afs: Locally edit directory data for mkdir/create/unlink/...

Locally edit the contents of an AFS directory upon a successful inode
operation that modifies that directory (such as mkdir, create and unlink)
so that we can avoid the current practice of re-downloading the directory
after each change.

This is viable provided that the directory version number we get back from
the modifying RPC op is exactly incremented by 1 from what we had
previously.  The data in the directory contents is in a defined format that
we have to parse locally to perform lookups and readdir, so modifying isn't
a problem.

If the edit fails, we just clear the VALID flag on the directory and it
will be reloaded next time it is needed.

Signed-off-by: David Howells <dhowells@redhat.com>

fs/afs/Makefile

@@ -12,6 +12,7 @@ kafs-objs := \
 	cell.o \
 	cmservice.o \
 	dir.o \
+	dir_edit.o \
 	dynroot.o \
 	file.o \
 	flock.o \

fs/afs/dir.c

@@ -130,17 +130,26 @@ static bool afs_dir_check_page(struct afs_vnode *dvnode, struct page *page,
 	qty /= sizeof(union afs_xdr_dir_block);
 
 	/* check them */
-	dbuf = page_address(page);
+	dbuf = kmap(page);
 	for (tmp = 0; tmp < qty; tmp++) {
 		if (dbuf->blocks[tmp].hdr.magic != AFS_DIR_MAGIC) {
 			printk("kAFS: %s(%lx): bad magic %d/%d is %04hx\n",
 			       __func__, dvnode->vfs_inode.i_ino, tmp, qty,
 			       ntohs(dbuf->blocks[tmp].hdr.magic));
 			trace_afs_dir_check_failed(dvnode, off, i_size);
+			kunmap(page);
 			goto error;
 		}
+
+		/* Make sure each block is NUL terminated so we can reasonably
+		 * use string functions on it.  The filenames in the page
+		 * *should* be NUL-terminated anyway.
+		 */
+		((u8 *)&dbuf->blocks[tmp])[AFS_DIR_BLOCK_SIZE - 1] = 0;
 	}
 
+	kunmap(page);
+
 checked:
 	afs_stat_v(dvnode, n_read_dir);
 	return true;
@@ -1114,6 +1123,7 @@ static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 	struct afs_vnode *dvnode = AFS_FS_I(dir);
 	struct afs_fid newfid;
 	struct key *key;
+	u64 data_version = dvnode->status.data_version;
 	int ret;
 
 	mode |= S_IFDIR;
@@ -1131,7 +1141,7 @@ static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
 		while (afs_select_fileserver(&fc)) {
 			fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
-			afs_fs_create(&fc, dentry->d_name.name, mode,
+			afs_fs_create(&fc, dentry->d_name.name, mode, data_version,
 				      &newfid, &newstatus, &newcb);
 		}
 
@@ -1145,6 +1155,11 @@ static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 		goto error_key;
 	}
 
+	if (ret == 0 &&
+	    test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
+		afs_edit_dir_add(dvnode, &dentry->d_name, &newfid,
+				 afs_edit_dir_for_create);
+
 	key_put(key);
 	_leave(" = 0");
 	return 0;
@@ -1168,6 +1183,7 @@ static void afs_dir_remove_subdir(struct dentry *dentry)
 		clear_nlink(&vnode->vfs_inode);
 		set_bit(AFS_VNODE_DELETED, &vnode->flags);
 		clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
+		clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
 	}
 }
 
@@ -1179,6 +1195,7 @@ static int afs_rmdir(struct inode *dir, struct dentry *dentry)
 	struct afs_fs_cursor fc;
 	struct afs_vnode *dvnode = AFS_FS_I(dir);
 	struct key *key;
+	u64 data_version = dvnode->status.data_version;
 	int ret;
 
 	_enter("{%x:%u},{%pd}",
@@ -1194,13 +1211,18 @@ static int afs_rmdir(struct inode *dir, struct dentry *dentry)
 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
 		while (afs_select_fileserver(&fc)) {
 			fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
-			afs_fs_remove(&fc, dentry->d_name.name, true);
+			afs_fs_remove(&fc, dentry->d_name.name, true,
+				      data_version);
 		}
 
 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
 		ret = afs_end_vnode_operation(&fc);
-		if (ret == 0)
+		if (ret == 0) {
 			afs_dir_remove_subdir(dentry);
+			if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
+				afs_edit_dir_remove(dvnode, &dentry->d_name,
+						    afs_edit_dir_for_rmdir);
+		}
 	}
 
 	key_put(key);
@@ -1265,6 +1287,7 @@ static int afs_unlink(struct inode *dir, struct dentry *dentry)
 	struct afs_vnode *dvnode = AFS_FS_I(dir), *vnode;
 	struct key *key;
 	unsigned long d_version = (unsigned long)dentry->d_fsdata;
+	u64 data_version = dvnode->status.data_version;
 	int ret;
 
 	_enter("{%x:%u},{%pd}",
@@ -1291,7 +1314,8 @@ static int afs_unlink(struct inode *dir, struct dentry *dentry)
 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
 		while (afs_select_fileserver(&fc)) {
 			fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
-			afs_fs_remove(&fc, dentry->d_name.name, false);
+			afs_fs_remove(&fc, dentry->d_name.name, false,
+				      data_version);
 		}
 
