On 05/12/2014 11:15 PM, Eric Blake wrote:

On 05/10/2014 10:35 AM, Jun Li wrote:

From: Jun Li <junmuzi@gmail.com>

I see three different "[PATCH v3]" mails with this subject.  To make
sure we are reviewing the right version, it might help to bump to version 4.

Yes, I have send v4 of this patch and changed the commit description.

This patch fixed the following bug:
https://bugzilla.redhat.com/show_bug.cgi?id=1084302 .

path_combine can not calculate the correct full path name for backing_file.
Such as:
create a snapshot chain:
sn2->sn1->$BASE_IMG
backing file is : /home/wookpecker/img.qcow2
sn1 : /home/woodpecker/tmp/sn1
sn2 : /home/woodpecker/tmp/sn2
when create sn2, path_combine can not got a correct path for $BASE_IMG.

I'm having a hard time parsing that.  We usually represent backing
chains with the base image on the left:

base <- sn1

Can you show the output of 'qemu-img info --backing-chain
/home/woodpecker/tmp/sn2' to make it a bit more obvious what your setup
is?  What command are you issuing that triggers a path_combine that is
getting the wrong result?

Sorry, this is the wrong commit description, please see it on v4. In version 4, I describe it just as followings:
create a snapshot chain:
$BASE_IMG<-sn1<-sn2
backing file is : ./img.qcow2
sn1 : ./tmp/sn1
sn2 : ./tmp/sn2

So,
# /opt/qemu-git-arm/bin/qemu-img info --backing-chain ./tmp/sn2
image: ./tmp/sn2
file format: qcow2
virtual size: 10G (10737418240 bytes)
disk size: 196K
cluster_size: 65536
backing file: ./tmp/sn1 (actual path: /home/lijun/Work/tmp/sn1)
Format specific information:
    compat: 1.1
    lazy refcounts: false

image: /home/lijun/Work/tmp/sn1
file format: qcow2
virtual size: 10G (10737418240 bytes)
disk size: 196K
cluster_size: 65536
backing file: ./img.qcow2 (actual path: /home/lijun/Work/img.qcow2)
Format specific information:
    compat: 1.1
    lazy refcounts: false

image: /home/lijun/Work/img.qcow2
file format: qcow2
virtual size: 10G (10737418240 bytes)
disk size: 196K
cluster_size: 65536
Format specific information:
    compat: 1.1
    lazy refcounts: false
------------------
Before this patch,
# qemu-img create -f qcow2 -b ./img.qcow2 ./tmp/sn1
# qemu-img create -f qcow2 -b ./tmp/sn1 ./tmp/sn2
qemu-img: Could not open './tmp/sn1': No such file or directory

In this patch, will check the backing_filename is a symlink or not firstly,
then return the full(absolute) path via realpath.

This feels fishy to me, and liable to do the wrong thing.  I need more
context on how to reproduce the issue you are attempting to fix before I
can even decide if your fix is the right approach.

This patch fixed the following bug(Bug 1084302 - Should improve error info when can't find backing file for snapshot):
https://bugzilla.redhat.com/show_bug.cgi?id=1084302

Signed-off-by: Jun Li <junmuzi@gmail.com>
---
 block.c | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

diff --git a/block.c b/block.c
index b749d31..6674719 100644
--- a/block.c
+++ b/block.c
@@ -304,10 +304,26 @@ void path_combine(char *dest, int dest_size,
 
 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz)
 {
+    struct stat sb;
+    char *linkname;
+
     if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) {
         pstrcpy(dest, sz, bs->backing_file);
     } else {
-        path_combine(dest, sz, bs->filename, bs->backing_file);
+        if (lstat(bs->backing_file, &sb) == -1) {
+            perror("lstat");
+            exit(EXIT_FAILURE);
+        }
+
+        /* Check linkname is a link or not */
+        if (S_ISLNK(sb.st_mode)) {
+            linkname = malloc(sb.st_size + 1);
+            readlink(bs->backing_file, linkname, sb.st_size + 1);
+            linkname[sb.st_size] = '\0';
+            realpath(linkname, dest);
+        } else {
+            realpath(bs->backing_file, dest);
+        }

Why are you tweaking just this caller, instead of path_combine() to
affect all other callers?

Hi Eric:

I have checked this.

$ grep -wr "path_combine" ./
./block.c:void path_combine(char *dest, int dest_size,
./block.c:        path_combine(dest, sz, bs->filename, bs->backing_file);
./block.c:            path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
./block.c:            path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
./block/vmdk.c:        path_combine(extent_path, sizeof(extent_path),
./include/block/block.h:void path_combine(char *dest, int dest_size,

path_combine() is called in above 4 places.

In function bdrv_find_backing_image(), the current realization of path_combine() is enough.

            /* If not an absolute filename path, make it relative to the current
             * image's filename path */
            path_combine(filename_tmp, PATH_MAX, backing_file);
            ...

            /* We need to make sure the backing filename we are comparing against
             * is relative to the current image filename (or absolute) */
            path_combine(filename_tmp, PATH_MAX, curr_bs->backing_file);

In function vmdk_parse_extents(), as only related to current image, so no need to change path_combine(), too.

Only bdrv_get_full_backing_filename() need to modify path_combine(). I have submit v4.

Best Regards,
Jun Li