On which hardware can I test this? Is it possible to reproduce in qemu?

In our case, we're testing on an embedded 64GB UFS disk, I'm not sure how to get those on the market, high end Android phones? Perhaps with proper qemu device models? I'm not familiar with qemu internals.

The only part that alerted me was the dmesg claiming that the logical blocks were 4096 bytes.

The kernel refuses to mount it if formatted with 512 byte clusters, depending on EFI bootloader firmware implementations, some might work anyway.

If there is no way for me to reproduce this I'd need your help in testing my patches.

First attempt (remove -S 512 in a hope that mkdosf will decide correct sector size):

diff --git a/scripts/lib/wic/partition.py b/scripts/lib/wic/partition.py
index 5aa68c9..0633589 100644
--- a/scripts/lib/wic/partition.py
+++ b/scripts/lib/wic/partition.py
@@ -312,8 +312,8 @@ class Partition():
         if self.fstype == 'msdos':
             size_str = "-F 16" # FAT 16
 
-        dosfs_cmd = "mkdosfs %s -S 512 %s -C %s %d" % (label_str, size_str,
-                                                       rootfs, rootfs_size)
+        dosfs_cmd = "mkdosfs %s %s -C %s %d" % (label_str, size_str,
+                                                rootfs, rootfs_size)
         exec_native_cmd(dosfs_cmd, native_sysroot)
 
         mcopy_cmd = "mcopy -i %s -s %s/* ::/" % (rootfs, rootfs_dir)


Let me know please if this works for you or not.

Looks like it still uses 512 byte clusters, and then errors out with "Attempting to create a too large filesystem" with larger volumes that would have succeeded if created with 4K clusters manually.

OK, next attempt then:

diff --git a/scripts/lib/wic/partition.py b/scripts/lib/wic/partition.py
index 5aa68c9..64f622f 100644
--- a/scripts/lib/wic/partition.py
+++ b/scripts/lib/wic/partition.py
@@ -312,8 +312,8 @@ class Partition():
         if self.fstype == 'msdos':
             size_str = "-F 16" # FAT 16
 
-        dosfs_cmd = "mkdosfs %s -S 512 %s -C %s %d" % (label_str, size_str,
-                                                       rootfs, rootfs_size)
+        dosfs_cmd = "mkdosfs %s -S 512 -s 8 %s -C %s %d" % (label_str, size_str,
+                                                            rootfs, rootfs_size)
         exec_native_cmd(dosfs_cmd, native_sysroot)
 
         mcopy_cmd = "mcopy -i %s -s %s/* ::/" % (rootfs, rootfs_dir)

BTW, you mentioned that you've changed -S 512 to -S 4096 in image-live.bbclass. However, -S specifies size of logical sector. That would not always means that cluster size will be 4096. So, I left logical sector 512 and specified amount of sectors per cluster. That should ensure that cluster size will be 4096. Let me know if this works and if this is what you wanted.

Nice, -S 512 -s 8 is working too, I was confused with the description of -s and -S in the man page.

OK, then I implement --mkfs-opts option for .wks and you can specify any options you want. Let me know if it would work for you.

Please answer Ed's question.

Hi, how do I get your new --mkfs-opts changes to test them? I don't see a patch. Btw, hddimg also needs to be handled.

Hold on, no "-S 512 -s 8" does not work, found out it was still booting the image as an emulated traditional USB drive image with 512-byte sectors due to a typo in my test.

Looks like it actually needs -S 4096.

Changing to a proper state and reassigning to Ed, hopefully that's right.

> Hold on, no "-S 512 -s 8" does not work

Can you tell me what does work? You mentioned you've changed 512 to 4096 in image-live.bbclass, but you didn't show the change.

Is it something like this?

--- a/meta/classes/image-live.bbclass
+++ b/meta/classes/image-live.bbclass
@@ -217,10 +217,10 @@ build_fat_img() {
        fi
 
        if [ -z "${HDDIMG_ID}" ]; then
-               mkdosfs ${FATSIZE} -n ${BOOTIMG_VOLUME_ID} -S 512 -C ${FATIMG} \
+               mkdosfs ${FATSIZE} -n ${BOOTIMG_VOLUME_ID} -S 4096 -C ${FATIMG} \
                        ${BLOCKS}
        else
-               mkdosfs ${FATSIZE} -n ${BOOTIMG_VOLUME_ID} -S 512 -C ${FATIMG} \
+               mkdosfs ${FATSIZE} -n ${BOOTIMG_VOLUME_ID} -S 4096 -C ${FATIMG} \
                ${BLOCKS} -i ${HDDIMG_ID}
        fi


Here is a similar patch for wic:
--- a/scripts/lib/wic/partition.py
+++ b/scripts/lib/wic/partition.py
@@ -312,7 +312,7 @@ class Partition():
         if self.fstype == 'msdos':
             size_str = "-F 16" # FAT 16
 
-        dosfs_cmd = "mkdosfs %s -S 512 %s -C %s %d" % (label_str, size_str,
+        dosfs_cmd = "mkdosfs %s -S 4096 %s -C %s %d" % (label_str, size_str,
                                                        rootfs, rootfs_size)
         exec_native_cmd(dosfs_cmd, native_sysroot)


Let me know if that works for you and I'll implement --mkfs-opts for wic and something similar for hddimg.

Yes, that is exactly what I changed to get the kernel to mount the filesystem properly.

Sent first patch for review: http://lists.openembedded.org/pipermail/openembedded-core/2017-July/139172.html

sent second patchset for review: http://lists.openembedded.org/pipermail/openembedded-core/2017-July/139178.html

both patchsets have been merged to poky master.