[PATCH 03/11] fs/ntfs3: Fix ntfs_create_inode()

From: Konstantin Komarov
Date: Wed Feb 15 2023 - 08:35:48 EST


Previous variant creates an inode that requires update the parent directory
(ea_packed_size). Operations in ntfs_create_inode have been rearranged
so we insert new directory entry with correct ea_packed_size and
new created inode does not require update it's parent directory.

Signed-off-by: Konstantin Komarov <almaz.alexandrovich@xxxxxxxxxxxxxxxxxxxx>
---
 fs/ntfs3/file.c    |  2 +-
 fs/ntfs3/inode.c   | 83 ++++++++++++++++++++++++----------------------
 fs/ntfs3/ntfs_fs.h |  2 +-
 fs/ntfs3/xattr.c   | 20 ++++++-----
 4 files changed, 58 insertions(+), 49 deletions(-)

diff --git a/fs/ntfs3/file.c b/fs/ntfs3/file.c
index 9cef189fc0c5..df7b76d1c127 100644
--- a/fs/ntfs3/file.c
+++ b/fs/ntfs3/file.c
@@ -703,7 +703,7 @@ int ntfs3_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
     }

     if (ia_valid & (ATTR_UID | ATTR_GID | ATTR_MODE))
-        ntfs_save_wsl_perm(inode);
+        ntfs_save_wsl_perm(inode, NULL);
     mark_inode_dirty(inode);
 out:
     return err;
diff --git a/fs/ntfs3/inode.c b/fs/ntfs3/inode.c
index 51e342ad79fd..752ad17685c0 100644
--- a/fs/ntfs3/inode.c
+++ b/fs/ntfs3/inode.c
@@ -1320,8 +1320,7 @@ struct inode *ntfs_create_inode(struct user_namespace *mnt_userns,
     inode_init_owner(mnt_userns, inode, dir, mode);
     mode = inode->i_mode;

-    inode->i_atime = inode->i_mtime = inode->i_ctime = ni->i_crtime =
-        current_time(inode);
+    ni->i_crtime = current_time(inode);

     rec = ni->mi.mrec;
     rec->hard_links = cpu_to_le16(1);
@@ -1362,10 +1361,9 @@ struct inode *ntfs_create_inode(struct user_namespace *mnt_userns,
     attr->res.data_size = cpu_to_le32(dsize);

     std5->cr_time = std5->m_time = std5->c_time = std5->a_time =
-        kernel2nt(&inode->i_atime);
+        kernel2nt(&ni->i_crtime);

-    ni->std_fa = fa;
-    std5->fa = fa;
+    std5->fa = ni->std_fa = fa;

     attr = Add2Ptr(attr, asize);

@@ -1564,11 +1562,15 @@ struct inode *ntfs_create_inode(struct user_namespace *mnt_userns,
             }

             asize = SIZEOF_NONRESIDENT + ALIGN(err, 8);
+            /* Write non resident data. */
+            err = ntfs_sb_write_run(sbi, &ni->file.run, 0, rp,
+                        nsize, 0);
+            if (err)
+                goto out5;
         } else {
             attr->res.data_off = SIZEOF_RESIDENT_LE;
             attr->res.data_size = cpu_to_le32(nsize);
             memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), rp, nsize);
-            nsize = 0;
         }
         /* Size of symlink equals the length of input string. */
         inode->i_size = size;
@@ -1589,19 +1591,8 @@ struct inode *ntfs_create_inode(struct user_namespace *mnt_userns,
     rec->used = cpu_to_le32(PtrOffset(rec, attr) + 8);
     rec->next_attr_id = cpu_to_le16(aid);

-    /* Step 2: Add new name in index. */
-    err = indx_insert_entry(&dir_ni->dir, dir_ni, new_de, sbi, fnd, 0);
-    if (err)
-        goto out6;
-
-    /* Unlock parent directory before ntfs_init_acl. */
-    if (!fnd)
-        ni_unlock(dir_ni);
-
     inode->i_generation = le16_to_cpu(rec->seq);

