afs: convert to new aops
authorNick Piggin <npiggin@suse.de>
Thu, 16 Oct 2008 05:04:32 +0000 (22:04 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 16 Oct 2008 18:21:48 +0000 (11:21 -0700)
Cannot assume writes will fully complete, so this conversion goes the easy
way and always brings the page uptodate before the write.

[dhowells@redhat.com: style tweaks]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/afs/file.c
fs/afs/internal.h
fs/afs/write.c

index 525f7c5..a390176 100644 (file)
@@ -50,8 +50,8 @@ const struct address_space_operations afs_fs_aops = {
        .launder_page   = afs_launder_page,
        .releasepage    = afs_releasepage,
        .invalidatepage = afs_invalidatepage,
-       .prepare_write  = afs_prepare_write,
-       .commit_write   = afs_commit_write,
+       .write_begin    = afs_write_begin,
+       .write_end      = afs_write_end,
        .writepage      = afs_writepage,
        .writepages     = afs_writepages,
 };
index 3cb6920..67f259d 100644 (file)
@@ -728,8 +728,12 @@ extern int afs_volume_release_fileserver(struct afs_vnode *,
  */
 extern int afs_set_page_dirty(struct page *);
 extern void afs_put_writeback(struct afs_writeback *);
-extern int afs_prepare_write(struct file *, struct page *, unsigned, unsigned);
-extern int afs_commit_write(struct file *, struct page *, unsigned, unsigned);
+extern int afs_write_begin(struct file *file, struct address_space *mapping,
+                       loff_t pos, unsigned len, unsigned flags,
+                       struct page **pagep, void **fsdata);
+extern int afs_write_end(struct file *file, struct address_space *mapping,
+                       loff_t pos, unsigned len, unsigned copied,
+                       struct page *page, void *fsdata);
 extern int afs_writepage(struct page *, struct writeback_control *);
 extern int afs_writepages(struct address_space *, struct writeback_control *);
 extern int afs_write_inode(struct inode *, int);
index 065b4e1..d6b85da 100644 (file)
@@ -84,15 +84,23 @@ void afs_put_writeback(struct afs_writeback *wb)
  * partly or wholly fill a page that's under preparation for writing
  */
 static int afs_fill_page(struct afs_vnode *vnode, struct key *key,
-                        unsigned start, unsigned len, struct page *page)
+                        loff_t pos, unsigned len, struct page *page)
 {
+       loff_t i_size;
+       unsigned eof;
        int ret;
 
-       _enter(",,%u,%u", start, len);
+       _enter(",,%llu,%u", (unsigned long long)pos, len);
 
-       ASSERTCMP(start + len, <=, PAGE_SIZE);
+       ASSERTCMP(len, <=, PAGE_CACHE_SIZE);
 
-       ret = afs_vnode_fetch_data(vnode, key, start, len, page);
+       i_size = i_size_read(&vnode->vfs_inode);
+       if (pos + len > i_size)
+               eof = i_size;
+       else
+               eof = PAGE_CACHE_SIZE;
+
+       ret = afs_vnode_fetch_data(vnode, key, 0, eof, page);
        if (ret < 0) {
                if (ret == -ENOENT) {
                        _debug("got NOENT from server"
@@ -107,109 +115,55 @@ static int afs_fill_page(struct afs_vnode *vnode, struct key *key,
 }
 
 /*
- * prepare a page for being written to
- */
-static int afs_prepare_page(struct afs_vnode *vnode, struct page *page,
-                           struct key *key, unsigned offset, unsigned to)
-{
-       unsigned eof, tail, start, stop, len;
-       loff_t i_size, pos;
-       void *p;
-       int ret;
-
-       _enter("");
-
-       if (offset == 0 && to == PAGE_SIZE)
-               return 0;
-
-       p = kmap_atomic(page, KM_USER0);
-
-       i_size = i_size_read(&vnode->vfs_inode);
-       pos = (loff_t) page->index << PAGE_SHIFT;
-       if (pos >= i_size) {
-               /* partial write, page beyond EOF */
-               _debug("beyond");
-               if (offset > 0)
-                       memset(p, 0, offset);
-               if (to < PAGE_SIZE)
-                       memset(p + to, 0, PAGE_SIZE - to);
-               kunmap_atomic(p, KM_USER0);
-               return 0;
-       }
-
-       if (i_size - pos >= PAGE_SIZE) {
-               /* partial write, page entirely before EOF */
-               _debug("before");
-               tail = eof = PAGE_SIZE;
-       } else {
-               /* partial write, page overlaps EOF */
-               eof = i_size - pos;
-               _debug("overlap %u", eof);
-               tail = max(eof, to);
-               if (tail < PAGE_SIZE)
-                       memset(p + tail, 0, PAGE_SIZE - tail);
-               if (offset > eof)
-                       memset(p + eof, 0, PAGE_SIZE - eof);
-       }
-
-       kunmap_atomic(p, KM_USER0);
-
-       ret = 0;
-       if (offset > 0 || eof > to) {
-               /* need to fill one or two bits that aren't going to be written
-                * (cover both fillers in one read if there are two) */
-               start = (offset > 0) ? 0 : to;
-               stop = (eof > to) ? eof : offset;
-               len = stop - start;
-               _debug("wr=%u-%u av=0-%u rd=%u@%u",
-                      offset, to, eof, start, len);
-               ret = afs_fill_page(vnode, key, start, len, page);
-       }
-
-       _leave(" = %d", ret);
-       return ret;
-}
-
-/*
  * prepare to perform part of a write to a page
- * - the caller holds the page locked, preventing it from being written out or
- *   modified by anyone else
  */
-int afs_prepare_write(struct file *file, struct page *page,
-                     unsigned offset, unsigned to)
+int afs_write_begin(struct file *file, struct address_space *mapping,
+                   loff_t pos, unsigned len, unsigned flags,
+                   struct page **pagep, void **fsdata)
 {
        struct afs_writeback *candidate, *wb;
        struct afs_vnode *vnode = AFS_FS_I(file->f_dentry->d_inode);
+       struct page *page;
        struct key *key = file->private_data;
-       pgoff_t index;
+       unsigned from = pos & (PAGE_CACHE_SIZE - 1);
+       unsigned to = from + len;
+       pgoff_t index = pos >> PAGE_CACHE_SHIFT;
        int ret;
 
