afs: Get rid of the afs_writeback record

Get rid of the afs_writeback record that kAFS is using to match keys with
writes made by that key.

Instead, keep a list of keys that have a file open for writing and/or
sync'ing and iterate through those.
Signed-off-by: David Howells <dhowells@redhat.com>

fs/afs/file.c

@@ -23,7 +23,6 @@ static int afs_readpage(struct file *file, struct page *page);
 static void afs_invalidatepage(struct page *page, unsigned int offset,
 			       unsigned int length);
 static int afs_releasepage(struct page *page, gfp_t gfp_flags);
-static int afs_launder_page(struct page *page);
 
 static int afs_readpages(struct file *filp, struct address_space *mapping,
 			 struct list_head *pages, unsigned nr_pages);
@@ -62,6 +61,50 @@ const struct address_space_operations afs_fs_aops = {
 	.writepages	= afs_writepages,
 };
 
+/*
+ * Discard a pin on a writeback key.
+ */
+void afs_put_wb_key(struct afs_wb_key *wbk)
+{
+	if (refcount_dec_and_test(&wbk->usage)) {
+		key_put(wbk->key);
+		kfree(wbk);
+	}
+}
+
+/*
+ * Cache key for writeback.
+ */
+int afs_cache_wb_key(struct afs_vnode *vnode, struct afs_file *af)
+{
+	struct afs_wb_key *wbk, *p;
+
+	wbk = kzalloc(sizeof(struct afs_wb_key), GFP_KERNEL);
+	if (!wbk)
+		return -ENOMEM;
+	refcount_set(&wbk->usage, 2);
+	wbk->key = af->key;
+
+	spin_lock(&vnode->wb_lock);
+	list_for_each_entry(p, &vnode->wb_keys, vnode_link) {
+		if (p->key == wbk->key)
+			goto found;
+	}
+
+	key_get(wbk->key);
+	list_add_tail(&wbk->vnode_link, &vnode->wb_keys);
+	spin_unlock(&vnode->wb_lock);
+	af->wb = wbk;
+	return 0;
+
+found:
+	refcount_inc(&p->usage);
+	spin_unlock(&vnode->wb_lock);
+	af->wb = p;
+	kfree(wbk);
+	return 0;
+}
+
 /*
  * open an AFS file or directory and attach a key to it
  */
@@ -85,12 +128,18 @@ int afs_open(struct inode *inode, struct file *file)
 		ret = -ENOMEM;
 		goto error_key;
 	}
+	af->key = key;
 
 	ret = afs_validate(vnode, key);
 	if (ret < 0)
 		goto error_af;
 
-	af->key = key;
+	if (file->f_mode & FMODE_WRITE) {
+		ret = afs_cache_wb_key(vnode, af);
+		if (ret < 0)
+			goto error_af;
+	}
+	
 	file->private_data = af;
 	_leave(" = 0");
 	return 0;
@@ -115,8 +164,11 @@ int afs_release(struct inode *inode, struct file *file)
 	_enter("{%x:%u},", vnode->fid.vid, vnode->fid.vnode);
 
 	file->private_data = NULL;
+	if (af->wb)
+		afs_put_wb_key(af->wb);
 	key_put(af->key);
 	kfree(af);
+	afs_prune_wb_keys(vnode);
 	_leave(" = 0");
 	return 0;
 }
@@ -516,16 +568,6 @@ static int afs_readpages(struct file *file, struct address_space *mapping,
 	return ret;
 }
 
-/*
- * write back a dirty page
- */
-static int afs_launder_page(struct page *page)
-{
-	_enter("{%lu}", page->index);
-
-	return 0;
-}
-
 /*
  * invalidate part or all of a page
  * - release a page and clean up its private data if offset is 0 (indicating
@@ -534,8 +576,6 @@ static int afs_launder_page(struct page *page)
 static void afs_invalidatepage(struct page *page, unsigned int offset,
 			       unsigned int length)
 {
-	struct afs_writeback *wb = (struct afs_writeback *) page_private(page);
-
 	_enter("{%lu},%u,%u", page->index, offset, length);
 
 	BUG_ON(!PageLocked(page));
@@ -551,13 +591,8 @@ static void afs_invalidatepage(struct page *page, unsigned int offset,
 #endif
 
 		if (PagePrivate(page)) {
-			if (wb && !PageWriteback(page)) {
-				set_page_private(page, 0);
-				afs_put_writeback(wb);
-			}
-
-			if (!page_private(page))
-				ClearPagePrivate(page);
+			set_page_private(page, 0);
+			ClearPagePrivate(page);
 		}
 	}
 
@@ -570,7 +605,6 @@ static void afs_invalidatepage(struct page *page, unsigned int offset,
  */
 static int afs_releasepage(struct page *page, gfp_t gfp_flags)
 {
-	struct afs_writeback *wb = (struct afs_writeback *) page_private(page);
 	struct afs_vnode *vnode = AFS_FS_I(page->mapping->host);
 
 	_enter("{{%x:%u}[%lu],%lx},%x",
@@ -587,10 +621,7 @@ static int afs_releasepage(struct page *page, gfp_t gfp_flags)
 #endif
 
 	if (PagePrivate(page)) {
-		if (wb) {
-			set_page_private(page, 0);
-			afs_put_writeback(wb);
-		}
+		set_page_private(page, 0);
 		ClearPagePrivate(page);
 	}
 

fs/afs/fsclient.c

@@ -1119,18 +1119,18 @@ static const struct afs_call_type afs_RXFSStoreData64 = {
  * store a set of pages to a very large file
  */
 static int afs_fs_store_data64(struct afs_fs_cursor *fc,
-			       struct afs_writeback *wb,
+			       struct address_space *mapping,
 			       pgoff_t first, pgoff_t last,
 			       unsigned offset, unsigned to,
 			       loff_t size, loff_t pos, loff_t i_size)
 {
-	struct afs_vnode *vnode = wb->vnode;
+	struct afs_vnode *vnode = fc->vnode;
 	struct afs_call *call;
 	struct afs_net *net = afs_v2net(vnode);
 	__be32 *bp;
 
