int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
{
int status = 0, size_change;
struct inode *inode = dentry->d_inode;
struct super_block *sb = inode->i_sb;
struct ocfs2_super *osb = OCFS2_SB(sb);
struct buffer_head *bh = NULL;
handle_t *handle = NULL;
mlog_entry("(0x%p, '%.*s')\n", dentry,
dentry->d_name.len, dentry->d_name.name);
if (attr->ia_valid & ATTR_MODE)
mlog(0, "mode change: %d\n", attr->ia_mode);
if (attr->ia_valid & ATTR_UID)
mlog(0, "uid change: %d\n", attr->ia_uid);
if (attr->ia_valid & ATTR_GID)
mlog(0, "gid change: %d\n", attr->ia_gid);
if (attr->ia_valid & ATTR_SIZE)
mlog(0, "size change...\n");
if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
mlog(0, "time change...\n");
#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
| ATTR_GID | ATTR_UID | ATTR_MODE)
if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
return 0;
}
status = inode_change_ok(inode, attr);
if (status)
return status;
size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
if (size_change) {
status = ocfs2_rw_lock(inode, 1);
if (status < 0) {
mlog_errno(status);
goto bail;
}
}
status = ocfs2_meta_lock(inode, &bh, 1);
if (status < 0) {
if (status != -ENOENT)
mlog_errno(status);
goto bail_unlock_rw;
}
if (size_change && attr->ia_size != i_size_read(inode)) {
if (i_size_read(inode) > attr->ia_size)
status = ocfs2_truncate_file(inode, bh, attr->ia_size);
else
status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
if (status < 0) {
if (status != -ENOSPC)
mlog_errno(status);
status = -ENOSPC;
goto bail_unlock;
}
}
handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
if (IS_ERR(handle)) {
status = PTR_ERR(handle);
mlog_errno(status);
goto bail_unlock;
}
status = inode_setattr(inode, attr);
if (status < 0) {
mlog_errno(status);
goto bail_commit;
}
status = ocfs2_mark_inode_dirty(handle, inode, bh);
if (status < 0)
mlog_errno(status);
bail_commit:
ocfs2_commit_trans(osb, handle);
bail_unlock:
ocfs2_meta_unlock(inode, 1);
bail_unlock_rw:
if (size_change)
ocfs2_rw_unlock(inode, 1);
bail:
if (bh)
brelse(bh);
mlog_exit(status);
return status;
}
/* XXX care about zeroing new clusters and final partially truncated
* clusters */
errcode_t ocfs2_truncate_full(ocfs2_filesys *fs, uint64_t ino,
uint64_t new_i_size,
errcode_t (*free_clusters)(ocfs2_filesys *fs,
uint32_t len,
uint64_t start,
void *free_data),
void *free_data)
{
errcode_t ret;
uint32_t new_clusters;
ocfs2_cached_inode *ci = NULL;
ret = ocfs2_read_cached_inode(fs, ino, &ci);
if (ret)
goto out;
/* in case of dio crashed, force do trucate since blocks may already
* be allocated
*/
if (ci->ci_inode->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) {
ci->ci_inode->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
ci->ci_inode->i_dio_orphaned_slot = 0;
new_i_size = ci->ci_inode->i_size;
goto truncate;
}
if (ci->ci_inode->i_size == new_i_size)
goto out;
if (ci->ci_inode->i_size < new_i_size) {
ret = ocfs2_extend_file(fs, ino, new_i_size);
goto out;
}
truncate:
if ((S_ISLNK(ci->ci_inode->i_mode) && !ci->ci_inode->i_clusters) ||
(ci->ci_inode->i_dyn_features & OCFS2_INLINE_DATA_FL))
ret = ocfs2_truncate_inline(fs, ino, new_i_size);
else {
ret = ocfs2_zero_tail_and_truncate_full(fs, ci, new_i_size,
&new_clusters,
free_clusters,
free_data);
if (ret)
goto out;
ci->ci_inode->i_clusters = new_clusters;
/* now all the clusters and extent blocks are freed.
* only when the file's content is empty, should the tree depth
* change.