        _enter("{%x:%u},{%lx},%u,%u",
-              vnode->fid.vid, vnode->fid.vnode, page->index, offset, to);
+              vnode->fid.vid, vnode->fid.vnode, index, from, to);
 
        candidate = kzalloc(sizeof(*candidate), GFP_KERNEL);
        if (!candidate)
                return -ENOMEM;
        candidate->vnode = vnode;
-       candidate->first = candidate->last = page->index;
-       candidate->offset_first = offset;
+       candidate->first = candidate->last = index;
+       candidate->offset_first = from;
        candidate->to_last = to;
        candidate->usage = 1;
        candidate->state = AFS_WBACK_PENDING;
        init_waitqueue_head(&candidate->waitq);
 
+       page = __grab_cache_page(mapping, index);
+       if (!page) {
+               kfree(candidate);
+               return -ENOMEM;
+       }
+       *pagep = page;
+       /* page won't leak in error case: it eventually gets cleaned off LRU */
+
        if (!PageUptodate(page)) {
                _debug("not up to date");
-               ret = afs_prepare_page(vnode, page, key, offset, to);
+               ret = afs_fill_page(vnode, key, pos, len, page);
                if (ret < 0) {
                        kfree(candidate);
                        _leave(" = %d [prep]", ret);
                        return ret;
                }
+               SetPageUptodate(page);
        }
 
 try_again:
-       index = page->index;
        spin_lock(&vnode->writeback_lock);
 
        /* see if this page is already pending a writeback under a suitable key
@@ -242,8 +196,8 @@ try_again:
 subsume_in_current_wb:
        _debug("subsume");
        ASSERTRANGE(wb->first, <=, index, <=, wb->last);
-       if (index == wb->first && offset < wb->offset_first)
-               wb->offset_first = offset;
+       if (index == wb->first && from < wb->offset_first)
+               wb->offset_first = from;
        if (index == wb->last && to > wb->to_last)
                wb->to_last = to;
        spin_unlock(&vnode->writeback_lock);
@@ -289,17 +243,17 @@ flush_conflicting_wb:
 /*
  * finalise part of a write to a page
  */
-int afs_commit_write(struct file *file, struct page *page,
-                    unsigned offset, unsigned to)
+int afs_write_end(struct file *file, struct address_space *mapping,
+                 loff_t pos, unsigned len, unsigned copied,
+                 struct page *page, void *fsdata)
 {
        struct afs_vnode *vnode = AFS_FS_I(file->f_dentry->d_inode);
        loff_t i_size, maybe_i_size;
 
-       _enter("{%x:%u},{%lx},%u,%u",
-              vnode->fid.vid, vnode->fid.vnode, page->index, offset, to);
+       _enter("{%x:%u},{%lx}",
+              vnode->fid.vid, vnode->fid.vnode, page->index);
 
-       maybe_i_size = (loff_t) page->index << PAGE_SHIFT;
-       maybe_i_size += to;
+       maybe_i_size = pos + copied;
 
        i_size = i_size_read(&vnode->vfs_inode);
        if (maybe_i_size > i_size) {
@@ -310,12 +264,13 @@ int afs_commit_write(struct file *file, struct page *page,
                spin_unlock(&vnode->writeback_lock);
        }
 
-       SetPageUptodate(page);
        set_page_dirty(page);
        if (PageDirty(page))
                _debug("dirtied");
+       unlock_page(page);
+       page_cache_release(page);
 
-       return 0;
+       return copied;
 }
 
 /*