 	_enter(",%x,{%x:%u},,",
-	       key_serial(wb->key), vnode->fid.vid, vnode->fid.vnode);
+	       key_serial(fc->key), vnode->fid.vid, vnode->fid.vnode);
 
 	call = afs_alloc_flat_call(net, &afs_RXFSStoreData64,
 				   (4 + 6 + 3 * 2) * 4,
@@ -1138,10 +1138,9 @@ static int afs_fs_store_data64(struct afs_fs_cursor *fc,
 	if (!call)
 		return -ENOMEM;
 
-	call->wb = wb;
-	call->key = wb->key;
+	call->key = fc->key;
+	call->mapping = mapping;
 	call->reply[0] = vnode;
-	call->mapping = vnode->vfs_inode.i_mapping;
 	call->first = first;
 	call->last = last;
 	call->first_offset = offset;
@@ -1177,18 +1176,18 @@ static int afs_fs_store_data64(struct afs_fs_cursor *fc,
 /*
  * store a set of pages
  */
-int afs_fs_store_data(struct afs_fs_cursor *fc, struct afs_writeback *wb,
+int afs_fs_store_data(struct afs_fs_cursor *fc, struct address_space *mapping,
 		      pgoff_t first, pgoff_t last,
 		      unsigned offset, unsigned to)
 {
-	struct afs_vnode *vnode = wb->vnode;
+	struct afs_vnode *vnode = fc->vnode;
 	struct afs_call *call;
 	struct afs_net *net = afs_v2net(vnode);
 	loff_t size, pos, i_size;
 	__be32 *bp;
 
 	_enter(",%x,{%x:%u},,",
-	       key_serial(wb->key), vnode->fid.vid, vnode->fid.vnode);
+	       key_serial(fc->key), vnode->fid.vid, vnode->fid.vnode);
 
 	size = (loff_t)to - (loff_t)offset;
 	if (first != last)
@@ -1205,7 +1204,7 @@ int afs_fs_store_data(struct afs_fs_cursor *fc, struct afs_writeback *wb,
 	       (unsigned long long) i_size);
 
 	if (pos >> 32 || i_size >> 32 || size >> 32 || (pos + size) >> 32)
-		return afs_fs_store_data64(fc, wb, first, last, offset, to,
+		return afs_fs_store_data64(fc, mapping, first, last, offset, to,
 					   size, pos, i_size);
 
 	call = afs_alloc_flat_call(net, &afs_RXFSStoreData,
@@ -1214,10 +1213,9 @@ int afs_fs_store_data(struct afs_fs_cursor *fc, struct afs_writeback *wb,
 	if (!call)
 		return -ENOMEM;
 
-	call->wb = wb;
-	call->key = wb->key;
+	call->key = fc->key;
+	call->mapping = mapping;
 	call->reply[0] = vnode;
-	call->mapping = vnode->vfs_inode.i_mapping;
 	call->first = first;
 	call->last = last;
 	call->first_offset = offset;

fs/afs/inode.c

@@ -482,7 +482,12 @@ void afs_evict_inode(struct inode *inode)
 		vnode->cb_interest = NULL;
 	}
 
-	ASSERT(list_empty(&vnode->writebacks));
+	while (!list_empty(&vnode->wb_keys)) {
+		struct afs_wb_key *wbk = list_entry(vnode->wb_keys.next,
+						    struct afs_wb_key, vnode_link);
+		list_del(&wbk->vnode_link);
+		afs_put_wb_key(wbk);
+	}
 
 #ifdef CONFIG_AFS_FSCACHE
 	fscache_relinquish_cookie(vnode->cache, 0);
@@ -514,10 +519,8 @@ int afs_setattr(struct dentry *dentry, struct iattr *attr)
 	}
 
 	/* flush any dirty data outstanding on a regular file */
-	if (S_ISREG(vnode->vfs_inode.i_mode)) {
+	if (S_ISREG(vnode->vfs_inode.i_mode))
 		filemap_write_and_wait(vnode->vfs_inode.i_mapping);
-		afs_writeback_all(vnode);
-	}
 
 	if (attr->ia_valid & ATTR_FILE) {
 		key = afs_file_key(attr->ia_file);

fs/afs/internal.h

@@ -89,8 +89,7 @@ struct afs_call {
 	struct afs_server	*cm_server;	/* Server affected by incoming CM call */
 	struct afs_cb_interest	*cbi;		/* Callback interest for server used */
 	void			*request;	/* request data (first part) */
-	struct address_space	*mapping;	/* page set */
-	struct afs_writeback	*wb;		/* writeback being performed */
+	struct address_space	*mapping;	/* Pages being written from */
 	void			*buffer;	/* reply receive buffer */
 	void			*reply[4];	/* Where to put the reply */
 	pgoff_t			first;		/* first page in mapping to deal with */
@@ -138,11 +137,21 @@ struct afs_call_type {
 	void (*work)(struct work_struct *work);
 };
 
+/*
+ * Key available for writeback on a file.
+ */
+struct afs_wb_key {
+	refcount_t		usage;
+	struct key		*key;
+	struct list_head	vnode_link;	/* Link in vnode->wb_keys */
+};
+
 /*
  * AFS open file information record.  Pointed to by file->private_data.
  */
 struct afs_file {
 	struct key		*key;		/* The key this file was opened with */
+	struct afs_wb_key	*wb;		/* Writeback key record for this file */
 };
 
 static inline struct key *afs_file_key(struct file *file)
@@ -167,32 +176,6 @@ struct afs_read {
 	struct page		*pages[];
 };
 
