Skip to contentSkip to conversation

Revert "ubifs: xattr: Don't operate on deleted inodes"

PATCH In review Richard Weinberger 2 participants 5 messages Sep 16, 2018 – Sep 21, 2018

Subsystems (3)
lore
Share & tools
lore

Richard Weinberger

author maintainer
#0 lore raw

PATCH 1/1 Revert "ubifs: xattr: Don't operate on deleted inodes"

This reverts commit 11a6fc3dc743e22fb50f2196ec55bee5140d3c52. UBIFS wants to assert that xattr operations are only issued on files with positive link count. The said patch made this operations return -ENOENT for unlinked files such that the asserts will no longer trigger. This was wrong since xattr operations are perfectly fine on unlinked files.
Instead the assertions need to be fixed/removed.

Fixes 11a6fc3dc743 ("ubifs: xattr: Don't operate on deleted inodes")
---
 fs/ubifs/xattr.c | 24 ------------------------
 1 file changed, 24 deletions(-)
fs/ubifs/xattr.c +0 -24
index 61afdfee4b28..f5ad1ede7990 100644
@@ -152,12 +152,6 @@ static int create_xattr(struct ubifs_info *c, struct inode *host,
 	ui->data_len = size;
 
 	mutex_lock(&host_ui->ui_mutex);
-
-	if (!host->i_nlink) {
-		err = -ENOENT;
-		goto out_noent;
-	}
-
 	host->i_ctime = current_time(host);
 	host_ui->xattr_cnt += 1;
 	host_ui->xattr_size += CALC_DENT_SIZE(fname_len(nm));
@@ -190,7 +184,6 @@ static int create_xattr(struct ubifs_info *c, struct inode *host,
 	host_ui->xattr_size -= CALC_XATTR_BYTES(size);
 	host_ui->xattr_names -= fname_len(nm);
 	host_ui->flags &= ~UBIFS_CRYPT_FL;
-out_noent:
 	mutex_unlock(&host_ui->ui_mutex);
 out_free:
 	make_bad_inode(inode);
@@ -242,12 +235,6 @@ static int change_xattr(struct ubifs_info *c, struct inode *host,
 	mutex_unlock(&ui->ui_mutex);
 
 	mutex_lock(&host_ui->ui_mutex);
-
-	if (!host->i_nlink) {
-		err = -ENOENT;
-		goto out_noent;
-	}
-
 	host->i_ctime = current_time(host);
 	host_ui->xattr_size -= CALC_XATTR_BYTES(old_size);
 	host_ui->xattr_size += CALC_XATTR_BYTES(size);
@@ -269,7 +256,6 @@ static int change_xattr(struct ubifs_info *c, struct inode *host,
 out_cancel:
 	host_ui->xattr_size -= CALC_XATTR_BYTES(size);
 	host_ui->xattr_size += CALC_XATTR_BYTES(old_size);
-out_noent:
 	mutex_unlock(&host_ui->ui_mutex);
 	make_bad_inode(inode);
 out_free:
@@ -496,12 +482,6 @@ static int remove_xattr(struct ubifs_info *c, struct inode *host,
 		return err;
 
 	mutex_lock(&host_ui->ui_mutex);
-
-	if (!host->i_nlink) {
-		err = -ENOENT;
-		goto out_noent;
-	}
-
 	host->i_ctime = current_time(host);
 	host_ui->xattr_cnt -= 1;
 	host_ui->xattr_size -= CALC_DENT_SIZE(fname_len(nm));
@@ -521,7 +501,6 @@ static int remove_xattr(struct ubifs_info *c, struct inode *host,
 	host_ui->xattr_size += CALC_DENT_SIZE(fname_len(nm));
 	host_ui->xattr_size += CALC_XATTR_BYTES(ui->data_len);
 	host_ui->xattr_names += fname_len(nm);
-out_noent:
 	mutex_unlock(&host_ui->ui_mutex);
 	ubifs_release_budget(c, &req);
 	make_bad_inode(inode);
@@ -561,9 +540,6 @@ static int ubifs_xattr_remove(struct inode *host, const char *name)
 
 	ubifs_assert(c, inode_is_locked(host));
 
