static int ocfs2_readpage(struct file *file, struct page *page)
{
struct inode *inode = page->mapping->host;
loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
int ret, unlock = 1;
mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0));
ret = ocfs2_meta_lock_with_page(inode, NULL, 0, page);
if (ret != 0) {
if (ret == AOP_TRUNCATED_PAGE)
unlock = 0;
mlog_errno(ret);
goto out;
}
if (down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem) == 0) {
ret = AOP_TRUNCATED_PAGE;
goto out_meta_unlock;
}
/*
* i_size might have just been updated as we grabed the meta lock. We
* might now be discovering a truncate that hit on another node.
* block_read_full_page->get_block freaks out if it is asked to read
* beyond the end of a file, so we check here. Callers
* (generic_file_read, vm_ops->fault) are clever enough to check i_size
* and notice that the page they just read isn't needed.
*
* XXX sys_readahead() seems to get that wrong?
*/
if (start >= i_size_read(inode)) {
zero_user_page(page, 0, PAGE_SIZE, KM_USER0);
SetPageUptodate(page);
ret = 0;
goto out_alloc;
}
ret = ocfs2_data_lock_with_page(inode, 0, page);
if (ret != 0) {
if (ret == AOP_TRUNCATED_PAGE)
unlock = 0;
mlog_errno(ret);
goto out_alloc;
}
ret = block_read_full_page(page, ocfs2_get_block);
unlock = 0;
ocfs2_data_unlock(inode, 0);
out_alloc:
up_read(&OCFS2_I(inode)->ip_alloc_sem);
out_meta_unlock:
ocfs2_meta_unlock(inode, 0);
out:
if (unlock)
unlock_page(page);
mlog_exit(ret);
return ret;
}
/*
* A tail_to_skip value > 0 indicates that we're being called from
* ocfs2_file_aio_write(). This has the following implications:
*
* - we don't want to update i_size
* - di_bh will be NULL, which is fine because it's only used in the
* case where we want to update i_size.
* - ocfs2_zero_extend() will then only be filling the hole created
* between i_size and the start of the write.
*/
static int ocfs2_extend_file(struct inode *inode,
struct buffer_head *di_bh,
u64 new_i_size,
size_t tail_to_skip)
{
int ret = 0;
u32 clusters_to_add = 0;
BUG_ON(!tail_to_skip && !di_bh);
/* setattr sometimes calls us like this. */
if (new_i_size == 0)
goto out;
if (i_size_read(inode) == new_i_size)
goto out;
BUG_ON(new_i_size < i_size_read(inode));
if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
BUG_ON(tail_to_skip != 0);
goto out_update_size;
}
clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) -
OCFS2_I(inode)->ip_clusters;
/*
* protect the pages that ocfs2_zero_extend is going to be
* pulling into the page cache.. we do this before the
* metadata extend so that we don't get into the situation
* where we've extended the metadata but can't get the data
* lock to zero.
*/
ret = ocfs2_data_lock(inode, 1);
if (ret < 0) {
mlog_errno(ret);
goto out;
}
if (clusters_to_add) {
ret = ocfs2_extend_allocation(inode, clusters_to_add);
if (ret < 0) {
mlog_errno(ret);
goto out_unlock;
}
}
/*
* Call this even if we don't add any clusters to the tree. We
* still need to zero the area between the old i_size and the
* new i_size.
*/
ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip);
if (ret < 0) {
mlog_errno(ret);
goto out_unlock;
}
out_update_size:
if (!tail_to_skip) {
/* We're being called from ocfs2_setattr() which wants
* us to update i_size */
ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
if (ret < 0)
mlog_errno(ret);
}
out_unlock:
if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
ocfs2_data_unlock(inode, 1);
out:
return ret;
}
static int ocfs2_truncate_file(struct inode *inode,
struct buffer_head *di_bh,
u64 new_i_size)
{
int status = 0;
struct ocfs2_dinode *fe = NULL;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
struct ocfs2_truncate_context *tc = NULL;
mlog_entry("(inode = %llu, new_i_size = %llu\n",
(unsigned long long)OCFS2_I(inode)->ip_blkno,
(unsigned long long)new_i_size);
unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
truncate_inode_pages(inode->i_mapping, new_i_size);
fe = (struct ocfs2_dinode *) di_bh->b_data;
if (!OCFS2_IS_VALID_DINODE(fe)) {
OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
status = -EIO;
goto bail;
}
mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
"Inode %llu, inode i_size = %lld != di "
"i_size = %llu, i_flags = 0x%x\n",
(unsigned long long)OCFS2_I(inode)->ip_blkno,
i_size_read(inode),
(unsigned long long)le64_to_cpu(fe->i_size),
le32_to_cpu(fe->i_flags));
if (new_i_size > le64_to_cpu(fe->i_size)) {
mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
(unsigned long long)le64_to_cpu(fe->i_size),
(unsigned long long)new_i_size);
status = -EINVAL;
mlog_errno(status);
goto bail;
}
mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
(unsigned long long)le64_to_cpu(fe->i_blkno),
(unsigned long long)le64_to_cpu(fe->i_size),
(unsigned long long)new_i_size);
/* lets handle the simple truncate cases before doing any more
* cluster locking. */
if (new_i_size == le64_to_cpu(fe->i_size))
goto bail;
/* This forces other nodes to sync and drop their pages. Do
* this even if we have a truncate without allocation change -
* ocfs2 cluster sizes can be much greater than page size, so
* we have to truncate them anyway. */
status = ocfs2_data_lock(inode, 1);
if (status < 0) {
mlog_errno(status);
goto bail;
}
/* alright, we're going to need to do a full blown alloc size
* change. Orphan the inode so that recovery can complete the
* truncate if necessary. This does the task of marking
* i_size. */
status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
if (status < 0) {
mlog_errno(status);
goto bail_unlock_data;
}
status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
if (status < 0) {
mlog_errno(status);
goto bail_unlock_data;
}
status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
if (status < 0) {
mlog_errno(status);
goto bail_unlock_data;
}
/* TODO: orphan dir cleanup here. */
bail_unlock_data:
ocfs2_data_unlock(inode, 1);
bail:
mlog_exit(status);
return status;
}