-/*
- * record of an outstanding writeback on a vnode
- */
-struct afs_writeback {
-	struct list_head	link;		/* link in vnode->writebacks */
-	struct work_struct	writer;		/* work item to perform the writeback */
-	struct afs_vnode	*vnode;		/* vnode to which this write applies */
-	struct key		*key;		/* owner of this write */
-	wait_queue_head_t	waitq;		/* completion and ready wait queue */
-	pgoff_t			first;		/* first page in batch */
-	pgoff_t			point;		/* last page in current store op */
-	pgoff_t			last;		/* last page in batch (inclusive) */
-	unsigned		offset_first;	/* offset into first page of start of write */
-	unsigned		to_last;	/* offset into last page of end of write */
-	int			num_conflicts;	/* count of conflicting writes in list */
-	int			usage;
-	bool			conflicts;	/* T if has dependent conflicts */
-	enum {
-		AFS_WBACK_SYNCING,		/* synchronisation being performed */
-		AFS_WBACK_PENDING,		/* write pending */
-		AFS_WBACK_CONFLICTING,		/* conflicting writes posted */
-		AFS_WBACK_WRITING,		/* writing back */
-		AFS_WBACK_COMPLETE		/* the writeback record has been unlinked */
-	} state __attribute__((packed));
-};
-
 /*
  * AFS superblock private data
  * - there's one superblock per volume
@@ -460,7 +443,7 @@ struct afs_vnode {
 	struct afs_permits	*permit_cache;	/* cache of permits so far obtained */
 	struct mutex		io_lock;	/* Lock for serialising I/O on this mutex */
 	struct mutex		validate_lock;	/* lock for validating this vnode */
-	spinlock_t		writeback_lock;	/* lock for writebacks */
+	spinlock_t		wb_lock;	/* lock for wb_keys */
 	spinlock_t		lock;		/* waitqueue/flags lock */
 	unsigned long		flags;
 #define AFS_VNODE_CB_PROMISED	0		/* Set if vnode has a callback promise */
@@ -476,7 +459,7 @@ struct afs_vnode {
 #define AFS_VNODE_AUTOCELL	10		/* set if Vnode is an auto mount point */
 #define AFS_VNODE_PSEUDODIR	11		/* set if Vnode is a pseudo directory */
 
-	struct list_head	writebacks;	/* alterations in pagecache that need writing */
+	struct list_head	wb_keys;	/* List of keys available for writeback */
 	struct list_head	pending_locks;	/* locks waiting to be granted */
 	struct list_head	granted_locks;	/* locks granted on this file */
 	struct delayed_work	lock_work;	/* work to be done in locking */
@@ -648,6 +631,8 @@ extern const struct address_space_operations afs_fs_aops;
 extern const struct inode_operations afs_file_inode_operations;
 extern const struct file_operations afs_file_operations;
 
+extern int afs_cache_wb_key(struct afs_vnode *, struct afs_file *);
+extern void afs_put_wb_key(struct afs_wb_key *);
 extern int afs_open(struct inode *, struct file *);
 extern int afs_release(struct inode *, struct file *);
 extern int afs_fetch_data(struct afs_vnode *, struct key *, struct afs_read *);
@@ -678,7 +663,7 @@ extern int afs_fs_symlink(struct afs_fs_cursor *, const char *, const char *,
 			  struct afs_fid *, struct afs_file_status *);
 extern int afs_fs_rename(struct afs_fs_cursor *, const char *,
 			 struct afs_vnode *, const char *);
-extern int afs_fs_store_data(struct afs_fs_cursor *, struct afs_writeback *,
+extern int afs_fs_store_data(struct afs_fs_cursor *, struct address_space *,
 			     pgoff_t, pgoff_t, unsigned, unsigned);
 extern int afs_fs_setattr(struct afs_fs_cursor *, struct iattr *);
 extern int afs_fs_get_volume_status(struct afs_fs_cursor *, struct afs_volume_status *);
@@ -889,7 +874,6 @@ extern int afs_check_volume_status(struct afs_volume *, struct key *);
  * write.c
  */
 extern int afs_set_page_dirty(struct page *);
-extern void afs_put_writeback(struct afs_writeback *);
 extern int afs_write_begin(struct file *file, struct address_space *mapping,
 			loff_t pos, unsigned len, unsigned flags,
 			struct page **pagep, void **fsdata);
@@ -900,9 +884,10 @@ extern int afs_writepage(struct page *, struct writeback_control *);
 extern int afs_writepages(struct address_space *, struct writeback_control *);
 extern void afs_pages_written_back(struct afs_vnode *, struct afs_call *);
 extern ssize_t afs_file_write(struct kiocb *, struct iov_iter *);
-extern int afs_writeback_all(struct afs_vnode *);
 extern int afs_flush(struct file *, fl_owner_t);
 extern int afs_fsync(struct file *, loff_t, loff_t, int);
+extern void afs_prune_wb_keys(struct afs_vnode *);
+extern int afs_launder_page(struct page *);
 
 /*
  * xattr.c

fs/afs/super.c

@@ -546,9 +546,9 @@ static void afs_i_init_once(void *_vnode)
 	inode_init_once(&vnode->vfs_inode);
 	mutex_init(&vnode->io_lock);
 	mutex_init(&vnode->validate_lock);
-	spin_lock_init(&vnode->writeback_lock);
+	spin_lock_init(&vnode->wb_lock);
 	spin_lock_init(&vnode->lock);
-	INIT_LIST_HEAD(&vnode->writebacks);
+	INIT_LIST_HEAD(&vnode->wb_keys);
 	INIT_LIST_HEAD(&vnode->pending_locks);
 	INIT_LIST_HEAD(&vnode->granted_locks);
 	INIT_DELAYED_WORK(&vnode->lock_work, afs_lock_work);

fs/afs/write.c

@@ -8,6 +8,7 @@
  * as published by the Free Software Foundation; either version
  * 2 of the License, or (at your option) any later version.
  */
+
 #include <linux/backing-dev.h>
 #include <linux/slab.h>
 #include <linux/fs.h>
@@ -16,8 +17,18 @@
 #include <linux/pagevec.h>
 #include "internal.h"
 
