| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <705A22C6-4583-47D4-8FB0-B2820F7051F3@dilger.ca>
Date: Fri, 14 Jul 2017 14:13:16 -0700
From: Andreas Dilger <adilger@...ger.ca>
To: Tahsin Erdogan <tahsin@...gle.com>
Cc: Theodore Ts'o <tytso@....edu>, linux-ext4@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] ext4: make xattr inode reads faster
On Jul 12, 2017, at 1:29 AM, Tahsin Erdogan <tahsin@...gle.com> wrote:
>
> ext4_xattr_inode_read() currently reads each block sequentially while
> waiting for io operation to complete before moving on to the next
> block. This prevents request merging in block layer.
>
> Fix this by starting reads for all blocks then wait for completions.
It surprises me that ext4/VFS doesn't already have a helper routine to do this.
It looks like ext4_find_entry() is doing something similar for read-ahead of
directory blocks, so it may be worthwhile to consider moving that code over
to use ext4_bread_batch(), but it looks like it would need to add a flag for
whether to wait on the buffers to be uptodate or not.
> Signed-off-by: Tahsin Erdogan <tahsin@...gle.com>
> ---
> fs/ext4/ext4.h | 2 ++
> fs/ext4/inode.c | 36 ++++++++++++++++++++++++++++++++++++
> fs/ext4/xattr.c | 50 +++++++++++++++++++++++++++++++-------------------
> 3 files changed, 69 insertions(+), 19 deletions(-)
>
> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> index 9ebde0cd632e..12f0a16ad500 100644
> --- a/fs/ext4/ext4.h
> +++ b/fs/ext4/ext4.h
> @@ -2462,6 +2462,8 @@ extern void ext4_process_freed_data(struct super_block *sb, tid_t commit_tid);
> int ext4_inode_is_fast_symlink(struct inode *inode);
> struct buffer_head *ext4_getblk(handle_t *, struct inode *, ext4_lblk_t, int);
> struct buffer_head *ext4_bread(handle_t *, struct inode *, ext4_lblk_t, int);
> +int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
> + struct buffer_head **bhs);
> int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
> struct buffer_head *bh_result, int create);
> int ext4_get_block(struct inode *inode, sector_t iblock,
> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index 3c600f02673f..5b8ae1b66f09 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -1015,6 +1015,42 @@ struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
> return ERR_PTR(-EIO);
> }
>
> +/* Read a contiguous batch of blocks. */
> +int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
> + struct buffer_head **bhs)
> +{
> + int i, err;
> +
> + for (i = 0; i < bh_count; i++) {
> + bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
It's nice to have the parameter annotation, since it can often be confusing if
there are multiple int/bool arguments to a function, but, I'd write this as:
ret = ext4_getblk(NULL, inode, block + i, /* map_flags = */ 0);
I don't know if there is some kind of convention for this kind of comment?
> + if (IS_ERR(bhs[i])) {
> + err = PTR_ERR(bhs[i]);
> + while (i--)
> + brelse(bhs[i]);
It would be better to do the error cleanup once at the end of the function,
instead of multiple times in the code, like:
if (IS_ERR(bhs[i])) {
err = PTR_ERR(bhs[i]);
bh_count = i;
goto out_brelse;
}
:
:
out_brelse:
for (i = 0; i < bh_count; i++) {
brelse[bhs[i]);
bhs[i] = NULL;
}
return err;
The "bhs[i] = NULL" isn't strictly necessary, but better to not leave stray
pointers around in this array.
> + return err;
> + }
> + }
> +
> + for (i = 0; i < bh_count; i++)
> + /* Note that NULL bhs[i] is valid because of holes. */
> + if (bhs[i] && !buffer_uptodate(bhs[i]))
> + ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
> + &bhs[i]);
> +
> + for (i = 0; i < bh_count; i++)
> + if (bhs[i])
> + wait_on_buffer(bhs[i]);
> +
> + for (i = 0; i < bh_count; i++) {
> + if (bhs[i] && !buffer_uptodate(bhs[i])) {
> + for (i = 0; i < bh_count; i++)
> + brelse(bhs[i]);
The "out_brelse:" label could also be used here.
> + return -EIO;
> + }
> + }
> + return 0;
> +}
> +
> int ext4_walk_page_buffers(handle_t *handle,
> struct buffer_head *head,
> unsigned from,
> diff --git a/fs/ext4/xattr.c b/fs/ext4/xattr.c
> index cff4f41ced61..f7364a842ff4 100644
> --- a/fs/ext4/xattr.c
> +++ b/fs/ext4/xattr.c
> @@ -317,28 +317,40 @@ static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
> */
> static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
> {
> - unsigned long block = 0;
> - struct buffer_head *bh;
> - int blocksize = ea_inode->i_sb->s_blocksize;
> - size_t csize, copied = 0;
> - void *copy_pos = buf;
> -
> - while (copied < size) {
> - csize = (size - copied) > blocksize ? blocksize : size - copied;
> - bh = ext4_bread(NULL, ea_inode, block, 0);
> - if (IS_ERR(bh))
> - return PTR_ERR(bh);
> - if (!bh)
> - return -EFSCORRUPTED;
> + int blocksize = 1 << ea_inode->i_blkbits;
> + int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
> + int tail_size = (size % blocksize) ?: blocksize;
> + struct buffer_head *bhs_inline[8];
> + struct buffer_head **bhs = bhs_inline;
> + int i, ret;
> +
> + if (bh_count > ARRAY_SIZE(bhs_inline)) {
> + bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
> + if (!bhs)
> + return -ENOMEM;
> + }
>
> - memcpy(copy_pos, bh->b_data, csize);
> - brelse(bh);
> + ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count, bhs);
ret = ext4_bread_batch(ea_inode, /* block = */ 0, bh_count, bhs); ?
> + if (ret)
> + goto free_bhs;
>
> - copy_pos += csize;
> - block += 1;
> - copied += csize;
> + for (i = 0; i < bh_count; i++) {
> + /* There shouldn't be any holes in ea_inode. */
> + if (!bhs[i]) {
> + ret = -EFSCORRUPTED;
> + goto put_bhs;
> + }
> + memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
> + i < bh_count - 1 ? blocksize : tail_size);
> }
> - return 0;
> + ret = 0;
> +put_bhs:
> + for (i = 0; i < bh_count; i++)
> + brelse(bhs[i]);
> +free_bhs:
> + if (bhs != bhs_inline)
> + kfree(bhs);
> + return ret;
> }
>
> static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
> --
> 2.13.2.932.g7449e964c-goog
>
Cheers, Andreas
Download attachment "signature.asc" of type "application/pgp-signature" (196 bytes)
Powered by blists - more mailing lists