-    dir->i_mtime = dir->i_ctime = inode->i_atime;
-
     if (S_ISDIR(mode)) {
         inode->i_op = &ntfs_dir_inode_operations;
         inode->i_fop = &ntfs_dir_operations;
@@ -1626,41 +1617,58 @@ struct inode *ntfs_create_inode(struct user_namespace *mnt_userns,
     if (!S_ISLNK(mode) && (sb->s_flags & SB_POSIXACL)) {
         err = ntfs_init_acl(mnt_userns, inode, dir);
         if (err)
-            goto out7;
+            goto out5;
     } else
 #endif
     {
         inode->i_flags |= S_NOSEC;
     }

-    /* Write non resident data. */
-    if (nsize) {
-        err = ntfs_sb_write_run(sbi, &ni->file.run, 0, rp, nsize, 0);
-        if (err)
-            goto out7;
+    /*
+     * ntfs_init_acl and ntfs_save_wsl_perm update extended attribute.
+     * The packed size of extended attribute is stored in direntry too.
+     * 'fname' here points to inside new_de.
+     */
+    ntfs_save_wsl_perm(inode, &fname->dup.ea_size);
+
+    /*
+     * update ea_size in file_name attribute too.
+     * Use ni_find_attr cause layout of MFT record may be changed
+     * in ntfs_init_acl and ntfs_save_wsl_perm.
+     */
+    attr = ni_find_attr(ni, NULL, NULL, ATTR_NAME, NULL, 0, NULL, NULL);
+    if (attr) {
+        struct ATTR_FILE_NAME *fn;
+
+        fn = resident_data_ex(attr, SIZEOF_ATTRIBUTE_FILENAME);
+        if (fn)
+            fn->dup.ea_size = fname->dup.ea_size;
     }

+    /* We do not need to update parent directory later */
+    ni->ni_flags &= ~NI_FLAG_UPDATE_PARENT;
+
+    /* Step 2: Add new name in index. */
+    err = indx_insert_entry(&dir_ni->dir, dir_ni, new_de, sbi, fnd, 0);
+    if (err)
+        goto out6;
+
     /*
      * Call 'd_instantiate' after inode->i_op is set
      * but before finish_open.
      */
     d_instantiate(dentry, inode);

-    ntfs_save_wsl_perm(inode);
+    /* Set original time. inode times (i_ctime) may be changed in ntfs_init_acl. */
+    inode->i_atime = inode->i_mtime = inode->i_ctime = dir->i_mtime =
+        dir->i_ctime = ni->i_crtime;
+
     mark_inode_dirty(dir);
     mark_inode_dirty(inode);

     /* Normal exit. */
     goto out2;

-out7:
-
-    /* Undo 'indx_insert_entry'. */
-    if (!fnd)
-        ni_lock_dir(dir_ni);
-    indx_delete_entry(&dir_ni->dir, dir_ni, new_de + 1,
-              le16_to_cpu(new_de->key_size), sbi);
-    /* ni_unlock(dir_ni); will be called later. */
 out6:
     if (rp_inserted)
         ntfs_remove_reparse(sbi, IO_REPARSE_TAG_SYMLINK, &new_de->ref);
@@ -1682,11 +1690,11 @@ struct inode *ntfs_create_inode(struct user_namespace *mnt_userns,
     kfree(rp);

 out1:
-    if (err) {
-        if (!fnd)
-            ni_unlock(dir_ni);
+    if (!fnd)
+        ni_unlock(dir_ni);
+
+    if (err)
         return ERR_PTR(err);
-    }

     unlock_new_inode(inode);

@@ -1783,9 +1791,6 @@ void ntfs_evict_inode(struct inode *inode)
 {
     truncate_inode_pages_final(&inode->i_data);

-    if (inode->i_nlink)
-        _ni_write_inode(inode, inode_needs_sync(inode));
-
     invalidate_inode_buffers(inode);
     clear_inode(inode);

diff --git a/fs/ntfs3/ntfs_fs.h b/fs/ntfs3/ntfs_fs.h
index 556b76f526cb..73a639716b45 100644
--- a/fs/ntfs3/ntfs_fs.h
+++ b/fs/ntfs3/ntfs_fs.h
@@ -870,7 +870,7 @@ int ntfs_acl_chmod(struct user_namespace *mnt_userns, struct dentry *dentry);
 ssize_t ntfs_listxattr(struct dentry *dentry, char *buffer, size_t size);
 extern const struct xattr_handler *ntfs_xattr_handlers[];