*/
if (new_clusters == 0)
ci->ci_inode->id2.i_list.l_tree_depth = 0;
ci->ci_inode->i_size = new_i_size;
ret = ocfs2_write_cached_inode(fs, ci);
}
if (!ret && !new_i_size && ci->ci_inode->i_refcount_loc &&
(ci->ci_inode->i_dyn_features & OCFS2_HAS_REFCOUNT_FL))
ret = ocfs2_detach_refcount_tree(fs, ino, ci->ci_inode->i_refcount_loc);
out:
if (ci)
ocfs2_free_cached_inode(fs, ci);
return ret;
}
static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
loff_t *ppos,
size_t count,
int appending,
int *direct_io)
{
int ret = 0, meta_level = appending;
struct inode *inode = dentry->d_inode;
u32 clusters;
loff_t newsize, saved_pos;
/*
* We sample i_size under a read level meta lock to see if our write
* is extending the file, if it is we back off and get a write level
* meta lock.
*/
for(;;) {
ret = ocfs2_meta_lock(inode, NULL, meta_level);
if (ret < 0) {
meta_level = -1;
mlog_errno(ret);
goto out;
}
/* Clear suid / sgid if necessary. We do this here
* instead of later in the write path because
* remove_suid() calls ->setattr without any hint that
* we may have already done our cluster locking. Since
* ocfs2_setattr() *must* take cluster locks to
* proceeed, this will lead us to recursively lock the
* inode. There's also the dinode i_size state which
* can be lost via setattr during extending writes (we
* set inode->i_size at the end of a write. */
if (should_remove_suid(dentry)) {
if (meta_level == 0) {
ocfs2_meta_unlock(inode, meta_level);
meta_level = 1;
continue;
}
ret = ocfs2_write_remove_suid(inode);
if (ret < 0) {
mlog_errno(ret);
goto out_unlock;
}
}
/* work on a copy of ppos until we're sure that we won't have
* to recalculate it due to relocking. */
if (appending) {
saved_pos = i_size_read(inode);
mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
} else {
saved_pos = *ppos;
}
if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
loff_t end = saved_pos + count;
/*
* Skip the O_DIRECT checks if we don't need
* them.
*/
if (!direct_io || !(*direct_io))
break;
/*
* Allowing concurrent direct writes means
* i_size changes wouldn't be synchronized, so
* one node could wind up truncating another
* nodes writes.
*/
if (end > i_size_read(inode)) {
*direct_io = 0;
break;
}
/*
* We don't fill holes during direct io, so
* check for them here. If any are found, the
* caller will have to retake some cluster
* locks and initiate the io as buffered.
*/
ret = ocfs2_check_range_for_holes(inode, saved_pos,
count);
if (ret == 1) {
*direct_io = 0;
ret = 0;
} else if (ret < 0)
mlog_errno(ret);
break;
}
/*
* The rest of this loop is concerned with legacy file
* systems which don't support sparse files.
*/
newsize = count + saved_pos;
mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
(long long) saved_pos, (long long) newsize,
(long long) i_size_read(inode));
/* No need for a higher level metadata lock if we're
* never going past i_size. */
if (newsize <= i_size_read(inode))
break;
if (meta_level == 0) {
ocfs2_meta_unlock(inode, meta_level);
meta_level = 1;
continue;
}
spin_lock(&OCFS2_I(inode)->ip_lock);
clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) -
OCFS2_I(inode)->ip_clusters;
spin_unlock(&OCFS2_I(inode)->ip_lock);
mlog(0, "Writing at EOF, may need more allocation: "
"i_size = %lld, newsize = %lld, need %u clusters\n",
(long long) i_size_read(inode), (long long) newsize,
clusters);
/* We only want to continue the rest of this loop if
* our extend will actually require more
* allocation. */
if (!clusters)
break;
ret = ocfs2_extend_file(inode, NULL, newsize, count);
if (ret < 0) {
if (ret != -ENOSPC)
mlog_errno(ret);
goto out_unlock;
}
break;
}
if (appending)
*ppos = saved_pos;
out_unlock:
ocfs2_meta_unlock(inode, meta_level);
out:
return ret;
}