-	if (!host->i_nlink)
-		return -ENOENT;
-
 	if (fname_len(&nm) > UBIFS_MAX_NLEN)
 		return -ENAMETOOLONG;
 
Signature · 1 line
2.19.0
Tested-by on v1 1/1
#1 lore raw
Quoted 9 lines · Richard Weinberger

This reverts commit 11a6fc3dc743e22fb50f2196ec55bee5140d3c52. UBIFS wants to assert that xattr operations are only issued on files with positive link count. The said patch made this operations return -ENOENT for unlinked files such that the asserts will no longer trigger. This was wrong since xattr operations are perfectly fine on unlinked files.
Instead the assertions need to be fixed/removed.

Quoted 4 lines
Fixes 11a6fc3dc743 ("ubifs: xattr: Don't operate on deleted inodes")

The bad commit 11a6fc3dc743 was included in 4.18.7, and broke our systems which use overlayfs on top of ubifs.

This patch didn't apply cleanly for me to 4.18.8 (an encoding issue?), but I did a revert of the offending commit which did the trick.

Cheers,

Joel

Quoted 84 lines
---
 fs/ubifs/xattr.c | 24 ------------------------
 1 file changed, 24 deletions(-)
fs/ubifs/xattr.c +0 -25
index 61afdfee4b28..f5ad1ede7990 100644
@@ -152,12 +152,6 @@ static int create_xattr(struct ubifs_info *c, struct inode *host,
        ui->data_len = size;
 
        mutex_lock(&host_ui->ui_mutex);
-
-       if (!host->i_nlink) {
-               err = -ENOENT;
-               goto out_noent;
-       }
-
        host->i_ctime = current_time(host);
        host_ui->xattr_cnt += 1;
        host_ui->xattr_size += CALC_DENT_SIZE(fname_len(nm));
@@ -190,7 +184,6 @@ static int create_xattr(struct ubifs_info *c, struct inode *host,
        host_ui->xattr_size -= CALC_XATTR_BYTES(size);
        host_ui->xattr_names -= fname_len(nm);
        host_ui->flags &= ~UBIFS_CRYPT_FL;
-out_noent:
        mutex_unlock(&host_ui->ui_mutex);
 out_free:
        make_bad_inode(inode);
@@ -242,12 +235,6 @@ static int change_xattr(struct ubifs_info *c, struct inode *host,
        mutex_unlock(&ui->ui_mutex);
 
        mutex_lock(&host_ui->ui_mutex);
-
-       if (!host->i_nlink) {
-               err = -ENOENT;
-               goto out_noent;
-       }
-
        host->i_ctime = current_time(host);
        host_ui->xattr_size -= CALC_XATTR_BYTES(old_size);
        host_ui->xattr_size += CALC_XATTR_BYTES(size);
@@ -269,7 +256,6 @@ static int change_xattr(struct ubifs_info *c, struct inode *host,
 out_cancel:
        host_ui->xattr_size -= CALC_XATTR_BYTES(size);
        host_ui->xattr_size += CALC_XATTR_BYTES(old_size);
-out_noent:
        mutex_unlock(&host_ui->ui_mutex);
        make_bad_inode(inode);
 out_free:
@@ -496,12 +482,6 @@ static int remove_xattr(struct ubifs_info *c, struct inode *host,
                return err;
 
        mutex_lock(&host_ui->ui_mutex);
-
-       if (!host->i_nlink) {
-               err = -ENOENT;
-               goto out_noent;
-       }
-
        host->i_ctime = current_time(host);
        host_ui->xattr_cnt -= 1;
        host_ui->xattr_size -= CALC_DENT_SIZE(fname_len(nm));
@@ -521,7 +501,6 @@ static int remove_xattr(struct ubifs_info *c, struct inode *host,
        host_ui->xattr_size += CALC_DENT_SIZE(fname_len(nm));
        host_ui->xattr_size += CALC_XATTR_BYTES(ui->data_len);
        host_ui->xattr_names += fname_len(nm);
-out_noent:
        mutex_unlock(&host_ui->ui_mutex);
        ubifs_release_budget(c, &req);
        make_bad_inode(inode);
@@ -561,9 +540,6 @@ static int ubifs_xattr_remove(struct inode *host, const char *name)
 
        ubifs_assert(c, inode_is_locked(host));
 