-int ntfs_save_wsl_perm(struct inode *inode);
+int ntfs_save_wsl_perm(struct inode *inode, __le16 *ea_size);
 void ntfs_get_wsl_perm(struct inode *inode);

 /* globals from lznt.c */
diff --git a/fs/ntfs3/xattr.c b/fs/ntfs3/xattr.c
index c42fbc56eb39..e7a66225361d 100644
--- a/fs/ntfs3/xattr.c
+++ b/fs/ntfs3/xattr.c
@@ -296,7 +296,8 @@ static int ntfs_get_ea(struct inode *inode, const char *name, size_t name_len,

 static noinline int ntfs_set_ea(struct inode *inode, const char *name,
                 size_t name_len, const void *value,
-                size_t val_size, int flags, bool locked)
+                size_t val_size, int flags, bool locked,
+                __le16 *ea_size)
 {
     struct ntfs_inode *ni = ntfs_i(inode);
     struct ntfs_sb_info *sbi = ni->mi.sbi;
@@ -504,6 +505,8 @@ static noinline int ntfs_set_ea(struct inode *inode, const char *name,

     if (ea_info.size_pack != size_pack)
         ni->ni_flags |= NI_FLAG_UPDATE_PARENT;
+    if (ea_size)
+        *ea_size = ea_info.size_pack;
     mark_inode_dirty(&ni->vfs_inode);

 out:
@@ -633,7 +636,7 @@ static noinline int ntfs_set_acl_ex(struct user_namespace *mnt_userns,
         flags = 0;
     }

-    err = ntfs_set_ea(inode, name, name_len, value, size, flags, 0);
+    err = ntfs_set_ea(inode, name, name_len, value, size, flags, 0, NULL);
     if (err == -ENODATA && !size)
         err = 0; /* Removing non existed xattr. */
     if (!err) {
@@ -923,7 +926,8 @@ static noinline int ntfs_setxattr(const struct xattr_handler *handler,
     }

     /* Deal with NTFS extended attribute. */
-    err = ntfs_set_ea(inode, name, strlen(name), value, size, flags, 0);
+    err = ntfs_set_ea(inode, name, strlen(name), value, size, flags, 0,
+              NULL);

 out:
     inode->i_ctime = current_time(inode);
@@ -937,7 +941,7 @@ static noinline int ntfs_setxattr(const struct xattr_handler *handler,
  *
  * save uid/gid/mode in xattr
  */
-int ntfs_save_wsl_perm(struct inode *inode)
+int ntfs_save_wsl_perm(struct inode *inode, __le16 *ea_size)
 {
     int err;
     __le32 value;
@@ -946,26 +950,26 @@ int ntfs_save_wsl_perm(struct inode *inode)
     ni_lock(ni);
     value = cpu_to_le32(i_uid_read(inode));
     err = ntfs_set_ea(inode, "$LXUID", sizeof("$LXUID") - 1, &value,
-              sizeof(value), 0, true); /* true == already locked. */
+              sizeof(value), 0, true, ea_size);
     if (err)
         goto out;

     value = cpu_to_le32(i_gid_read(inode));
     err = ntfs_set_ea(inode, "$LXGID", sizeof("$LXGID") - 1, &value,
-              sizeof(value), 0, true);
+              sizeof(value), 0, true, ea_size);
     if (err)
         goto out;

     value = cpu_to_le32(inode->i_mode);
     err = ntfs_set_ea(inode, "$LXMOD", sizeof("$LXMOD") - 1, &value,
-              sizeof(value), 0, true);
+              sizeof(value), 0, true, ea_size);
     if (err)
         goto out;

     if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
         value = cpu_to_le32(inode->i_rdev);
         err = ntfs_set_ea(inode, "$LXDEV", sizeof("$LXDEV") - 1, &value,
-                  sizeof(value), 0, true);
+                  sizeof(value), 0, true, ea_size);
         if (err)
             goto out;
     }
--
2.34.1