/filesystems/udf.txt.
Pekka
Quoted 13 lines · Pekka J Enberg
Hi Phillip,
Quoted 10 lines · Phillip Susi
My bad, I meant to remove those ifs, not just prepend them with an else. Try this one:
c8393f6e4fe6159fd916f3c68091e76bbfdc5fd8fs/udf/inode.c +0
-0@@ -1045,10 +1045,12 @@ static void udf_fill_inode(struct inode }
Quoted 8 lines
fs/udf/inode.c static void udf_fill_inode(struct inode } · line 1049
Quoted 5 lines
fs/udf/inode.c +2
-1inode->i_uid =le32_to_cpu (fe->uid); -if ( inode->i_uid ==-1 ) inode->i_uid =UDF_SB (inode->i_sb)->s_uid; +if ( inode->i_uid ==-1 ||UDF_QUERY_FLAG (inode->i_sb,UDF_FLAG_UID_IGNORE ) ) + inode->i_uid =UDF_SB (inode->i_sb)->s_uid;Formatting.
What exactly do you mean by formatting?
Quoted 14 lines
fs/udf/inode.c static void udf_fill_inode(struct inode } · line 1053
Quoted 5 lines
fs/udf/inode.c +2
-1inode->i_gid =le32_to_cpu (fe->gid); -if ( inode->i_gid ==-1 ) inode->i_gid =UDF_SB (inode->i_sb)->s_gid; +if ( inode->i_gid ==-1 ||UDF_QUERY_FLAG (inode->i_sb,UDF_FLAG_GID_IGNORE ) ) + inode->i_gid =UDF_SB (inode->i_sb)->s_gid;Same here.
Quoted 6 lines
fs/udf/inode.c +0
-0inode->i_nlink =le16_to_cpu (fe->fileLinkCount);if (!inode->i_nlink)@@ -1335,11 +1337,13 @@ udf_update_inode(struct inode *inode, inreturn err; }
Quoted 11 lines
fs/udf/inode.c udf_update_inode(struct inode *inode, in · line 1342
Quoted 8 lines
fs/udf/inode.c +3
-5-if (inode->i_uid !=UDF_SB (inode->i_sb)->s_uid) - fe->uid =cpu_to_le32 (inode->i_uid); - -if (inode->i_gid !=UDF_SB (inode->i_sb)->s_gid) - fe->gid =cpu_to_le32 (inode->i_gid); +if (UDF_QUERY_FLAG (inode->i_sb,UDF_FLAG_UID_FORGET )) + fe->uid =cpu_to_le32 (-1 ); +else fe->uid =cpu_to_le32 (inode->i_uid);This is better, but if id was -1 on disk, we're overriding it unless the forget mount option was specified. Do we want that? I think my patch is a better fix (if it works anyway) and yours should be on top of that. Did you have the chance to test it?
Yes, if you chown a file that is owned by -1 on disk, you do want it to be saved back with the new id, unless you set the forget option.
Quoted 7 lines
Also, formatting is, wrong, just make it
if (forget) fe->uid = ...; else fe->uid = ...;
You mean use hard tabstops instead of two spaces to indent? Is there a way to set emacs to do that? It automatically uses the two spaces.
Quoted 11 lines
fs/udf/inode.c udf_update_inode(struct inode *inode, in · line 1348
Quoted 6 lines
fs/udf/inode.c +3
-0+if (UDF_QUERY_FLAG (inode->i_sb,UDF_FLAG_GID_FORGET )) + fe->gid =cpu_to_le32 (-1 ); +else fe->gid =cpu_to_le32 (inode->i_gid); udfperms = ((inode->i_mode &S_IRWXO ) ) | ((inode->i_mode & S_IRWXG) << 2) |Same here.
Please document the new mount options in
Documentation/filesystems/udf.txt.
Working on it.
Quoted 10 lines · Phillip Susi
Quoted 8 lines
Quoted 6 lines
inode->i_uid = le32_to_cpu(fe->uid);- if (
inode->i_uid== -1 )inode->i_uid=
UDF_SB(inode->i_sb)->s_uid;
+ if (inode->i_uid== -1 || UDF_QUERY_FLAG(inode->i_sb,UDF_FLAG_UID_IGNORE) )
+inode->i_uid= UDF_SB(inode->i_sb)->s_uid;Formatting.
What exactly do you mean by formatting?
Looks like you're using two spaces. Indentation is one tab and one tab is exactly eight characters (see Documentation/CodingStyle).
Quoted 2 lines · Phillip Susi
Yes, if you chown a file that is owned by -1 on disk, you do want it to be saved back with the new id, unless you set the forget option.
Okay, fair enough. I see akpm has taken your patch. Please make sure the mount options are documented. Thanks!
(Please note that we didn't fix the unconditional memset now, so there's
dead code in udf_update_inode(). The check for TAG_IDENT_USE will always fail.)
Pekka
Quoted 2 lines · Pekka J Enberg
Looks like you're using two spaces. Indentation is one tab and one tab is exactly eight characters (see
Documentation/CodingStyle).
Wow! 8 spaces? I always go with 4, 8 wastes way too much screen space.
Is there a handy incantation to get emacs to auto indent with a hard
tab rather than 2 spaces like it does by default?
Quoted 2 lines
Okay, fair enough. I see akpm has taken your patch. Please make sure the mount options are documented. Thanks!
I wrote up some docs for it last night.
Quoted 3 lines
(Please note that we didn't fix the unconditional memset now, so there's
dead code inudf_update_inode(). The check forTAG_IDENT_USEwill always fail.)
Hrm... I'll take a look at that tonight and see what I come up with. Probably will resubmit the patch tomorrow.
Quoted 6 lines · On Tue, 07 Mar 2006 10:49:14 -0500 Phillip Susi wrote:
Quoted 3 lines · Pekka J Enberg
Looks like you're using two spaces. Indentation is one tab and one tab is exactly eight characters (see
Documentation/CodingStyle).Wow! 8 spaces? I always go with 4, 8 wastes way too much screen space.
See Documentation/CodingStyle for reasons.
Quoted 2 lines
Is there a handy incantation to get emacs to auto indent with a hard
tab rather than 2 spaces like it does by default?
C-c . linux RET (C-c . runs c-set-style; suggestions inDocumentation/CodingStyle are outdated).