2007-02-12 14:30:31

by Dave McCracken

Subject: [PATCH 1/1] Enable fully functional truncate for hugetlbfs

This patch enables the full functionality of truncate for hugetlbfs
files. Truncate was originally limited to reducing the file size
because page faults were not supported for hugetlbfs. Now that page
faults have been implemented it is now possible to fully support
truncate.

Signed-off-by: Dave McCracken <[email protected]>

---
inode.c | 21 ++++++++++++++++++++-
1 file changed, 20 insertions(+), 1 deletion(-)

--- 2.6.20/./fs/hugetlbfs/inode.c 2007-02-04 12:44:54.000000000 -0600
+++ 2.6.20-htrunc/./fs/hugetlbfs/inode.c 2007-02-05 13:02:00.000000000 -0600
@@ -298,9 +298,10 @@ static int hugetlb_vmtruncate(struct ino
{
pgoff_t pgoff;
struct address_space *mapping = inode->i_mapping;
+ unsigned long limit;

if (offset > inode->i_size)
- return -EINVAL;
+ goto do_expand;

BUG_ON(offset & ~HPAGE_MASK);
pgoff = offset >> PAGE_SHIFT;
@@ -312,6 +313,24 @@ static int hugetlb_vmtruncate(struct ino
spin_unlock(&mapping->i_mmap_lock);
truncate_hugepages(inode, offset);
return 0;
+
+do_expand:
+ limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
+ if (limit != RLIM_INFINITY && offset > limit)
+ goto out_sig;
+ if (offset > inode->i_sb->s_maxbytes)
+ goto out_big;
+ if (hugetlb_reserve_pages(inode, inode->i_size >> HPAGE_SHIFT,
+ offset >> HPAGE_SHIFT))
+ goto out_mem;
+ i_size_write(inode, offset);
+ return 0;
+out_sig:
+ send_sig(SIGXFSZ, current, 0);
+out_big:
+ return -EFBIG;
+out_mem:
+ return -ENOMEM;
}

static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)

2007-02-12 22:15:00

by Ken Chen

Subject: Re: [PATCH 1/1] Enable fully functional truncate for hugetlbfs

On 2/12/07, Dave McCracken <[email protected]> wrote:
> This patch enables the full functionality of truncate for hugetlbfs
> files. Truncate was originally limited to reducing the file size
> because page faults were not supported for hugetlbfs. Now that page
> faults have been implemented it is now possible to fully support
> truncate.

ye!

> --- 2.6.20/./fs/hugetlbfs/inode.c 2007-02-04 12:44:54.000000000 -0600
> +++ 2.6.20-htrunc/./fs/hugetlbfs/inode.c 2007-02-05 13:02:00.000000000 -0600
> + if (hugetlb_reserve_pages(inode, inode->i_size >> HPAGE_SHIFT,
> + offset >> HPAGE_SHIFT))
> + goto out_mem;
> + i_size_write(inode, offset);

hugetlb_reserve_pages() is used to reserve pages for shared mapping
and is used only for the size of mapping at the time of mmap(). We
shouldn't call this function when expanding hugetlb file.

- Ken