-static int afs_write_back_from_locked_page(struct afs_writeback *wb,
-					   struct page *page);
+/*
+ * We use page->private to hold the amount of the page that we've written to,
+ * splitting the field into two parts.  However, we need to represent a range
+ * 0...PAGE_SIZE inclusive, so we can't support 64K pages on a 32-bit system.
+ */
+#if PAGE_SIZE > 32768
+#define AFS_PRIV_MAX	0xffffffff
+#define AFS_PRIV_SHIFT	32
+#else
+#define AFS_PRIV_MAX	0xffff
+#define AFS_PRIV_SHIFT	16
+#endif
 
 /*
  * mark a page as having been made dirty and thus needing writeback
@@ -28,58 +39,6 @@ int afs_set_page_dirty(struct page *page)
 	return __set_page_dirty_nobuffers(page);
 }
 
-/*
- * unlink a writeback record because its usage has reached zero
- * - must be called with the wb->vnode->writeback_lock held
- */
-static void afs_unlink_writeback(struct afs_writeback *wb)
-{
-	struct afs_writeback *front;
-	struct afs_vnode *vnode = wb->vnode;
-
-	list_del_init(&wb->link);
-	if (!list_empty(&vnode->writebacks)) {
-		/* if an fsync rises to the front of the queue then wake it
-		 * up */
-		front = list_entry(vnode->writebacks.next,
-				   struct afs_writeback, link);
-		if (front->state == AFS_WBACK_SYNCING) {
-			_debug("wake up sync");
-			front->state = AFS_WBACK_COMPLETE;
-			wake_up(&front->waitq);
-		}
-	}
-}
-
-/*
- * free a writeback record
- */
-static void afs_free_writeback(struct afs_writeback *wb)
-{
-	_enter("");
-	key_put(wb->key);
-	kfree(wb);
-}
-
-/*
- * dispose of a reference to a writeback record
- */
-void afs_put_writeback(struct afs_writeback *wb)
-{
-	struct afs_vnode *vnode = wb->vnode;
-
-	_enter("{%d}", wb->usage);
-
-	spin_lock(&vnode->writeback_lock);
-	if (--wb->usage == 0)
-		afs_unlink_writeback(wb);
-	else
-		wb = NULL;
-	spin_unlock(&vnode->writeback_lock);
-	if (wb)
-		afs_free_writeback(wb);
-}
-
 /*
  * partly or wholly fill a page that's under preparation for writing
  */
@@ -125,42 +84,32 @@ 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_inode(file));
 	struct page *page;
 	struct key *key = afs_file_key(file);
-	unsigned from = pos & (PAGE_SIZE - 1);
-	unsigned to = from + len;
+	unsigned long priv;
+	unsigned f, from = pos & (PAGE_SIZE - 1);
+	unsigned t, to = from + len;
 	pgoff_t index = pos >> PAGE_SHIFT;
 	int ret;
 
 	_enter("{%x:%u},{%lx},%u,%u",
 	       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 = index;
-	candidate->offset_first = from;
-	candidate->to_last = to;
-	INIT_LIST_HEAD(&candidate->link);
-	candidate->usage = 1;
-	candidate->state = AFS_WBACK_PENDING;
-	init_waitqueue_head(&candidate->waitq);
+	/* We want to store information about how much of a page is altered in
+	 * page->private.
+	 */
+	BUILD_BUG_ON(PAGE_SIZE > 32768 && sizeof(page->private) < 8);
 
 	page = grab_cache_page_write_begin(mapping, index, flags);
-	if (!page) {
-		kfree(candidate);
+	if (!page)
 		return -ENOMEM;
-	}
 
 	if (!PageUptodate(page) && len != PAGE_SIZE) {
 		ret = afs_fill_page(vnode, key, pos & PAGE_MASK, PAGE_SIZE, page);
 		if (ret < 0) {
 			unlock_page(page);
 			put_page(page);
-			kfree(candidate);
 			_leave(" = %d [prep]", ret);
 			return ret;
 		}
@@ -171,79 +120,52 @@ int afs_write_begin(struct file *file, struct address_space *mapping,
 	*pagep = page;
 
 try_again:
-	spin_lock(&vnode->writeback_lock);
-
-	/* see if this page is already pending a writeback under a suitable key
-	 * - if so we can just join onto that one */
-	wb = (struct afs_writeback *) page_private(page);
-	if (wb) {
-		if (wb->key == key && wb->state == AFS_WBACK_PENDING)
-			goto subsume_in_current_wb;
-		goto flush_conflicting_wb;
+	/* See if this page is already partially written in a way that we can
+	 * merge the new write with.
+	 */
+	t = f = 0;
+	if (PagePrivate(page)) {
+		priv = page_private(page);
+		f = priv & AFS_PRIV_MAX;
+		t = priv >> AFS_PRIV_SHIFT;
+		ASSERTCMP(f, <=, t);
 	}
 
-	if (index > 0) {
-		/* see if we can find an already pending writeback that we can
-		 * append this page to */
-		list_for_each_entry(wb, &vnode->writebacks, link) {
-			if (wb->last == index - 1 && wb->key == key &&
-			    wb->state == AFS_WBACK_PENDING)
-				goto append_to_previous_wb;
-		}
+	if (f != t) {
+		if (to < f || from > t)
+			goto flush_conflicting_write;
+		if (from < f)
+			f = from;
+		if (to > t)
+			t = to;
+	} else {
+		f = from;
+		t = to;
 	}
 
