Linux-CIFS Archive on
 help / color / Atom feed
From: Pavel Shilovsky <>
To: ronnie sahlberg <>
Cc: Steve French <>,
	Ronnie Sahlberg <>,
	linux-cifs <>
Subject: Re: [PATCH] cifs: add support for fallocate mode 0 for non-sparse files
Date: Thu, 16 Jan 2020 16:27:16 -0800
Message-ID: <> (raw)
In-Reply-To: <>

чт, 16 янв. 2020 г. в 01:05, ronnie sahlberg <>:
> The bug is basically that if we extend a file by fallocate mode==0
> and immediately afterwards mmap() the file we will only mmap() as much
> as end-of-file was
> prior to the truncate  and then if we try to touch any
> address in this extended region userspace dies with bus error.
> The patch added "extend a file with fallocate mode==0 for NON-Sparse
> files" and caused xfstest to fail.
> The fix is to force us to revalidate the file attributes (the size is
> the important one) when we extend the file so
> mmap() will work properly.
> I have fixed this in the patch and will resend tomorrow after some more testing.
> Looking for other SMB2_set_eof() callsites I see we already had the
> same bug for the case of extending a SPARSE

I agree that regardless of file being sparse or not, we should somehow
update a size in the VFS after calling SMB2_set_eof().

> file using fallocate mode==0. I fixed that too and will audit all
> other plases where we use SMB2_set_eof()
> to see if they are safe or not before resending.

One of those places where SMB2_set_eof() is called is
cifs_set_file_size() which does call the following after getting a
successful response from the server:

2250 >-------if (rc == 0) {
2251 >------->-------cifsInode->server_eof = attrs->ia_size;
2252 >------->-------cifs_setsize(inode, attrs->ia_size);
2253 >------->-------cifs_truncate_page(inode->i_mapping, inode->i_size);
2254 >-------}

This is called by cifs_setattr_[no]unix() which does the following afterwards:

2569 >-------if ((attrs->ia_valid & ATTR_SIZE) &&
2570 >-------    attrs->ia_size != i_size_read(inode))
2571 >------->-------truncate_setsize(inode, attrs->ia_size);

truncate_setsize() does  similar things as cifs_setsize() besides
setting cinode->time to 0. This code path probably needs to be
refactored. But putting this aside, for the current fallocate fix I
think we should use the same existing mechanism to update a file size
in the VFS without revalidating the inode.

Best regards,
Pavel Shilovsky

  reply index

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-01-15  1:23 [PATCH 0/1] add support for fallocate mode 0 Ronnie Sahlberg
2020-01-15  1:23 ` [PATCH] cifs: add support for fallocate mode 0 for non-sparse files Ronnie Sahlberg
2020-01-15  1:25   ` Steve French
2020-01-15  2:25     ` ronnie sahlberg
2020-01-15 20:14       ` Steve French
2020-01-16  2:03         ` Steve French
2020-01-16  8:42           ` ronnie sahlberg
2020-01-17  0:27             ` Pavel Shilovsky [this message]
2020-01-17  0:56               ` ronnie sahlberg
2020-01-17  0:04 Ronnie Sahlberg
2020-01-17  1:15 Ronnie Sahlberg

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='' \ \ \ \ \ \

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

Linux-CIFS Archive on

Archives are clonable:
	git clone --mirror linux-cifs/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-cifs linux-cifs/ \
	public-inbox-index linux-cifs

Example config snippet for mirrors

Newsgroup available over NNTP:

AGPL code for this site: git clone