-       if (!host->i_nlink)
-               return -ENOENT;
-
        if (fname_len(&nm) > UBIFS_MAX_NLEN)
                return -ENAMETOOLONG;
 
--

2.19.0

author maintainer
#2 lore raw

Am Donnerstag, 20. September 2018, 04:30:41 CEST schrieb Joel Stanley:

Quoted 15 lines · Richard Weinberger
Quoted 9 lines

This reverts commit 11a6fc3dc743e22fb50f2196ec55bee5140d3c52. UBIFS wants to assert that xattr operations are only issued on files with positive link count. The said patch made this operations return -ENOENT for unlinked files such that the asserts will no longer trigger. This was wrong since xattr operations are perfectly fine on unlinked files.
Instead the assertions need to be fixed/removed.

Quoted 4 lines
Fixes 11a6fc3dc743 ("ubifs: xattr: Don't operate on deleted inodes")
Quoted 7 lines

The bad commit 11a6fc3dc743 was included in 4.18.7, and broke our systems which use overlayfs on top of ubifs.

This patch didn't apply cleanly for me to 4.18.8 (an encoding issue?), but I did a revert of the offending commit which did the trick.

Thanks a lot for testing and sorry for the breakage. :-(

Thanks,
//richard

#3 lore raw
Quoted 20 lines · Richard Weinberger

Am Donnerstag, 20. September 2018, 04:30:41 CEST schrieb Joel Stanley:

Quoted 18 lines · Richard Weinberger
Quoted 9 lines

This reverts commit 11a6fc3dc743e22fb50f2196ec55bee5140d3c52. UBIFS wants to assert that xattr operations are only issued on files with positive link count. The said patch made this operations return -ENOENT for unlinked files such that the asserts will no longer trigger. This was wrong since xattr operations are perfectly fine on unlinked files.
Instead the assertions need to be fixed/removed.

Quoted 4 lines
Fixes 11a6fc3dc743 ("ubifs: xattr: Don't operate on deleted inodes")

The bad commit 11a6fc3dc743 was included in 4.18.7, and broke our systems which use overlayfs on top of ubifs.

Quoted 6 lines
Quoted 4 lines

This patch didn't apply cleanly for me to 4.18.8 (an encoding issue?), but I did a revert of the offending commit which did the trick.

Thanks a lot for testing and sorry for the breakage. :-(

No worries. How do you plan to get this fix in? It would be good to get it in master asap, so we can get it backported in the next wave of stable releases.

Cheers,

Joel

author maintainer
#4 lore raw

Am Freitag, 21. September 2018, 03:42:25 CEST schrieb Joel Stanley:

Quoted 24 lines · Richard Weinberger
Quoted 23 lines

Am Donnerstag, 20. September 2018, 04:30:41 CEST schrieb Joel Stanley:

Quoted 21 lines · Richard Weinberger
Quoted 9 lines

This reverts commit 11a6fc3dc743e22fb50f2196ec55bee5140d3c52. UBIFS wants to assert that xattr operations are only issued on files with positive link count. The said patch made this operations return -ENOENT for unlinked files such that the asserts will no longer trigger. This was wrong since xattr operations are perfectly fine on unlinked files.
Instead the assertions need to be fixed/removed.

Quoted 4 lines
Fixes 11a6fc3dc743 ("ubifs: xattr: Don't operate on deleted inodes")

The bad commit 11a6fc3dc743 was included in 4.18.7, and broke our systems which use overlayfs on top of ubifs.

This patch didn't apply cleanly for me to 4.18.8 (an encoding issue?), but I did a revert of the offending commit which did the trick.

Quoted 7 lines
Quoted 3 lines
Quoted 1 line

Thanks a lot for testing and sorry for the breakage. :-(

No worries. How do you plan to get this fix in? It would be good to get it in master asap, so we can get it backported in the next wave of stable releases.

Since yesterday it is in -next, I'll send a pull request to Linus^WGreg later this day.

Thanks,
//richard