-	list_add_tail(&candidate->link, &vnode->writebacks);
-	candidate->key = key_get(key);
-	spin_unlock(&vnode->writeback_lock);
+	priv = (unsigned long)t << AFS_PRIV_SHIFT;
+	priv |= f;
 	SetPagePrivate(page);
-	set_page_private(page, (unsigned long) candidate);
-	_leave(" = 0 [new]");
-	return 0;
-
-subsume_in_current_wb:
-	_debug("subsume");
-	ASSERTRANGE(wb->first, <=, index, <=, wb->last);
-	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);
-	kfree(candidate);
-	_leave(" = 0 [sub]");
-	return 0;
-
-append_to_previous_wb:
-	_debug("append into %lx-%lx", wb->first, wb->last);
-	wb->usage++;
-	wb->last++;
-	wb->to_last = to;
-	spin_unlock(&vnode->writeback_lock);
-	SetPagePrivate(page);
-	set_page_private(page, (unsigned long) wb);
-	kfree(candidate);
-	_leave(" = 0 [app]");
+	set_page_private(page, priv);
+	_leave(" = 0");
 	return 0;
 
-	/* the page is currently bound to another context, so if it's dirty we
-	 * need to flush it before we can use the new context */
-flush_conflicting_wb:
+	/* The previous write and this write aren't adjacent or overlapping, so
+	 * flush the page out.
+	 */
+flush_conflicting_write:
 	_debug("flush conflict");
-	if (wb->state == AFS_WBACK_PENDING)
-		wb->state = AFS_WBACK_CONFLICTING;
-	spin_unlock(&vnode->writeback_lock);
-	if (clear_page_dirty_for_io(page)) {
-		ret = afs_write_back_from_locked_page(wb, page);
-		if (ret < 0) {
-			afs_put_writeback(candidate);
-			_leave(" = %d", ret);
-			return ret;
-		}
+	ret = write_one_page(page);
+	if (ret < 0) {
+		_leave(" = %d", ret);
+		return ret;
 	}
 
-	/* the page holds a ref on the writeback record */
-	afs_put_writeback(wb);
-	set_page_private(page, 0);
-	ClearPagePrivate(page);
+	ret = lock_page_killable(page);
+	if (ret < 0) {
+		_leave(" = %d", ret);
+		return ret;
+	}
 	goto try_again;
 }
 
@@ -266,11 +188,11 @@ int afs_write_end(struct file *file, struct address_space *mapping,
 
 	i_size = i_size_read(&vnode->vfs_inode);
 	if (maybe_i_size > i_size) {
-		spin_lock(&vnode->writeback_lock);
+		spin_lock(&vnode->wb_lock);
 		i_size = i_size_read(&vnode->vfs_inode);
 		if (maybe_i_size > i_size)
 			i_size_write(&vnode->vfs_inode, maybe_i_size);
-		spin_unlock(&vnode->writeback_lock);
+		spin_unlock(&vnode->wb_lock);
 	}
 
 	if (!PageUptodate(page)) {
@@ -299,9 +221,10 @@ int afs_write_end(struct file *file, struct address_space *mapping,
 /*
  * kill all the pages in the given range
  */
-static void afs_kill_pages(struct afs_vnode *vnode, bool error,
+static void afs_kill_pages(struct address_space *mapping,
 			   pgoff_t first, pgoff_t last)
 {
+	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
 	struct pagevec pv;
 	unsigned count, loop;
 
@@ -316,23 +239,62 @@ static void afs_kill_pages(struct afs_vnode *vnode, bool error,
 		count = last - first + 1;
 		if (count > PAGEVEC_SIZE)
 			count = PAGEVEC_SIZE;
-		pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
-					      first, count, pv.pages);
+		pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
 		ASSERTCMP(pv.nr, ==, count);
 
 		for (loop = 0; loop < count; loop++) {
 			struct page *page = pv.pages[loop];
 			ClearPageUptodate(page);
-			if (error)
-				SetPageError(page);
-			if (PageWriteback(page))
-				end_page_writeback(page);
+			SetPageError(page);
+			end_page_writeback(page);
 			if (page->index >= first)
 				first = page->index + 1;
+			lock_page(page);
+			generic_error_remove_page(mapping, page);
 		}
 
 		__pagevec_release(&pv);
-	} while (first < last);
+	} while (first <= last);
+
+	_leave("");
+}
+
+/*
+ * Redirty all the pages in a given range.
+ */
+static void afs_redirty_pages(struct writeback_control *wbc,
+			      struct address_space *mapping,
+			      pgoff_t first, pgoff_t last)
+{
+	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
+	struct pagevec pv;
+	unsigned count, loop;
+
+	_enter("{%x:%u},%lx-%lx",
+	       vnode->fid.vid, vnode->fid.vnode, first, last);
+
+	pagevec_init(&pv, 0);
+
+	do {
+		_debug("redirty %lx-%lx", first, last);
+
+		count = last - first + 1;
+		if (count > PAGEVEC_SIZE)
+			count = PAGEVEC_SIZE;
+		pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
+		ASSERTCMP(pv.nr, ==, count);
+
+		for (loop = 0; loop < count; loop++) {
+			struct page *page = pv.pages[loop];
+
+			redirty_page_for_writepage(wbc, page);
+			end_page_writeback(page);
+			if (page->index >= first)
+				first = page->index + 1;
+		}
+
+		__pagevec_release(&pv);
+	} while (first <= last);
 
 	_leave("");
 }
@@ -340,26 +302,55 @@ static void afs_kill_pages(struct afs_vnode *vnode, bool error,
 /*
  * write to a file
  */
-static int afs_store_data(struct afs_writeback *wb, pgoff_t first, pgoff_t last,
+static int afs_store_data(struct address_space *mapping,
+			  pgoff_t first, pgoff_t last,
 			  unsigned offset, unsigned to)
 {
+	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
 	struct afs_fs_cursor fc;
-	struct afs_vnode *vnode = wb->vnode;
-	int ret;
+	struct afs_wb_key *wbk = NULL;
+	struct list_head *p;
+	int ret = -ENOKEY, ret2;
 
-	_enter("%s{%x:%u.%u},%x,%lx,%lx,%x,%x",
+	_enter("%s{%x:%u.%u},%lx,%lx,%x,%x",
 	       vnode->volume->name,
 	       vnode->fid.vid,
 	       vnode->fid.vnode,
 	       vnode->fid.unique,
-	       key_serial(wb->key),
 	       first, last, offset, to);
 