 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
@@ -1300,6 +1324,10 @@ static int afs_unlink(struct inode *dir, struct dentry *dentry)
 			ret = afs_dir_remove_link(
 				dentry, key, d_version,
 				(unsigned long)dvnode->status.data_version);
+		if (ret == 0 &&
+		    test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
+			afs_edit_dir_remove(dvnode, &dentry->d_name,
+					    afs_edit_dir_for_unlink);
 	}
 
 error_key:
@@ -1321,6 +1349,7 @@ static int afs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 	struct afs_vnode *dvnode = AFS_FS_I(dir);
 	struct afs_fid newfid;
 	struct key *key;
+	u64 data_version = dvnode->status.data_version;
 	int ret;
 
 	mode |= S_IFREG;
@@ -1342,7 +1371,7 @@ static int afs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
 		while (afs_select_fileserver(&fc)) {
 			fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
-			afs_fs_create(&fc, dentry->d_name.name, mode,
+			afs_fs_create(&fc, dentry->d_name.name, mode, data_version,
 				      &newfid, &newstatus, &newcb);
 		}
 
@@ -1356,6 +1385,10 @@ static int afs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 		goto error_key;
 	}
 
+	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
+		afs_edit_dir_add(dvnode, &dentry->d_name, &newfid,
+				 afs_edit_dir_for_create);
+
 	key_put(key);
 	_leave(" = 0");
 	return 0;
@@ -1377,10 +1410,12 @@ static int afs_link(struct dentry *from, struct inode *dir,
 	struct afs_fs_cursor fc;
 	struct afs_vnode *dvnode, *vnode;
 	struct key *key;
+	u64 data_version;
 	int ret;
 
 	vnode = AFS_FS_I(d_inode(from));
 	dvnode = AFS_FS_I(dir);
+	data_version = dvnode->status.data_version;
 
 	_enter("{%x:%u},{%x:%u},{%pd}",
 	       vnode->fid.vid, vnode->fid.vnode,
@@ -1407,7 +1442,7 @@ static int afs_link(struct dentry *from, struct inode *dir,
 		while (afs_select_fileserver(&fc)) {
 			fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
 			fc.cb_break_2 = vnode->cb_break + vnode->cb_s_break;
-			afs_fs_link(&fc, vnode, dentry->d_name.name);
+			afs_fs_link(&fc, vnode, dentry->d_name.name, data_version);
 		}
 
 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
@@ -1423,6 +1458,10 @@ static int afs_link(struct dentry *from, struct inode *dir,
 		goto error_key;
 	}
 
+	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
+		afs_edit_dir_add(dvnode, &dentry->d_name, &vnode->fid,
+				 afs_edit_dir_for_link);
+
 	key_put(key);
 	_leave(" = 0");
 	return 0;
@@ -1446,6 +1485,7 @@ static int afs_symlink(struct inode *dir, struct dentry *dentry,
 	struct afs_vnode *dvnode = AFS_FS_I(dir);
 	struct afs_fid newfid;
 	struct key *key;
+	u64 data_version = dvnode->status.data_version;
 	int ret;
 
 	_enter("{%x:%u},{%pd},%s",
@@ -1470,7 +1510,8 @@ static int afs_symlink(struct inode *dir, struct dentry *dentry,
 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
 		while (afs_select_fileserver(&fc)) {
 			fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
-			afs_fs_symlink(&fc, dentry->d_name.name, content,
+			afs_fs_symlink(&fc, dentry->d_name.name,
+				       content, data_version,
 				       &newfid, &newstatus);
 		}
 
@@ -1484,6 +1525,10 @@ static int afs_symlink(struct inode *dir, struct dentry *dentry,
 		goto error_key;
 	}
 
+	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
+		afs_edit_dir_add(dvnode, &dentry->d_name, &newfid,
+				 afs_edit_dir_for_symlink);
+
 	key_put(key);
 	_leave(" = 0");
 	return 0;
@@ -1506,6 +1551,8 @@ static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	struct afs_fs_cursor fc;
 	struct afs_vnode *orig_dvnode, *new_dvnode, *vnode;
 	struct key *key;
+	u64 orig_data_version, new_data_version;
+	bool new_negative = d_is_negative(new_dentry);
 	int ret;
 
 	if (flags)
@@ -1514,6 +1561,8 @@ static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	vnode = AFS_FS_I(d_inode(old_dentry));
 	orig_dvnode = AFS_FS_I(old_dir);
 	new_dvnode = AFS_FS_I(new_dir);
+	orig_data_version = orig_dvnode->status.data_version;
+	new_data_version = new_dvnode->status.data_version;
 
 	_enter("{%x:%u},{%x:%u},{%x:%u},{%pd}",
 	       orig_dvnode->fid.vid, orig_dvnode->fid.vnode,
@@ -1539,7 +1588,8 @@ static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
 			fc.cb_break = orig_dvnode->cb_break + orig_dvnode->cb_s_break;
 			fc.cb_break_2 = new_dvnode->cb_break + new_dvnode->cb_s_break;
 			afs_fs_rename(&fc, old_dentry->d_name.name,
-				      new_dvnode, new_dentry->d_name.name);
+				      new_dvnode, new_dentry->d_name.name,
+				      orig_data_version, new_data_version);
 		}
 
 		afs_vnode_commit_status(&fc, orig_dvnode, fc.cb_break);
@@ -1551,6 +1601,21 @@ static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
 			goto error_key;
 	}
 
