xattr: Option to disable meta-data block cache

mbcache provides absolutely no value for Lustre xattrs (because
they are unique and cannot be shared between files) and as we can
see it has a noticable overhead in some cases. In the past there
was a CONFIG_MBCACHE option that would allow it to be disabled,
but this was removed in newer kernels, so we will need to patch
ldiskfs to fix this.

References: <https://bugzilla.kernel.org/show_bug.cgi?id=107301>
References: <https://git.hpdd.intel.com/fs/lustre-release.git/blob_plain/HEAD:/ldiskfs/kernel_patches/patches/rhel7/ext4-disable-mb-cache.patch>
References: CVE-2015-8952

On 13.12.2016 at 15:58 Ben Hutchings wrote:
> I decided not to apply this as it's a userland ABI extension that we
> would then need to carry indefinitely.
Signed-off-by: Philipp Hahn <hahn@univention.de>
Signed-off-by: Sasha Levin <alexander.levin@verizon.com>

fs/ext4/ext4.h

@@ -1008,6 +1008,7 @@ struct ext4_inode_info {
 /*
  * Mount flags set via mount options or defaults
  */
+#define EXT4_MOUNT_NO_MBCACHE		0x00001 /* Disable mbcache */
 #define EXT4_MOUNT_GRPID		0x00004	/* Create files with directory's group */
 #define EXT4_MOUNT_DEBUG		0x00008	/* Some debugging messages */
 #define EXT4_MOUNT_ERRORS_CONT		0x00010	/* Continue on errors */

fs/ext4/super.c

@@ -1144,6 +1144,7 @@ enum {
 	Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
 	Opt_inode_readahead_blks, Opt_journal_ioprio,
 	Opt_dioread_nolock, Opt_dioread_lock,
+	Opt_no_mbcache,
 	Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
 	Opt_max_dir_size_kb, Opt_nojournal_checksum,
 };
@@ -1222,6 +1223,7 @@ static const match_table_t tokens = {
 	{Opt_discard, "discard"},
 	{Opt_nodiscard, "nodiscard"},
 	{Opt_init_itable, "init_itable=%u"},
+	{Opt_no_mbcache, "no_mbcache"},
 	{Opt_init_itable, "init_itable"},
 	{Opt_noinit_itable, "noinit_itable"},
 	{Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
@@ -1385,6 +1387,7 @@ static const struct mount_opts {
 	{Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
 	{Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
 	{Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
+	{Opt_no_mbcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
 	{Opt_commit, 0, MOPT_GTE0},
 	{Opt_max_batch_time, 0, MOPT_GTE0},
 	{Opt_min_batch_time, 0, MOPT_GTE0},

fs/ext4/xattr.c

@@ -80,7 +80,7 @@
 # define ea_bdebug(bh, fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
 #endif
 
-static void ext4_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
+static void ext4_xattr_cache_insert(struct inode *, struct buffer_head *);
 static struct buffer_head *ext4_xattr_cache_find(struct inode *,
 						 struct ext4_xattr_header *,
 						 struct mb_cache_entry **);
@@ -278,7 +278,6 @@ ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 	struct ext4_xattr_entry *entry;
 	size_t size;
 	int error;
-	struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 
 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 		  name_index, name, buffer, (long)buffer_size);
@@ -300,7 +299,7 @@ ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 		error = -EIO;
 		goto cleanup;
 	}
-	ext4_xattr_cache_insert(ext4_mb_cache, bh);
+	ext4_xattr_cache_insert(inode, bh);
 	entry = BFIRST(bh);
 	error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
 	if (error == -EIO)
@@ -426,7 +425,6 @@ ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 	struct inode *inode = d_inode(dentry);
 	struct buffer_head *bh = NULL;
 	int error;
-	struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 
 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 		  buffer, (long)buffer_size);
@@ -448,7 +446,7 @@ ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 		error = -EIO;
 		goto cleanup;
 	}
-	ext4_xattr_cache_insert(ext4_mb_cache, bh);
+	ext4_xattr_cache_insert(inode, bh);
 	error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
 
 cleanup:
@@ -547,6 +545,7 @@ ext4_xattr_release_block(handle_t *handle, struct inode *inode,
 	int error = 0;
 	struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 
+	if (!test_opt(inode->i_sb, NO_MBCACHE))
 		ce = mb_cache_entry_get(ext4_mb_cache, bh->b_bdev, bh->b_blocknr);
 	BUFFER_TRACE(bh, "get_write_access");
 	error = ext4_journal_get_write_access(handle, bh);
@@ -788,6 +787,7 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 	if (i->value && i->value_len > sb->s_blocksize)
 		return -ENOSPC;
 	if (s->base) {
+		if (!test_opt(inode->i_sb, NO_MBCACHE))
 			ce = mb_cache_entry_get(ext4_mb_cache, bs->bh->b_bdev,
 						bs->bh->b_blocknr);
 		BUFFER_TRACE(bs->bh, "get_write_access");
@@ -807,7 +807,7 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 				if (!IS_LAST_ENTRY(s->first))
 					ext4_xattr_rehash(header(s->base),
 							  s->here);
-				ext4_xattr_cache_insert(ext4_mb_cache,
+				ext4_xattr_cache_insert(inode,
 					bs->bh);
 			}
 			unlock_buffer(bs->bh);
@@ -892,6 +892,7 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 				if (error)
 					goto cleanup_dquot;
 			}
+			if (ce)
 				mb_cache_entry_release(ce);
 			ce = NULL;
 		} else if (bs->bh && s->base == bs->bh->b_data) {
@@ -939,7 +940,7 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 			memcpy(new_bh->b_data, s->base, new_bh->b_size);
 			set_buffer_uptodate(new_bh);
 			unlock_buffer(new_bh);
-			ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
+			ext4_xattr_cache_insert(inode, new_bh);
 			error = ext4_handle_dirty_xattr_block(handle,
 							      inode, new_bh);
 			if (error)
@@ -1529,12 +1530,17 @@ ext4_xattr_put_super(struct super_block *sb)
  * Returns 0, or a negative error number on failure.
  */
 static void
-ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
+ext4_xattr_cache_insert(struct inode *inode, struct buffer_head *bh)
 {
+	struct super_block *sb = inode->i_sb;
+	struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 	__u32 hash = le32_to_cpu(BHDR(bh)->h_hash);
 	struct mb_cache_entry *ce;
 	int error;
 
+	if (test_opt(sb, NO_MBCACHE))
+		return;
+
 	ce = mb_cache_entry_alloc(ext4_mb_cache, GFP_NOFS);
 	if (!ce) {
 		ea_bdebug(bh, "out of memory");
@@ -1609,6 +1615,8 @@ ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
 	struct mb_cache_entry *ce;
 	struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 
+	if (test_opt(inode->i_sb, NO_MBCACHE))
+		return NULL;
 	if (!header->h_hash)
 		return NULL;  /* never share */
 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);