+	spin_lock(&vnode->wb_lock);
+	p = vnode->wb_keys.next;
+
+	/* Iterate through the list looking for a valid key to use. */
+try_next_key:
+	while (p != &vnode->wb_keys) {
+		wbk = list_entry(p, struct afs_wb_key, vnode_link);
+		_debug("wbk %u", key_serial(wbk->key));
+		ret2 = key_validate(wbk->key);
+		if (ret2 == 0)
+			goto found_key;
+		if (ret == -ENOKEY)
+			ret = ret2;
+		p = p->next;
+	}
+
+	spin_unlock(&vnode->wb_lock);
+	afs_put_wb_key(wbk);
+	_leave(" = %d [no keys]", ret);
+	return ret;
+
+found_key:
+	refcount_inc(&wbk->usage);
+	spin_unlock(&vnode->wb_lock);
+
+	_debug("USE WB KEY %u", key_serial(wbk->key));
+
 	ret = -ERESTARTSYS;
-	if (afs_begin_vnode_operation(&fc, vnode, wb->key)) {
+	if (afs_begin_vnode_operation(&fc, vnode, wbk->key)) {
 		while (afs_select_fileserver(&fc)) {
 			fc.cb_break = vnode->cb_break + vnode->cb_s_break;
-			afs_fs_store_data(&fc, wb, first, last, offset, to);
+			afs_fs_store_data(&fc, mapping, first, last, offset, to);
 		}
 
 		afs_check_for_remote_deletion(&fc, fc.vnode);
@@ -367,20 +358,37 @@ static int afs_store_data(struct afs_writeback *wb, pgoff_t first, pgoff_t last,
 		ret = afs_end_vnode_operation(&fc);
 	}
 
+	switch (ret) {
+	case -EACCES:
+	case -EPERM:
+	case -ENOKEY:
+	case -EKEYEXPIRED:
+	case -EKEYREJECTED:
+	case -EKEYREVOKED:
+		_debug("next");
+		spin_lock(&vnode->wb_lock);
+		p = wbk->vnode_link.next;
+		afs_put_wb_key(wbk);
+		goto try_next_key;
+	}
+
+	afs_put_wb_key(wbk);
 	_leave(" = %d", ret);
 	return ret;
 }
 
 /*
- * synchronously write back the locked page and any subsequent non-locked dirty
- * pages also covered by the same writeback record
+ * Synchronously write back the locked page and any subsequent non-locked dirty
+ * pages.
  */
-static int afs_write_back_from_locked_page(struct afs_writeback *wb,
-					   struct page *primary_page)
+static int afs_write_back_from_locked_page(struct address_space *mapping,
+					   struct writeback_control *wbc,
+					   struct page *primary_page,
+					   pgoff_t final_page)
 {
 	struct page *pages[8], *page;
-	unsigned long count;
-	unsigned n, offset, to;
+	unsigned long count, priv;
+	unsigned n, offset, to, f, t;
 	pgoff_t start, first, last;
 	int loop, ret;
 
@@ -390,20 +398,28 @@ static int afs_write_back_from_locked_page(struct afs_writeback *wb,
 	if (test_set_page_writeback(primary_page))
 		BUG();
 
-	/* find all consecutive lockable dirty pages, stopping when we find a
-	 * page that is not immediately lockable, is not dirty or is missing,
-	 * or we reach the end of the range */
+	/* Find all consecutive lockable dirty pages that have contiguous
+	 * written regions, stopping when we find a page that is not
+	 * immediately lockable, is not dirty or is missing, or we reach the
+	 * end of the range.
+	 */
 	start = primary_page->index;
-	if (start >= wb->last)
+	priv = page_private(primary_page);
+	offset = priv & AFS_PRIV_MAX;
+	to = priv >> AFS_PRIV_SHIFT;
+
+	WARN_ON(offset == to);
+
+	if (start >= final_page || to < PAGE_SIZE)
 		goto no_more;
+
 	start++;
 	do {
 		_debug("more %lx [%lx]", start, count);
-		n = wb->last - start + 1;
+		n = final_page - start + 1;
 		if (n > ARRAY_SIZE(pages))
 			n = ARRAY_SIZE(pages);
-		n = find_get_pages_contig(wb->vnode->vfs_inode.i_mapping,
-					  start, n, pages);
+		n = find_get_pages_contig(mapping, start, ARRAY_SIZE(pages), pages);
 		_debug("fgpc %u", n);
 		if (n == 0)
 			goto no_more;
@@ -415,16 +431,27 @@ static int afs_write_back_from_locked_page(struct afs_writeback *wb,
 		}
 
 		for (loop = 0; loop < n; loop++) {
+			if (to != PAGE_SIZE)
+				break;
 			page = pages[loop];
-			if (page->index > wb->last)
+			if (page->index > final_page)
 				break;
 			if (!trylock_page(page))
 				break;
-			if (!PageDirty(page) ||
-			    page_private(page) != (unsigned long) wb) {
+			if (!PageDirty(page) || PageWriteback(page)) {
+				unlock_page(page);
+				break;
+			}
+
+			priv = page_private(page);
+			f = priv & AFS_PRIV_MAX;
+			t = priv >> AFS_PRIV_SHIFT;
+			if (f != 0) {
 				unlock_page(page);
 				break;
 			}
+			to = t;
+
 			if (!clear_page_dirty_for_io(page))
 				BUG();
 			if (test_set_page_writeback(page))
@@ -440,50 +467,55 @@ static int afs_write_back_from_locked_page(struct afs_writeback *wb,
 		}
 
 		start += loop;
-	} while (start <= wb->last && count < 65536);
+	} while (start <= final_page && count < 65536);
 
 no_more:
-	/* we now have a contiguous set of dirty pages, each with writeback set
-	 * and the dirty mark cleared; the first page is locked and must remain
-	 * so, all the rest are unlocked */
+	/* We now have a contiguous set of dirty pages, each with writeback
+	 * set; the first page is still locked at this point, but all the rest
+	 * have been unlocked.
+	 */
+	unlock_page(primary_page);
+
 	first = primary_page->index;
 	last = first + count - 1;
 
