linux-ext4 - [PATCH 1/1] ext4: correct partial write discard size calculation

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:	Tue,  6 Dec 2011 22:19:48 +0000
From:	Andy Whitcroft <apw@...onical.com>
To:	"Theodore Ts'o" <tytso@....edu>,
	Andreas Dilger <adilger.kernel@...ger.ca>
Cc:	linux-ext4@...r.kernel.org,
	Allison Henderson <achender@...ux.vnet.ibm.com>,
	linux-kernel@...r.kernel.org, Andy Whitcroft <apw@...onical.com>
Subject: [PATCH 1/1] ext4: correct partial write discard size calculation

When copying large numbers of files we are seeing occasional write failures
with errno EINVAL.  These are being returned from ext4_da_write_end()
when attempting to discard the end portion of a partial write.  The error
is detected and reported by the page index check below:

    int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
		    struct inode *inode, struct page *page, loff_t from,
		    loff_t length, int flags)
    {
    [...]
        if (index != page->index)
                return -EINVAL;
    [...]

This code was introduced by the commit below:

  commit 02fac1297eb3f471a27368271aadd285548297b0
  Author: Allison Henderson <achender@...ux.vnet.ibm.com>
  Date:   Tue Sep 6 21:53:01 2011 -0400

    ext4: fix partial page writes

This error is triggering when a write occurs at pos == 0 and results in
0 bytes being written (copied == 0):

    page_len = PAGE_CACHE_SIZE -
                    ((pos + copied - 1) & (PAGE_CACHE_SIZE - 1));
    if (page_len > 0) {
            ret = ext4_discard_partial_page_buffers_no_lock(handle,
                    inode, page, pos + copied - 1, page_len,
    [...]

In this case we will calculate that we need to clear out only one byte of
the page.  As we are aligned at the page boundary and wrote 0 bytes we
actually need to clear the entire page.  Also note that when we attempt
to apply the discard we will apply it at offset -1 (0 + 0 - 1), which is
the wrong place:

    page_len = 4096 - ((0 + 0 - 1) & 4095)
    page_len = 1

Firstly fix up the offset calculation.  Once this is done the erroring
case will correctly believe that the entire page needs to be discarded.
However in this case we did not actually write to the page so the page
is not instantiated and no discard is required.  So also only apply the
discard where we are not discarding the entire page.

Signed-off-by: Andy Whitcroft <apw@...onical.com>
---
 fs/ext4/inode.c |    6 +++---
 1 files changed, 3 insertions(+), 3 deletions(-)

This issue is most easily reproducible within a VM on a fast, lightly
loaded host.  In that configuration I can trigger a failure with about
1/2GB of medium sized files (.debs in this case).  Without the patch
the copy will fail 'EINVAL' 99% of the time, always failing within two
iterations.  With the patch I have run 100 iterations of the same copy
without failure.

diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 240f6e2..c137168 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -2533,11 +2533,11 @@ static int ext4_da_write_end(struct file *file,
 							page, fsdata);

 	page_len = PAGE_CACHE_SIZE -
-			((pos + copied - 1) & (PAGE_CACHE_SIZE - 1));
+			((pos + copied) & (PAGE_CACHE_SIZE - 1));

-	if (page_len > 0) {
+	if (page_len > 0 && page_len < PAGE_CACHE_SIZE) {
 		ret = ext4_discard_partial_page_buffers_no_lock(handle,
-			inode, page, pos + copied - 1, page_len,
+			inode, page, pos + copied, page_len,
 			EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED);
 	}

-- 
1.7.5.4

--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html