+	if (ret == 0) {
+		if (test_bit(AFS_VNODE_DIR_VALID, &orig_dvnode->flags))
+		    afs_edit_dir_remove(orig_dvnode, &old_dentry->d_name,
+					afs_edit_dir_for_rename);
+
+		if (!new_negative &&
+		    test_bit(AFS_VNODE_DIR_VALID, &new_dvnode->flags))
+			afs_edit_dir_remove(new_dvnode, &new_dentry->d_name,
+					    afs_edit_dir_for_rename);
+
+		if (test_bit(AFS_VNODE_DIR_VALID, &new_dvnode->flags))
+			afs_edit_dir_add(new_dvnode, &new_dentry->d_name,
+					 &vnode->fid,  afs_edit_dir_for_rename);
+	}
+
 error_key:
 	key_put(key);
 error:

fs/afs/dir_edit.c

@@ -0,0 +1,505 @@
+/* AFS filesystem directory editing
+ *
+ * Copyright (C) 2018 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public Licence
+ * as published by the Free Software Foundation; either version
+ * 2 of the Licence, or (at your option) any later version.
+ */
+
+#include <linux/kernel.h>
+#include <linux/fs.h>
+#include <linux/namei.h>
+#include <linux/pagemap.h>
+#include <linux/iversion.h>
+#include "internal.h"
+#include "xdr_fs.h"
+
+/*
+ * Find a number of contiguous clear bits in a directory block bitmask.
+ *
+ * There are 64 slots, which means we can load the entire bitmap into a
+ * variable.  The first bit doesn't count as it corresponds to the block header
+ * slot.  nr_slots is between 1 and 9.
+ */
+static int afs_find_contig_bits(union afs_xdr_dir_block *block, unsigned int nr_slots)
+{
+	u64 bitmap;
+	u32 mask;
+	int bit, n;
+
+	bitmap  = (u64)block->hdr.bitmap[0] << 0 * 8;
+	bitmap |= (u64)block->hdr.bitmap[1] << 1 * 8;
+	bitmap |= (u64)block->hdr.bitmap[2] << 2 * 8;
+	bitmap |= (u64)block->hdr.bitmap[3] << 3 * 8;
+	bitmap |= (u64)block->hdr.bitmap[4] << 4 * 8;
+	bitmap |= (u64)block->hdr.bitmap[5] << 5 * 8;
+	bitmap |= (u64)block->hdr.bitmap[6] << 6 * 8;
+	bitmap |= (u64)block->hdr.bitmap[7] << 7 * 8;
+	bitmap >>= 1; /* The first entry is metadata */
+	bit = 1;
+	mask = (1 << nr_slots) - 1;
+
+	do {
+		if (sizeof(unsigned long) == 8)
+			n = ffz(bitmap);
+		else
+			n = ((u32)bitmap) != 0 ?
+				ffz((u32)bitmap) :
+				ffz((u32)(bitmap >> 32)) + 32;
+		bitmap >>= n;
+		bit += n;
+
+		if ((bitmap & mask) == 0) {
+			if (bit > 64 - nr_slots)
+				return -1;
+			return bit;
+		}
+
+		n = __ffs(bitmap);
+		bitmap >>= n;
+		bit += n;
+	} while (bitmap);
+
+	return -1;
+}
+
+/*
+ * Set a number of contiguous bits in the directory block bitmap.
+ */
+static void afs_set_contig_bits(union afs_xdr_dir_block *block,
+				int bit, unsigned int nr_slots)
+{
+	u64 mask, before, after;
+
+	mask = (1 << nr_slots) - 1;
+	mask <<= bit;
+
+	before = *(u64 *)block->hdr.bitmap;
+
+	block->hdr.bitmap[0] |= (u8)(mask >> 0 * 8);
+	block->hdr.bitmap[1] |= (u8)(mask >> 1 * 8);
+	block->hdr.bitmap[2] |= (u8)(mask >> 2 * 8);
+	block->hdr.bitmap[3] |= (u8)(mask >> 3 * 8);
+	block->hdr.bitmap[4] |= (u8)(mask >> 4 * 8);
+	block->hdr.bitmap[5] |= (u8)(mask >> 5 * 8);
+	block->hdr.bitmap[6] |= (u8)(mask >> 6 * 8);
+	block->hdr.bitmap[7] |= (u8)(mask >> 7 * 8);
+
+	after = *(u64 *)block->hdr.bitmap;
+}
+
+/*
+ * Clear a number of contiguous bits in the directory block bitmap.
+ */
+static void afs_clear_contig_bits(union afs_xdr_dir_block *block,
+				  int bit, unsigned int nr_slots)
+{
+	u64 mask, before, after;
+
+	mask = (1 << nr_slots) - 1;
+	mask <<= bit;
+
+	before = *(u64 *)block->hdr.bitmap;
+
+	block->hdr.bitmap[0] &= ~(u8)(mask >> 0 * 8);
+	block->hdr.bitmap[1] &= ~(u8)(mask >> 1 * 8);
+	block->hdr.bitmap[2] &= ~(u8)(mask >> 2 * 8);
+	block->hdr.bitmap[3] &= ~(u8)(mask >> 3 * 8);
+	block->hdr.bitmap[4] &= ~(u8)(mask >> 4 * 8);
+	block->hdr.bitmap[5] &= ~(u8)(mask >> 5 * 8);
+	block->hdr.bitmap[6] &= ~(u8)(mask >> 6 * 8);
+	block->hdr.bitmap[7] &= ~(u8)(mask >> 7 * 8);
+
+	after = *(u64 *)block->hdr.bitmap;
+}
+
+/*
+ * Scan a directory block looking for a dirent of the right name.
+ */
+static int afs_dir_scan_block(union afs_xdr_dir_block *block, struct qstr *name,
+			      unsigned int blocknum)
+{
+	union afs_xdr_dirent *de;
+	u64 bitmap;
+	int d, len, n;
+
+	_enter("");
+
+	bitmap  = (u64)block->hdr.bitmap[0] << 0 * 8;
+	bitmap |= (u64)block->hdr.bitmap[1] << 1 * 8;
+	bitmap |= (u64)block->hdr.bitmap[2] << 2 * 8;
+	bitmap |= (u64)block->hdr.bitmap[3] << 3 * 8;
+	bitmap |= (u64)block->hdr.bitmap[4] << 4 * 8;
+	bitmap |= (u64)block->hdr.bitmap[5] << 5 * 8;
+	bitmap |= (u64)block->hdr.bitmap[6] << 6 * 8;
+	bitmap |= (u64)block->hdr.bitmap[7] << 7 * 8;
+
+	for (d = (blocknum == 0 ? AFS_DIR_RESV_BLOCKS0 : AFS_DIR_RESV_BLOCKS);
+	     d < AFS_DIR_SLOTS_PER_BLOCK;
+	     d++) {
+		if (!((bitmap >> d) & 1))
+			continue;
+		de = &block->dirents[d];
+		if (de->u.valid != 1)
+			continue;
+
+		/* The block was NUL-terminated by afs_dir_check_page(). */
+		len = strlen(de->u.name);
+		if (len == name->len &&
+		    memcmp(de->u.name, name->name, name->len) == 0)
+			return d;
+
+		n = round_up(12 + len + 1 + 4, AFS_DIR_DIRENT_SIZE);
+		n /= AFS_DIR_DIRENT_SIZE;
+		d += n - 1;
+	}
+
+	return -1;
+}
+
+/*
+ * Initialise a new directory block.  Note that block 0 is special and contains
+ * some extra metadata.
+ */
+static void afs_edit_init_block(union afs_xdr_dir_block *meta,
+				union afs_xdr_dir_block *block, int block_num)
+{
+	memset(block, 0, sizeof(*block));
+	block->hdr.npages = htons(1);
+	block->hdr.magic = AFS_DIR_MAGIC;
+	block->hdr.bitmap[0] = 1;
+
+	if (block_num == 0) {
+		block->hdr.bitmap[0] = 0xff;
+		block->hdr.bitmap[1] = 0x1f;
+		memset(block->meta.alloc_ctrs,
+		       AFS_DIR_SLOTS_PER_BLOCK,
+		       sizeof(block->meta.alloc_ctrs));
+		meta->meta.alloc_ctrs[0] =
+			AFS_DIR_SLOTS_PER_BLOCK - AFS_DIR_RESV_BLOCKS0;
+	}
+
+	if (block_num < AFS_DIR_BLOCKS_WITH_CTR)
+		meta->meta.alloc_ctrs[block_num] =
+			AFS_DIR_SLOTS_PER_BLOCK - AFS_DIR_RESV_BLOCKS;
+}
+
+/*
+ * Edit a directory's file data to add a new directory entry.  Doing this after
+ * create, mkdir, symlink, link or rename if the data version number is
+ * incremented by exactly one avoids the need to re-download the entire
+ * directory contents.
+ *
+ * The caller must hold the inode locked.
+ */
+void afs_edit_dir_add(struct afs_vnode *vnode,
+		      struct qstr *name, struct afs_fid *new_fid,
+		      enum afs_edit_dir_reason why)
+{
+	union afs_xdr_dir_block *meta, *block;
+	struct afs_xdr_dir_page *meta_page, *dir_page;
+	union afs_xdr_dirent *de;
+	struct page *page0, *page;
+	unsigned int need_slots, nr_blocks, b;
+	pgoff_t index;
+	loff_t i_size;
+	gfp_t gfp;
+	int slot;
+
+	_enter(",,{%d,%s},", name->len, name->name);
+
+	i_size = i_size_read(&vnode->vfs_inode);
+	if (i_size > AFS_DIR_BLOCK_SIZE * AFS_DIR_MAX_BLOCKS ||
+	    (i_size & (AFS_DIR_BLOCK_SIZE - 1))) {
+		clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
+		return;
+	}
+
+	gfp = vnode->vfs_inode.i_mapping->gfp_mask;
+	page0 = find_or_create_page(vnode->vfs_inode.i_mapping, 0, gfp);
+	if (!page0) {
+		clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
+		_leave(" [fgp]");
+		return;
+	}
+
+	/* Work out how many slots we're going to need. */
+	need_slots = round_up(12 + name->len + 1 + 4, AFS_DIR_DIRENT_SIZE);
+	need_slots /= AFS_DIR_DIRENT_SIZE;
+
+	meta_page = kmap(page0);
+	meta = &meta_page->blocks[0];
+	if (i_size == 0)
+		goto new_directory;
+	nr_blocks = i_size / AFS_DIR_BLOCK_SIZE;
+
+	/* Find a block that has sufficient slots available.  Each VM page
+	 * contains two or more directory blocks.
+	 */
+	for (b = 0; b < nr_blocks + 1; b++) {
+		/* If the directory extended into a new page, then we need to
+		 * tack a new page on the end.
+		 */
+		index = b / AFS_DIR_BLOCKS_PER_PAGE;
+		if (index == 0) {
+			page = page0;
+			dir_page = meta_page;
+		} else {
+			if (nr_blocks >= AFS_DIR_MAX_BLOCKS)
+				goto error;
+			gfp = vnode->vfs_inode.i_mapping->gfp_mask;
+			page = find_or_create_page(vnode->vfs_inode.i_mapping,
+						   index, gfp);
+			if (!page)
+				goto error;
+			if (!PagePrivate(page)) {
+				set_page_private(page, 1);
+				SetPagePrivate(page);
+			}
+			dir_page = kmap(page);
+		}
+
+		/* Abandon the edit if we got a callback break. */
+		if (!test_bit(AFS_VNODE_DIR_VALID, &vnode->flags))
+			goto invalidated;
+
+		block = &dir_page->blocks[b % AFS_DIR_BLOCKS_PER_PAGE];
+
+		_debug("block %u: %2u %3u %u",
+		       b,
+		       (b < AFS_DIR_BLOCKS_WITH_CTR) ? meta->meta.alloc_ctrs[b] : 99,
+		       ntohs(block->hdr.npages),
+		       ntohs(block->hdr.magic));
+
+		/* Initialise the block if necessary. */
+		if (b == nr_blocks) {
+			_debug("init %u", b);
+			afs_edit_init_block(meta, block, b);
+			i_size_write(&vnode->vfs_inode, (b + 1) * AFS_DIR_BLOCK_SIZE);
+		}
+
+		/* Only lower dir pages have a counter in the header. */
+		if (b >= AFS_DIR_BLOCKS_WITH_CTR ||
+		    meta->meta.alloc_ctrs[b] >= need_slots) {
+			/* We need to try and find one or more consecutive
+			 * slots to hold the entry.
+			 */
+			slot = afs_find_contig_bits(block, need_slots);
+			if (slot >= 0) {
+				_debug("slot %u", slot);
+				goto found_space;
+			}
+		}
+
+		if (page != page0) {
+			unlock_page(page);
+			kunmap(page);
+			put_page(page);
+		}
+	}
+
+	/* There are no spare slots of sufficient size, yet the operation
+	 * succeeded.  Download the directory again.
+	 */
+	trace_afs_edit_dir(vnode, why, afs_edit_dir_create_nospc, 0, 0, 0, 0, name->name);
+	clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
+	goto out_unmap;
+
+new_directory:
+	afs_edit_init_block(meta, meta, 0);
+	i_size = AFS_DIR_BLOCK_SIZE;
+	i_size_write(&vnode->vfs_inode, i_size);
+	slot = AFS_DIR_RESV_BLOCKS0;
+	page = page0;
+	block = meta;
+	nr_blocks = 1;
+	b = 0;
+
+found_space:
+	/* Set the dirent slot. */
+	trace_afs_edit_dir(vnode, why, afs_edit_dir_create, b, slot,
+			   new_fid->vnode, new_fid->unique, name->name);
+	de = &block->dirents[slot];
+	de->u.valid	= 1;
+	de->u.unused[0]	= 0;
+	de->u.hash_next	= 0; // TODO: Really need to maintain this
+	de->u.vnode	= htonl(new_fid->vnode);
+	de->u.unique	= htonl(new_fid->unique);
+	memcpy(de->u.name, name->name, name->len + 1);
+	de->u.name[name->len] = 0;
+
+	/* Adjust the bitmap. */
+	afs_set_contig_bits(block, slot, need_slots);
+	if (page != page0) {
+		unlock_page(page);
+		kunmap(page);
+		put_page(page);
+	}
+
+	/* Adjust the allocation counter. */
+	if (b < AFS_DIR_BLOCKS_WITH_CTR)
+		meta->meta.alloc_ctrs[b] -= need_slots;
+
+	inode_inc_iversion_raw(&vnode->vfs_inode);
+	afs_stat_v(vnode, n_dir_cr);
+	_debug("Insert %s in %u[%u]", name->name, b, slot);
+
+out_unmap:
+	unlock_page(page0);
+	kunmap(page0);
+	put_page(page0);
+	_leave("");
+	return;
+
+invalidated:
+	trace_afs_edit_dir(vnode, why, afs_edit_dir_create_inval, 0, 0, 0, 0, name->name);
+	clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
+	if (page != page0) {
+		kunmap(page);
+		put_page(page);
+	}
+	goto out_unmap;
+
+error:
+	trace_afs_edit_dir(vnode, why, afs_edit_dir_create_error, 0, 0, 0, 0, name->name);
+	clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
+	goto out_unmap;
+}
+
+/*
+ * Edit a directory's file data to remove a new directory entry.  Doing this
+ * after unlink, rmdir or rename if the data version number is incremented by
+ * exactly one avoids the need to re-download the entire directory contents.
+ *
+ * The caller must hold the inode locked.
+ */
+void afs_edit_dir_remove(struct afs_vnode *vnode,
+			 struct qstr *name, enum afs_edit_dir_reason why)
+{
+	struct afs_xdr_dir_page *meta_page, *dir_page;
+	union afs_xdr_dir_block *meta, *block;
+	union afs_xdr_dirent *de;
+	struct page *page0, *page;
+	unsigned int need_slots, nr_blocks, b;
+	pgoff_t index;
+	loff_t i_size;
+	int slot;
+
+	_enter(",,{%d,%s},", name->len, name->name);
+
+	i_size = i_size_read(&vnode->vfs_inode);
+	if (i_size < AFS_DIR_BLOCK_SIZE ||
+	    i_size > AFS_DIR_BLOCK_SIZE * AFS_DIR_MAX_BLOCKS ||
+	    (i_size & (AFS_DIR_BLOCK_SIZE - 1))) {
+		clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
+		return;
+	}
+	nr_blocks = i_size / AFS_DIR_BLOCK_SIZE;
+
+	page0 = find_lock_page(vnode->vfs_inode.i_mapping, 0);
+	if (!page0) {
+		clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
+		_leave(" [fgp]");
+		return;
+	}
+
+	/* Work out how many slots we're going to discard. */
+	need_slots = round_up(12 + name->len + 1 + 4, AFS_DIR_DIRENT_SIZE);
+	need_slots /= AFS_DIR_DIRENT_SIZE;
+
+	meta_page = kmap(page0);
+	meta = &meta_page->blocks[0];
+
+	/* Find a page that has sufficient slots available.  Each VM page
+	 * contains two or more directory blocks.
+	 */
+	for (b = 0; b < nr_blocks; b++) {
+		index = b / AFS_DIR_BLOCKS_PER_PAGE;
+		if (index != 0) {
+			page = find_lock_page(vnode->vfs_inode.i_mapping, index);
+			if (!page)
+				goto error;
+			dir_page = kmap(page);
+		} else {
+			page = page0;
+			dir_page = meta_page;
+		}
+
+		/* Abandon the edit if we got a callback break. */
+		if (!test_bit(AFS_VNODE_DIR_VALID, &vnode->flags))
+			goto invalidated;
+
+		block = &dir_page->blocks[b % AFS_DIR_BLOCKS_PER_PAGE];
+
+		if (b > AFS_DIR_BLOCKS_WITH_CTR ||
+		    meta->meta.alloc_ctrs[b] <= AFS_DIR_SLOTS_PER_BLOCK - 1 - need_slots) {
+			slot = afs_dir_scan_block(block, name, b);
+			if (slot >= 0)
+				goto found_dirent;
+		}
+
+		if (page != page0) {
+			unlock_page(page);
+			kunmap(page);
+			put_page(page);
+		}
+	}
+
+	/* Didn't find the dirent to clobber.  Download the directory again. */
+	trace_afs_edit_dir(vnode, why, afs_edit_dir_delete_noent,
+			   0, 0, 0, 0, name->name);
+	clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
+	goto out_unmap;
+
+found_dirent:
+	de = &block->dirents[slot];
+
+	trace_afs_edit_dir(vnode, why, afs_edit_dir_delete, b, slot,
+			   ntohl(de->u.vnode), ntohl(de->u.unique),
+			   name->name);
+
+	memset(de, 0, sizeof(*de) * need_slots);
+
+	/* Adjust the bitmap. */
+	afs_clear_contig_bits(block, slot, need_slots);
+	if (page != page0) {
+		unlock_page(page);
+		kunmap(page);
+		put_page(page);
+	}
+
+	/* Adjust the allocation counter. */
+	if (b < AFS_DIR_BLOCKS_WITH_CTR)
+		meta->meta.alloc_ctrs[b] += need_slots;
+
+	inode_set_iversion_raw(&vnode->vfs_inode, vnode->status.data_version);
+	afs_stat_v(vnode, n_dir_rm);
+	_debug("Remove %s from %u[%u]", name->name, b, slot);
+
+out_unmap:
+	unlock_page(page0);
+	kunmap(page0);
+	put_page(page0);
+	_leave("");
+	return;
+
+invalidated:
+	trace_afs_edit_dir(vnode, why, afs_edit_dir_delete_inval,
+			   0, 0, 0, 0, name->name);
+	clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
+	if (page != page0) {
+		unlock_page(page);
+		kunmap(page);
+		put_page(page);
+	}
+	goto out_unmap;
+
+error:
+	trace_afs_edit_dir(vnode, why, afs_edit_dir_delete_error,
+			   0, 0, 0, 0, name->name);
+	clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
+	goto out_unmap;
+}

fs/afs/fsclient.c

@@ -107,6 +107,13 @@ void afs_update_inode_from_status(struct afs_vnode *vnode,
 			} else {
 				set_bit(AFS_VNODE_ZAP_DATA, &vnode->flags);
 			}
+		} else if (vnode->status.type == AFS_FTYPE_DIR) {
+			/* Expected directory change is handled elsewhere so
+			 * that we can locally edit the directory and save on a
+			 * download.
+			 */
+			if (test_bit(AFS_VNODE_DIR_VALID, &vnode->flags))
+				flags &= ~AFS_VNODE_DATA_CHANGED;
 		}
 	}
 
@@ -190,10 +197,7 @@ static int xdr_decode_AFSFetchStatus(const __be32 **_bp,
 
 	size  = (u64)ntohl(xdr->size_lo);
 	size |= (u64)ntohl(xdr->size_hi) << 32;
-	if (size != status->size) {
+	status->size = size;
-		status->size = size;
-		flags |= AFS_VNODE_DATA_CHANGED;
-	}
 
 	data_version  = (u64)ntohl(xdr->data_version_lo);
 	data_version |= (u64)ntohl(xdr->data_version_hi) << 32;
@@ -736,6 +740,7 @@ static const struct afs_call_type afs_RXFSMakeDir = {
 int afs_fs_create(struct afs_fs_cursor *fc,
 		  const char *name,
 		  umode_t mode,
+		  u64 current_data_version,
 		  struct afs_fid *newfid,
 		  struct afs_file_status *newstatus,
 		  struct afs_callback *newcb)
@@ -763,7 +768,7 @@ int afs_fs_create(struct afs_fs_cursor *fc,
 	call->reply[1] = newfid;
 	call->reply[2] = newstatus;
 	call->reply[3] = newcb;
-	call->expected_version = vnode->status.data_version;
+	call->expected_version = current_data_version + 1;
 
 	/* marshall the parameters */
 	bp = call->request;
@@ -836,7 +841,8 @@ static const struct afs_call_type afs_RXFSRemoveDir = {
 /*
  * remove a file or directory
  */
-int afs_fs_remove(struct afs_fs_cursor *fc, const char *name, bool isdir)
+int afs_fs_remove(struct afs_fs_cursor *fc, const char *name, bool isdir,
+		  u64 current_data_version)
 {
 	struct afs_vnode *vnode = fc->vnode;
 	struct afs_call *call;
@@ -858,7 +864,7 @@ int afs_fs_remove(struct afs_fs_cursor *fc, const char *name, bool isdir)
 
 	call->key = fc->key;
 	call->reply[0] = vnode;
-	call->expected_version = vnode->status.data_version;
+	call->expected_version = current_data_version + 1;
 
 	/* marshall the parameters */
 	bp = call->request;
@@ -920,7 +926,7 @@ static const struct afs_call_type afs_RXFSLink = {
  * make a hard link
  */
 int afs_fs_link(struct afs_fs_cursor *fc, struct afs_vnode *vnode,
-		const char *name)
+		const char *name, u64 current_data_version)
 {
 	struct afs_vnode *dvnode = fc->vnode;
 	struct afs_call *call;
@@ -941,7 +947,7 @@ int afs_fs_link(struct afs_fs_cursor *fc, struct afs_vnode *vnode,
 	call->key = fc->key;
 	call->reply[0] = dvnode;
 	call->reply[1] = vnode;
-	call->expected_version = vnode->status.data_version;
+	call->expected_version = current_data_version + 1;
 
 	/* marshall the parameters */
 	bp = call->request;
@@ -1009,6 +1015,7 @@ static const struct afs_call_type afs_RXFSSymlink = {
 int afs_fs_symlink(struct afs_fs_cursor *fc,
 		   const char *name,
 		   const char *contents,
+		   u64 current_data_version,
 		   struct afs_fid *newfid,
 		   struct afs_file_status *newstatus)
 {
@@ -1037,7 +1044,7 @@ int afs_fs_symlink(struct afs_fs_cursor *fc,
 	call->reply[0] = vnode;
 	call->reply[1] = newfid;
 	call->reply[2] = newstatus;
-	call->expected_version = vnode->status.data_version;
+	call->expected_version = current_data_version + 1;
 
 	/* marshall the parameters */
 	bp = call->request;
@@ -1117,7 +1124,9 @@ static const struct afs_call_type afs_RXFSRename = {
 int afs_fs_rename(struct afs_fs_cursor *fc,
 		  const char *orig_name,
 		  struct afs_vnode *new_dvnode,
-		  const char *new_name)
+		  const char *new_name,
+		  u64 current_orig_data_version,
+		  u64 current_new_data_version)
 {
 	struct afs_vnode *orig_dvnode = fc->vnode;
 	struct afs_call *call;
@@ -1145,8 +1154,8 @@ int afs_fs_rename(struct afs_fs_cursor *fc,
 	call->key = fc->key;
 	call->reply[0] = orig_dvnode;
 	call->reply[1] = new_dvnode;
-	call->expected_version = orig_dvnode->status.data_version;
+	call->expected_version = current_orig_data_version + 1;
-	call->expected_version_2 = new_dvnode->status.data_version;
+	call->expected_version_2 = current_new_data_version + 1;
 
 	/* marshall the parameters */
 	bp = call->request;

fs/afs/inode.c

@@ -395,8 +395,11 @@ int afs_validate(struct afs_vnode *vnode, struct key *key)
 	if (test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
 		if (vnode->cb_s_break != vnode->cb_interest->server->cb_s_break) {
 			vnode->cb_s_break = vnode->cb_interest->server->cb_s_break;
-		} else if (test_bit(AFS_VNODE_DIR_VALID, &vnode->flags) &&
+		} else if (vnode->status.type == AFS_FTYPE_DIR &&
-			   !test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags) &&
+			   test_bit(AFS_VNODE_DIR_VALID, &vnode->flags) &&
+			   vnode->cb_expires_at - 10 > now) {
+				valid = true;
+		} else if (!test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags) &&
 			   vnode->cb_expires_at - 10 > now) {
 				valid = true;
 		}

fs/afs/internal.h

@@ -271,6 +271,8 @@ struct afs_net {
 	atomic_t		n_inval;	/* Number of invalidations by the server */
 	atomic_t		n_relpg;	/* Number of invalidations by releasepage */
 	atomic_t		n_read_dir;	/* Number of directory pages read */
+	atomic_t		n_dir_cr;	/* Number of directory entry creation edits */
+	atomic_t		n_dir_rm;	/* Number of directory entry removal edits */
 };
 
 extern const char afs_init_sysname[];
@@ -679,6 +681,13 @@ extern const struct dentry_operations afs_fs_dentry_operations;
 
 extern void afs_d_release(struct dentry *);
 
+/*
+ * dir_edit.c
+ */
+extern void afs_edit_dir_add(struct afs_vnode *, struct qstr *, struct afs_fid *,
+			     enum afs_edit_dir_reason);
+extern void afs_edit_dir_remove(struct afs_vnode *, struct qstr *, enum afs_edit_dir_reason);
+
 /*
  * dynroot.c
  */
@@ -725,14 +734,14 @@ extern void afs_update_inode_from_status(struct afs_vnode *, struct afs_file_sta
 extern int afs_fs_fetch_file_status(struct afs_fs_cursor *, struct afs_volsync *, bool);
 extern int afs_fs_give_up_callbacks(struct afs_net *, struct afs_server *);
 extern int afs_fs_fetch_data(struct afs_fs_cursor *, struct afs_read *);
-extern int afs_fs_create(struct afs_fs_cursor *, const char *, umode_t,
+extern int afs_fs_create(struct afs_fs_cursor *, const char *, umode_t, u64,
 			 struct afs_fid *, struct afs_file_status *, struct afs_callback *);
-extern int afs_fs_remove(struct afs_fs_cursor *, const char *, bool);
+extern int afs_fs_remove(struct afs_fs_cursor *, const char *, bool, u64);
-extern int afs_fs_link(struct afs_fs_cursor *, struct afs_vnode *, const char *);
+extern int afs_fs_link(struct afs_fs_cursor *, struct afs_vnode *, const char *, u64);
-extern int afs_fs_symlink(struct afs_fs_cursor *, const char *, const char *,
+extern int afs_fs_symlink(struct afs_fs_cursor *, const char *, const char *, u64,
 			  struct afs_fid *, struct afs_file_status *);
 extern int afs_fs_rename(struct afs_fs_cursor *, const char *,
-			 struct afs_vnode *, const char *);
+			 struct afs_vnode *, const char *, u64, u64);
 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 *);

fs/afs/proc.c

@@ -918,6 +918,10 @@ static int afs_proc_stats_show(struct seq_file *m, void *v)
 
 	seq_printf(m, "dir-data: rdpg=%u\n",
 		   atomic_read(&net->n_read_dir));
+
+	seq_printf(m, "dir-edit: cr=%u rm=%u\n",
+		   atomic_read(&net->n_dir_cr),
+		   atomic_read(&net->n_dir_rm));
 	return 0;
 }
 

include/trace/events/afs.h

@@ -63,6 +63,27 @@ enum afs_vl_operation {
 	afs_VL_GetCapabilities	= 65537,	/* AFS Get VL server capabilities */
 };
 
+enum afs_edit_dir_op {
+	afs_edit_dir_create,
+	afs_edit_dir_create_error,
+	afs_edit_dir_create_inval,
+	afs_edit_dir_create_nospc,
+	afs_edit_dir_delete,
+	afs_edit_dir_delete_error,
+	afs_edit_dir_delete_inval,
+	afs_edit_dir_delete_noent,
+};
+
+enum afs_edit_dir_reason {
+	afs_edit_dir_for_create,
+	afs_edit_dir_for_link,
+	afs_edit_dir_for_mkdir,
+	afs_edit_dir_for_rename,
+	afs_edit_dir_for_rmdir,
+	afs_edit_dir_for_symlink,
+	afs_edit_dir_for_unlink,
+};
+
 #endif /* end __AFS_DECLARE_TRACE_ENUMS_ONCE_ONLY */
 
 /*
@@ -106,6 +127,25 @@ enum afs_vl_operation {
 	EM(afs_YFSVL_GetEndpoints,		"YFSVL.GetEndpoints") \
 	E_(afs_VL_GetCapabilities,		"VL.GetCapabilities")
 
+#define afs_edit_dir_ops				  \
+	EM(afs_edit_dir_create,			"create") \
+	EM(afs_edit_dir_create_error,		"c_fail") \
+	EM(afs_edit_dir_create_inval,		"c_invl") \
+	EM(afs_edit_dir_create_nospc,		"c_nspc") \
+	EM(afs_edit_dir_delete,			"delete") \
+	EM(afs_edit_dir_delete_error,		"d_err ") \
+	EM(afs_edit_dir_delete_inval,		"d_invl") \
+	E_(afs_edit_dir_delete_noent,		"d_nent")
+
+#define afs_edit_dir_reasons				  \
+	EM(afs_edit_dir_for_create,		"Create") \
+	EM(afs_edit_dir_for_link,		"Link  ") \
+	EM(afs_edit_dir_for_mkdir,		"MkDir ") \
+	EM(afs_edit_dir_for_rename,		"Rename") \
+	EM(afs_edit_dir_for_rmdir,		"RmDir ") \
+	EM(afs_edit_dir_for_symlink,		"Symlnk") \
+	E_(afs_edit_dir_for_unlink,		"Unlink")
+
 
 /*
  * Export enum symbols via userspace.
@@ -118,6 +158,8 @@ enum afs_vl_operation {
 afs_call_traces;
 afs_fs_operations;
 afs_vl_operations;
+afs_edit_dir_ops;
+afs_edit_dir_reasons;
 
 /*
  * Now redefine the EM() and E_() macros to map the enums to the strings that
@@ -464,6 +506,54 @@ TRACE_EVENT(afs_call_state,
 		      __entry->ret, __entry->abort)
 	    );
 
+TRACE_EVENT(afs_edit_dir,
+	    TP_PROTO(struct afs_vnode *dvnode,
+		     enum afs_edit_dir_reason why,
+		     enum afs_edit_dir_op op,
+		     unsigned int block,
+		     unsigned int slot,
+		     unsigned int f_vnode,
+		     unsigned int f_unique,
+		     const char *name),
+
+	    TP_ARGS(dvnode, why, op, block, slot, f_vnode, f_unique, name),
+
+	    TP_STRUCT__entry(
+		    __field(unsigned int,		vnode		)
+		    __field(unsigned int,		unique		)
+		    __field(enum afs_edit_dir_reason,	why		)
+		    __field(enum afs_edit_dir_op,	op		)
+		    __field(unsigned int,		block		)
+		    __field(unsigned short,		slot		)
+		    __field(unsigned int,		f_vnode		)
+		    __field(unsigned int,		f_unique	)
+		    __array(char,			name, 18	)
+			     ),
+
+	    TP_fast_assign(
+		    int __len = strlen(name);
+		    __len = min(__len, 17);
+		    __entry->vnode	= dvnode->fid.vnode;
+		    __entry->unique	= dvnode->fid.unique;
+		    __entry->why	= why;
+		    __entry->op		= op;
+		    __entry->block	= block;
+		    __entry->slot	= slot;
+		    __entry->f_vnode	= f_vnode;
+		    __entry->f_unique	= f_unique;
+		    memcpy(__entry->name, name, __len);
+		    __entry->name[__len] = 0;
+			   ),
+
+	    TP_printk("d=%x:%x %s %s %u[%u] f=%x:%x %s",
+		      __entry->vnode, __entry->unique,
+		      __print_symbolic(__entry->why, afs_edit_dir_reasons),
+		      __print_symbolic(__entry->op, afs_edit_dir_ops),
+		      __entry->block, __entry->slot,
+		      __entry->f_vnode, __entry->f_unique,
+		      __entry->name)
+	    );
+
 #endif /* _TRACE_AFS_H */
 
 /* This part must be outside protection */