-	offset = (first == wb->first) ? wb->offset_first : 0;
-	to = (last == wb->last) ? wb->to_last : PAGE_SIZE;
-
 	_debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
 
-	ret = afs_store_data(wb, first, last, offset, to);
-	if (ret < 0) {
-		switch (ret) {
-		case -EDQUOT:
-		case -ENOSPC:
-			mapping_set_error(wb->vnode->vfs_inode.i_mapping, -ENOSPC);
-			break;
-		case -EROFS:
-		case -EIO:
-		case -EREMOTEIO:
-		case -EFBIG:
-		case -ENOENT:
-		case -ENOMEDIUM:
-		case -ENXIO:
-			afs_kill_pages(wb->vnode, true, first, last);
-			mapping_set_error(wb->vnode->vfs_inode.i_mapping, -EIO);
-			break;
-		case -EACCES:
-		case -EPERM:
-		case -ENOKEY:
-		case -EKEYEXPIRED:
-		case -EKEYREJECTED:
-		case -EKEYREVOKED:
-			afs_kill_pages(wb->vnode, false, first, last);
-			break;
-		default:
-			break;
-		}
-	} else {
+	ret = afs_store_data(mapping, first, last, offset, to);
+	switch (ret) {
+	case 0:
 		ret = count;
+		break;
+
+	default:
+		pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
+		/* Fall through */
+	case -EACCES:
+	case -EPERM:
+	case -ENOKEY:
+	case -EKEYEXPIRED:
+	case -EKEYREJECTED:
+	case -EKEYREVOKED:
+		afs_redirty_pages(wbc, mapping, first, last);
+		mapping_set_error(mapping, ret);
+		break;
+
+	case -EDQUOT:
+	case -ENOSPC:
+		afs_redirty_pages(wbc, mapping, first, last);
+		mapping_set_error(mapping, -ENOSPC);
+		break;
+
+	case -EROFS:
+	case -EIO:
+	case -EREMOTEIO:
+	case -EFBIG:
+	case -ENOENT:
+	case -ENOMEDIUM:
+	case -ENXIO:
+		afs_kill_pages(mapping, first, last);
+		mapping_set_error(mapping, ret);
+		break;
 	}
 
 	_leave(" = %d", ret);
@@ -496,16 +528,12 @@ static int afs_write_back_from_locked_page(struct afs_writeback *wb,
  */
 int afs_writepage(struct page *page, struct writeback_control *wbc)
 {
-	struct afs_writeback *wb;
 	int ret;
 
 	_enter("{%lx},", page->index);
 
-	wb = (struct afs_writeback *) page_private(page);
-	ASSERT(wb != NULL);
-
-	ret = afs_write_back_from_locked_page(wb, page);
-	unlock_page(page);
+	ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
+					      wbc->range_end >> PAGE_SHIFT);
 	if (ret < 0) {
 		_leave(" = %d", ret);
 		return 0;
@@ -524,7 +552,6 @@ static int afs_writepages_region(struct address_space *mapping,
 				 struct writeback_control *wbc,
 				 pgoff_t index, pgoff_t end, pgoff_t *_next)
 {
-	struct afs_writeback *wb;
 	struct page *page;
 	int ret, n;
 
@@ -550,7 +577,12 @@ static int afs_writepages_region(struct address_space *mapping,
 		 * (changing page->mapping to NULL), or even swizzled back from
 		 * swapper_space to tmpfs file mapping
 		 */
-		lock_page(page);
+		ret = lock_page_killable(page);
+		if (ret < 0) {
+			put_page(page);
+			_leave(" = %d", ret);
+			return ret;
+		}
 
 		if (page->mapping != mapping || !PageDirty(page)) {
 			unlock_page(page);
@@ -566,17 +598,9 @@ static int afs_writepages_region(struct address_space *mapping,
 			continue;
 		}
 
-		wb = (struct afs_writeback *) page_private(page);
-		ASSERT(wb != NULL);
-
-		spin_lock(&wb->vnode->writeback_lock);
-		wb->state = AFS_WBACK_WRITING;
-		spin_unlock(&wb->vnode->writeback_lock);
-
 		if (!clear_page_dirty_for_io(page))
 			BUG();
-		ret = afs_write_back_from_locked_page(wb, page);
-		unlock_page(page);
+		ret = afs_write_back_from_locked_page(mapping, wbc, page, end);
 		put_page(page);
 		if (ret < 0) {
 			_leave(" = %d", ret);
@@ -632,17 +656,13 @@ int afs_writepages(struct address_space *mapping,
  */
 void afs_pages_written_back(struct afs_vnode *vnode, struct afs_call *call)
 {
-	struct afs_writeback *wb = call->wb;
 	struct pagevec pv;
 	unsigned count, loop;
 	pgoff_t first = call->first, last = call->last;
-	bool free_wb;
 
 	_enter("{%x:%u},{%lx-%lx}",
 	       vnode->fid.vid, vnode->fid.vnode, first, last);
 
-	ASSERT(wb != NULL);
-
 	pagevec_init(&pv, 0);
 
 	do {
@@ -651,35 +671,19 @@ void afs_pages_written_back(struct afs_vnode *vnode, struct afs_call *call)
 		count = last - first + 1;
 		if (count > PAGEVEC_SIZE)
 			count = PAGEVEC_SIZE;
-		pv.nr = find_get_pages_contig(call->mapping, first, count,
-					      pv.pages);
+		pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
+					      first, count, pv.pages);
 		ASSERTCMP(pv.nr, ==, count);
 
