#include "ocfs2.h"
+#include "aops.h"
#include "dlmglue.h"
#include "file.h"
#include "inode.h"
#include "mmap.h"
-static struct page *ocfs2_nopage(struct vm_area_struct * area,
- unsigned long address,
- int *type)
+static inline int ocfs2_vm_op_block_sigs(sigset_t *blocked, sigset_t *oldset)
+{
+ /* The best way to deal with signals in the vm path is
+ * to block them upfront, rather than allowing the
+ * locking paths to return -ERESTARTSYS. */
+ sigfillset(blocked);
+
+ /* We should technically never get a bad return value
+ * from sigprocmask */
+ return sigprocmask(SIG_BLOCK, blocked, oldset);
+}
+
+static inline int ocfs2_vm_op_unblock_sigs(sigset_t *oldset)
+{
+ return sigprocmask(SIG_SETMASK, oldset, NULL);
+}
+
+static int ocfs2_fault(struct vm_area_struct *area, struct vm_fault *vmf)
{
- struct inode *inode = area->vm_file->f_dentry->d_inode;
- struct page *page = NOPAGE_SIGBUS;
sigset_t blocked, oldset;
+ int error, ret;
+
+ mlog_entry("(area=%p, page offset=%lu)\n", area, vmf->pgoff);
+
+ error = ocfs2_vm_op_block_sigs(&blocked, &oldset);
+ if (error < 0) {
+ mlog_errno(error);
+ ret = VM_FAULT_SIGBUS;
+ goto out;
+ }
+
+ ret = filemap_fault(area, vmf);
+
+ error = ocfs2_vm_op_unblock_sigs(&oldset);
+ if (error < 0)
+ mlog_errno(error);
+out:
+ mlog_exit_ptr(vmf->page);
+ return ret;
+}
+
+static int __ocfs2_page_mkwrite(struct inode *inode, struct buffer_head *di_bh,
+ struct page *page)
+{
int ret;
+ struct address_space *mapping = inode->i_mapping;
+ loff_t pos = page_offset(page);
+ unsigned int len = PAGE_CACHE_SIZE;
+ pgoff_t last_index;
+ struct page *locked_page = NULL;
+ void *fsdata;
+ loff_t size = i_size_read(inode);
- mlog_entry("(inode %lu, address %lu)\n", inode->i_ino, address);
+ /*
+ * Another node might have truncated while we were waiting on
+ * cluster locks.
+ */
+ last_index = size >> PAGE_CACHE_SHIFT;
+ if (page->index > last_index) {
+ ret = -EINVAL;
+ goto out;
+ }
- /* The best way to deal with signals in this path is
- * to block them upfront, rather than allowing the
- * locking paths to return -ERESTARTSYS. */
- sigfillset(&blocked);
+ /*
+ * The i_size check above doesn't catch the case where nodes
+ * truncated and then re-extended the file. We'll re-check the
+ * page mapping after taking the page lock inside of
+ * ocfs2_write_begin_nolock().
+ */
+ if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
+ /*
+ * the page has been umapped in ocfs2_data_downconvert_worker.
+ * So return 0 here and let VFS retry.
+ */
+ ret = 0;
+ goto out;
+ }
- /* We should technically never get a bad ret return
- * from sigprocmask */
- ret = sigprocmask(SIG_BLOCK, &blocked, &oldset);
+ /*
+ * Call ocfs2_write_begin() and ocfs2_write_end() to take
+ * advantage of the allocation code there. We pass a write
+ * length of the whole page (chopped to i_size) to make sure
+ * the whole thing is allocated.
+ *
+ * Since we know the page is up to date, we don't have to
+ * worry about ocfs2_write_begin() skipping some buffer reads
+ * because the "write" would invalidate their data.
+ */
+ if (page->index == last_index)
+ len = size & ~PAGE_CACHE_MASK;
+
+ ret = ocfs2_write_begin_nolock(mapping, pos, len, 0, &locked_page,
+ &fsdata, di_bh, page);
+ if (ret) {
+ if (ret != -ENOSPC)
+ mlog_errno(ret);
+ goto out;
+ }
+
+ ret = ocfs2_write_end_nolock(mapping, pos, len, len, locked_page,
+ fsdata);
if (ret < 0) {
mlog_errno(ret);
goto out;
}
+ BUG_ON(ret != len);
+ ret = 0;
+out:
+ return ret;
+}
- page = filemap_nopage(area, address, type);
+static int ocfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
+{
+ struct page *page = vmf->page;
+ struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
+ struct buffer_head *di_bh = NULL;
+ sigset_t blocked, oldset;
+ int ret, ret2;
+
+ ret = ocfs2_vm_op_block_sigs(&blocked, &oldset);
+ if (ret < 0) {
+ mlog_errno(ret);
+ return ret;
+ }
- ret = sigprocmask(SIG_SETMASK, &oldset, NULL);
- if (ret < 0)
+ /*
+ * The cluster locks taken will block a truncate from another
+ * node. Taking the data lock will also ensure that we don't
+ * attempt page truncation as part of a downconvert.
+ */
+ ret = ocfs2_inode_lock(inode, &di_bh, 1);
+ if (ret < 0) {
mlog_errno(ret);
+ goto out;
+ }
+
+ /*
+ * The alloc sem should be enough to serialize with
+ * ocfs2_truncate_file() changing i_size as well as any thread
+ * modifying the inode btree.
+ */
+ down_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+ ret = __ocfs2_page_mkwrite(inode, di_bh, page);
+
+ up_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+ brelse(di_bh);
+ ocfs2_inode_unlock(inode, 1);
+
out:
- mlog_exit_ptr(page);
- return page;
+ ret2 = ocfs2_vm_op_unblock_sigs(&oldset);
+ if (ret2 < 0)
+ mlog_errno(ret2);
+ if (ret)
+ ret = VM_FAULT_SIGBUS;
+ return ret;
}
-static struct vm_operations_struct ocfs2_file_vm_ops = {
- .nopage = ocfs2_nopage,
+static const struct vm_operations_struct ocfs2_file_vm_ops = {
+ .fault = ocfs2_fault,
+ .page_mkwrite = ocfs2_page_mkwrite,
};
int ocfs2_mmap(struct file *file, struct vm_area_struct *vma)
{
- /* We don't want to support shared writable mappings yet. */
- if (((vma->vm_flags & VM_SHARED) || (vma->vm_flags & VM_MAYSHARE))
- && ((vma->vm_flags & VM_WRITE) || (vma->vm_flags & VM_MAYWRITE))) {
- mlog(0, "disallow shared writable mmaps %lx\n", vma->vm_flags);
- /* This is -EINVAL because generic_file_readonly_mmap
- * returns it in a similar situation. */
- return -EINVAL;
- }
+ int ret = 0, lock_level = 0;
- file_accessed(file);
+ ret = ocfs2_inode_lock_atime(file->f_dentry->d_inode,
+ file->f_vfsmnt, &lock_level);
+ if (ret < 0) {
+ mlog_errno(ret);
+ goto out;
+ }
+ ocfs2_inode_unlock(file->f_dentry->d_inode, lock_level);
+out:
vma->vm_ops = &ocfs2_file_vm_ops;
+ vma->vm_flags |= VM_CAN_NONLINEAR;
return 0;
}
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff