yeahuh / rpms / qemu-kvm

Forked from rpms/qemu-kvm 2 years ago
Clone

Blame SOURCES/kvm-virtiofsd-Fix-data-corruption-with-O_APPEND-write-in.patch

902636
From 9b5fbc95a287b2ce9448142194b161d8360d5e4e Mon Sep 17 00:00:00 2001
902636
From: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
902636
Date: Mon, 27 Jan 2020 19:02:15 +0100
902636
Subject: [PATCH 104/116] virtiofsd: Fix data corruption with O_APPEND write in
902636
 writeback mode
902636
MIME-Version: 1.0
902636
Content-Type: text/plain; charset=UTF-8
902636
Content-Transfer-Encoding: 8bit
902636
902636
RH-Author: Dr. David Alan Gilbert <dgilbert@redhat.com>
902636
Message-id: <20200127190227.40942-101-dgilbert@redhat.com>
902636
Patchwork-id: 93556
902636
O-Subject: [RHEL-AV-8.2 qemu-kvm PATCH 100/112] virtiofsd: Fix data corruption with O_APPEND write in writeback mode
902636
Bugzilla: 1694164
902636
RH-Acked-by: Philippe Mathieu-Daudé <philmd@redhat.com>
902636
RH-Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
902636
RH-Acked-by: Sergio Lopez Pascual <slp@redhat.com>
902636
902636
From: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
902636
902636
When writeback mode is enabled (-o writeback), O_APPEND handling is
902636
done in kernel. Therefore virtiofsd clears O_APPEND flag when open.
902636
Otherwise O_APPEND flag takes precedence over pwrite() and write
902636
data may corrupt.
902636
902636
Currently clearing O_APPEND flag is done in lo_open(), but we also
902636
need the same operation in lo_create(). So, factor out the flag
902636
update operation in lo_open() to update_open_flags() and call it
902636
in both lo_open() and lo_create().
902636
902636
This fixes the failure of xfstest generic/069 in writeback mode
902636
(which tests O_APPEND write data integrity).
902636
902636
Signed-off-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
902636
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
902636
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
902636
(cherry picked from commit 8e4e41e39eac5ee5f378d66f069a2f70a1734317)
902636
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
902636
---
902636
 tools/virtiofsd/passthrough_ll.c | 66 ++++++++++++++++++++--------------------
902636
 1 file changed, 33 insertions(+), 33 deletions(-)
902636
902636
diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c
902636
index 948cb19..4c61ac5 100644
902636
--- a/tools/virtiofsd/passthrough_ll.c
902636
+++ b/tools/virtiofsd/passthrough_ll.c
902636
@@ -1692,6 +1692,37 @@ static void lo_releasedir(fuse_req_t req, fuse_ino_t ino,
902636
     fuse_reply_err(req, 0);
902636
 }
902636
 
902636
+static void update_open_flags(int writeback, struct fuse_file_info *fi)
902636
+{
902636
+    /*
902636
+     * With writeback cache, kernel may send read requests even
902636
+     * when userspace opened write-only
902636
+     */
902636
+    if (writeback && (fi->flags & O_ACCMODE) == O_WRONLY) {
902636
+        fi->flags &= ~O_ACCMODE;
902636
+        fi->flags |= O_RDWR;
902636
+    }
902636
+
902636
+    /*
902636
+     * With writeback cache, O_APPEND is handled by the kernel.
902636
+     * This breaks atomicity (since the file may change in the
902636
+     * underlying filesystem, so that the kernel's idea of the
902636
+     * end of the file isn't accurate anymore). In this example,
902636
+     * we just accept that. A more rigorous filesystem may want
902636
+     * to return an error here
902636
+     */
902636
+    if (writeback && (fi->flags & O_APPEND)) {
902636
+        fi->flags &= ~O_APPEND;
902636
+    }
902636
+
902636
+    /*
902636
+     * O_DIRECT in guest should not necessarily mean bypassing page
902636
+     * cache on host as well. If somebody needs that behavior, it
902636
+     * probably should be a configuration knob in daemon.
902636
+     */
902636
+    fi->flags &= ~O_DIRECT;
902636
+}
902636
+
902636
 static void lo_create(fuse_req_t req, fuse_ino_t parent, const char *name,
902636
                       mode_t mode, struct fuse_file_info *fi)
902636
 {
902636
@@ -1721,12 +1752,7 @@ static void lo_create(fuse_req_t req, fuse_ino_t parent, const char *name,
902636
         goto out;
902636
     }
902636
 
902636
-    /*
902636
-     * O_DIRECT in guest should not necessarily mean bypassing page
902636
-     * cache on host as well. If somebody needs that behavior, it
902636
-     * probably should be a configuration knob in daemon.
902636
-     */
902636
-    fi->flags &= ~O_DIRECT;
902636
+    update_open_flags(lo->writeback, fi);
902636
 
902636
     fd = openat(parent_inode->fd, name, (fi->flags | O_CREAT) & ~O_NOFOLLOW,
902636
                 mode);
902636
@@ -1936,33 +1962,7 @@ static void lo_open(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi)
902636
     fuse_log(FUSE_LOG_DEBUG, "lo_open(ino=%" PRIu64 ", flags=%d)\n", ino,
902636
              fi->flags);
902636
 
902636
-    /*
902636
-     * With writeback cache, kernel may send read requests even
902636
-     * when userspace opened write-only
902636
-     */
902636
-    if (lo->writeback && (fi->flags & O_ACCMODE) == O_WRONLY) {
902636
-        fi->flags &= ~O_ACCMODE;
902636
-        fi->flags |= O_RDWR;
902636
-    }
902636
-
902636
-    /*
902636
-     * With writeback cache, O_APPEND is handled by the kernel.
902636
-     * This breaks atomicity (since the file may change in the
902636
-     * underlying filesystem, so that the kernel's idea of the
902636
-     * end of the file isn't accurate anymore). In this example,
902636
-     * we just accept that. A more rigorous filesystem may want
902636
-     * to return an error here
902636
-     */
902636
-    if (lo->writeback && (fi->flags & O_APPEND)) {
902636
-        fi->flags &= ~O_APPEND;
902636
-    }
902636
-
902636
-    /*
902636
-     * O_DIRECT in guest should not necessarily mean bypassing page
902636
-     * cache on host as well. If somebody needs that behavior, it
902636
-     * probably should be a configuration knob in daemon.
902636
-     */
902636
-    fi->flags &= ~O_DIRECT;
902636
+    update_open_flags(lo->writeback, fi);
902636
 
902636
     sprintf(buf, "%i", lo_fd(req, ino));
902636
     fd = openat(lo->proc_self_fd, buf, fi->flags & ~O_NOFOLLOW);
902636
-- 
902636
1.8.3.1
902636