-		spin_lock(&vnode->writeback_lock);
 		for (loop = 0; loop < count; loop++) {
-			struct page *page = pv.pages[loop];
-			end_page_writeback(page);
-			if (page_private(page) == (unsigned long) wb) {
-				set_page_private(page, 0);
-				ClearPagePrivate(page);
-				wb->usage--;
-			}
+			set_page_private(pv.pages[loop], 0);
+			end_page_writeback(pv.pages[loop]);
 		}
-		free_wb = false;
-		if (wb->usage == 0) {
-			afs_unlink_writeback(wb);
-			free_wb = true;
-		}
-		spin_unlock(&vnode->writeback_lock);
 		first += count;
-		if (free_wb) {
-			afs_free_writeback(wb);
-			wb = NULL;
-		}
-
 		__pagevec_release(&pv);
 	} while (first <= last);
 
+	afs_prune_wb_keys(vnode);
 	_leave("");
 }
 
@@ -710,28 +714,6 @@ ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
 	return result;
 }
 
-/*
- * flush the vnode to the fileserver
- */
-int afs_writeback_all(struct afs_vnode *vnode)
-{
-	struct address_space *mapping = vnode->vfs_inode.i_mapping;
-	struct writeback_control wbc = {
-		.sync_mode	= WB_SYNC_ALL,
-		.nr_to_write	= LONG_MAX,
-		.range_cyclic	= 1,
-	};
-	int ret;
-
-	_enter("");
-
-	ret = mapping->a_ops->writepages(mapping, &wbc);
-	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
-
-	_leave(" = %d", ret);
-	return ret;
-}
-
 /*
  * flush any dirty pages for this process, and check for write errors.
  * - the return status from this call provides a reliable indication of
@@ -740,61 +722,13 @@ int afs_writeback_all(struct afs_vnode *vnode)
 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 {
 	struct inode *inode = file_inode(file);
-	struct afs_writeback *wb, *xwb;
 	struct afs_vnode *vnode = AFS_FS_I(inode);
-	int ret;
 
 	_enter("{%x:%u},{n=%pD},%d",
 	       vnode->fid.vid, vnode->fid.vnode, file,
 	       datasync);
 
-	ret = file_write_and_wait_range(file, start, end);
-	if (ret)
-		return ret;
-	inode_lock(inode);
-
-	/* use a writeback record as a marker in the queue - when this reaches
-	 * the front of the queue, all the outstanding writes are either
-	 * completed or rejected */
-	wb = kzalloc(sizeof(*wb), GFP_KERNEL);
-	if (!wb) {
-		ret = -ENOMEM;
-		goto out;
-	}
-	wb->vnode = vnode;
-	wb->first = 0;
-	wb->last = -1;
-	wb->offset_first = 0;
-	wb->to_last = PAGE_SIZE;
-	wb->usage = 1;
-	wb->state = AFS_WBACK_SYNCING;
-	init_waitqueue_head(&wb->waitq);
-
-	spin_lock(&vnode->writeback_lock);
-	list_for_each_entry(xwb, &vnode->writebacks, link) {
-		if (xwb->state == AFS_WBACK_PENDING)
-			xwb->state = AFS_WBACK_CONFLICTING;
-	}
-	list_add_tail(&wb->link, &vnode->writebacks);
-	spin_unlock(&vnode->writeback_lock);
-
-	/* push all the outstanding writebacks to the server */
-	ret = afs_writeback_all(vnode);
-	if (ret < 0) {
-		afs_put_writeback(wb);
-		_leave(" = %d [wb]", ret);
-		goto out;
-	}
-
-	/* wait for the preceding writes to actually complete */
-	ret = wait_event_interruptible(wb->waitq,
-				       wb->state == AFS_WBACK_COMPLETE ||
-				       vnode->writebacks.next == &wb->link);
-	afs_put_writeback(wb);
-	_leave(" = %d", ret);
-out:
-	inode_unlock(inode);
-	return ret;
+	return file_write_and_wait_range(file, start, end);
 }
 
 /*
@@ -831,3 +765,68 @@ int afs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
 	_leave(" = 0");
 	return 0;
 }
+
+/*
+ * Prune the keys cached for writeback.  The caller must hold vnode->wb_lock.
+ */
+void afs_prune_wb_keys(struct afs_vnode *vnode)
+{
+	LIST_HEAD(graveyard);
+	struct afs_wb_key *wbk, *tmp;
+
+	/* Discard unused keys */
+	spin_lock(&vnode->wb_lock);
+
+	if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
+	    !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
+		list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
+			if (refcount_read(&wbk->usage) == 1)
+				list_move(&wbk->vnode_link, &graveyard);
+		}
+	}
+
+	spin_unlock(&vnode->wb_lock);
+
+	while (!list_empty(&graveyard)) {
+		wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
+		list_del(&wbk->vnode_link);
+		afs_put_wb_key(wbk);
+	}
+}
+
+/*
+ * Clean up a page during invalidation.
+ */
+int afs_launder_page(struct page *page)
+{
+	struct address_space *mapping = page->mapping;
+	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
+	unsigned long priv;
+	unsigned int f, t;
+	int ret = 0;
+
+	_enter("{%lx}", page->index);
+
+	priv = page_private(page);
+	if (clear_page_dirty_for_io(page)) {
+		f = 0;
+		t = PAGE_SIZE;
+		if (PagePrivate(page)) {
+			f = priv & AFS_PRIV_MAX;
+			t = priv >> AFS_PRIV_SHIFT;
+		}
+
+		ret = afs_store_data(mapping, page->index, page->index, t, f);
+	}
+
+	set_page_private(page, 0);
+	ClearPagePrivate(page);
+
+#ifdef CONFIG_AFS_FSCACHE
+	if (PageFsCache(page)) {
+		fscache_wait_on_page_write(vnode->cache, page);
+		fscache_uncache_page(vnode->cache, page);
+	}
+#endif
+	return ret;
+}

mm/filemap.c

@@ -1041,6 +1041,7 @@ int wait_on_page_bit_killable(struct page *page, int bit_nr)
 	wait_queue_head_t *q = page_waitqueue(page);
 	return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
 }
+EXPORT_SYMBOL(wait_on_page_bit_killable);
 
 /**
  * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue