Auto Expand Root

[jlpoolen]

I downloaded the Raspberry Pi 4B, 3B/B+ 64-bit Lite Version alpha9 (576Kb) posted on the project page gentoo-on-rpi-64bit. I installed the extracted image, genpi64-arm64-openrc-lite-alpha9.img (4,349 KB), on a SanDisk 64GB SDHC. I've been working with several Raspberry Pi builds and had become lulled into the expectation that when I burned an image onto a SDHC, the operating system would on the first run expand to consume the extra disk space.

Here's what is on my system:

GenPi64 /usr/local/src/gstreamer # ls /dev
ashmem           fd         loop2         net    ram4            snd     tty19  tty33  tty48  tty62      vcs1    vcsa12    vcsu3        video15
autofs           full       loop3         null   ram5            stderr  tty2   tty34  tty49  tty63      vcs10   vcsa2     vcsu4        video16
block            fuse       loop4         port   ram6            stdin   tty20  tty35  tty5   tty7       vcs11   vcsa3     vcsu5        video18
btrfs-control    gpiochip0  loop5         ppp    ram7            stdout  tty21  tty36  tty50  tty8       vcs12   vcsa4     vcsu6        video20
bus              gpiochip1  loop6         ptmx   ram8            tty     tty22  tty37  tty51  tty9       vcs2    vcsa5     vcsu7        video21
cachefiles       gpiomem    loop7         pts    ram9            tty0    tty23  tty38  tty52  ttyAMA0    vcs3    vcsa6     vcsu8        video22
cec0             hidraw0    loop-control  ram0   random          tty1    tty24  tty39  tty53  ttyprintk  vcs4    vcsa7     vcsu9        video23
cec1             hidraw1    mapper        ram1   raw             tty10   tty25  tty4   tty54  ttyS0      vcs5    vcsa8     vga_arbiter  watchdog
char             hidraw2    media0        ram10  rfkill          tty11   tty26  tty40  tty55  uhid       vcs6    vcsa9     vhci         watchdog0
console          hwrng      media1        ram11  rpivid-h264mem  tty12   tty27  tty41  tty56  uinput     vcs7    vcsm-cma  vhost-net    zero
cpu_dma_latency  input      media2        ram12  rpivid-hevcmem  tty13   tty28  tty42  tty57  urandom    vcs8    vcsu      vhost-vsock
cuse             kmsg       mem           ram13  rpivid-intcmem  tty14   tty29  tty43  tty58  v4l        vcs9    vcsu1     video10
disk             kvm        mmcblk0       ram14  rpivid-vp9mem   tty15   tty3   tty44  tty59  vchiq      vcsa    vcsu10    video11
dma_heap         log        mmcblk0p1     ram15  serial0         tty16   tty30  tty45  tty6   vcio       vcsa1   vcsu11    video12
dri              loop0      mmcblk0p2     ram2   serial1         tty17   tty31  tty46  tty60  vc-mem     vcsa10  vcsu12    video13
fb0              loop1      mqueue        ram3   shm             tty18   tty32  tty47  tty61  vcs        vcsa11  vcsu2     video14
GenPi64 /

In my case, fdisk tells me partition 2 is consuming 59.2 GB:

GenPi64 /usr/local/src/gstreamer # fdisk -l /dev/mmcblk0
Disk /dev/mmcblk0: 59.48 GiB, 63864569856 bytes, 124735488 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: dos
Disk identifier: 0xdd08d348

Device         Boot  Start       End   Sectors  Size Id Type
/dev/mmcblk0p1 *      2048    524287    522240  255M  c W95 FAT32 (LBA)
/dev/mmcblk0p2      524288 124735487 124211200 59.2G 83 Linux
GenPi64 /usr/local/src/gstreamer #

Yet, df tells a different story:

GenPi64 /usr/local/src/gstreamer # df -lh
Filesystem      Size  Used Avail Use% Mounted on
/dev/mmcblk0p2  3.9G  3.5G  528K 100% /
devtmpfs         10M     0   10M   0% /dev
tmpfs           1.6G  760K  1.6G   1% /run
cgroup_root      10M     0   10M   0% /sys/fs/cgroup
shm             3.9G     0  3.9G   0% /dev/shm
/dev/mmcblk0p1  255M   70M  186M  28% /boot
tmpfs           782M     0  782M   0% /run/user/0
tmpfs           782M     0  782M   0% /run/user/1001
GenPi64 /usr/local/src/gstreamer #

In discussion with @jonesmz on Discord, he indicated /boot/autoexpand_root.sh "needs to be moved into the initramfs"

In my glee about having Gentoo working on the Rapsberry Pi Zero 2 W, I raced towards getting gstreamer compiled only to discover I had a 4 GB memory limitation and my build stopped at task 2587 of the 6226 needed:

 ../subprojects/gst-plugins-bad/gst/dvdspu/gstspu-vobsub.c:524:1: fatal error: error writing to /tmp/ccvKHYmt.s: No space left on device
  524 | }
      | ^
compilation terminated.
[2587/6226] 

[jlpoolen]

I'm definitely missing something here, probably has to do with file systems: ext4 vs. btrfs. I usually use ext4 for all my Linux systems. I consulted https://en.wikipedia.org/wiki/Btrfs and https://linoxide.com/btrfs-vs-ext4/. I recall reading somewhere that btrfs is superior for high capacity storage disks which serves as the disk space for Raspberry Pi and like systems. With that in mind, I tried using the steps in the script autoexpand_root.sh. For posterity, here's what autoexpand_root.sh currently has:

GenPi64 /usr/local/src/gstreamer # cat -n /boot/autoexpand_root.sh
     1  #!/bin/sh
     2  #
     3  # Automatically attempt to extend root partition (calculated from
     4  # the filesystem mounted at '/') to fill remaining free space on the device
     5  #
     6  # Checks for a sentinel file "dont_autoexpand_root" on /boot
     7  # and will proceed only if not found.
     8
     9  SENTINEL="/boot/dont_autoexpand_root"
    10  if [ -f "${SENTINEL}" ]; then
    11          # sentinel found, service should really be disabled
    12          echo "Sentinel file detected; consider deleting ${SVCNAME} from boot runlevel"
    13          return 0
    14  fi
    15
    16  # Find the partition mounted as / according to the kernel
    17  ROOTPART="$(awk '/\s\/\s/{print $1}' /proc/mounts)"
    18  # Find the root partitions filesystem type
    19  ROOTFS="$(awk '/\s\/\s/{print $3}' /proc/mounts)"
    20  # Discard the partition number to find the underlying drive
    21  ROOTDRIVE="$(sed 's/[0-9]*$//;s/\([0-9]\)p$/\1/' <<<"${ROOTPART}")"
    22  # Get the partition number too
    23  PNUM="$(egrep -o "[0-9]+$" <<<"${ROOTPART}")"
    24
    25  # Ensure we get some kind of data from the above
    26  if [[ -z "${ROOTPART}" || -z "${ROOTFS}" || -z "${ROOTDRIVE}" || -z "${PNUM}" ]]; then
    27          echo "Unable to determine root path components"
    28          return 1
    29  fi
    30
    31  # first boot, need to resize root partition to fill disk
    32  # then, since this is /, need to reboot to get kernel to see it
    33  # after which its ext4 filesystem can be resized online
    34
    35  echo ""
    36  echo "*******************************************************************"
    37  echo "* Auto resizing root partition to fill block device - please wait *"
    38  echo "*******************************************************************"
    39  echo ""
    40
    41  touch "${SENTINEL}"
    42  if ! [ -f "${SENTINEL}" ]; then
    43          # might mean an ro filesystem, and we don't want an
    44          # infinite loop
    45          echo "Failed to delete sentinel file '${SENTINEL}'"
    46          return 1
    47  fi
    48
    49  # turn off swapfiles, for safety
    50  swapoff -a
    51
    52  # sync filesystems before we begin, to minimize any damage if things go wrong
    53  if ! sync &>/dev/null; then
    54          echo "Failed to sync to disk"
    55          return 1
    56  fi
    57
    58  # begin by resizing the root partition
    59  if ! sfdisk --no-reread --no-tell-kernel -N ${PNUM} <<<", +" "${ROOTDRIVE}" &>/dev/null; then
    60          echo "Failed to resize root partition"
    61          return 1
    62  fi
    63
    64  # Tell the kernel about the change. Seperate for sake of logging
    65  if ! partprobe "${ROOTPART}" &>/dev/null; then
    66          echo "Failed to read root partition change"
    67          return 1
    68  fi
    69
    70  # Do an online resize of the root partition file system
    71  if [ "${ROOTFS}" = "btrfs" ]; then
    72          if ! btrfs filesystem resize max "/" &>/dev/null; then
    73                  echo "Failed to resize root filesystem"
    74                  return 1
    75          fi
    76  else
    77          if ! resize2fs -f "${ROOTPART}" &>/dev/null; then
    78                  echo "Failed to resize root filesystem"
    79                  return 1
    80          fi
    81  fi
    82
    83  # Flush everything to disk
    84  if ! sync &>/dev/null; then
    85          echo "Failed to sync to disk"
    86          return 1
    87  fi
    88
    89  # Swap is OK to use again
    90  swapon -a
    91
    92  echo ""
    93  echo "*******************************************************************"
    94  echo "*  Resize completed                                               *"
    95  echo "*******************************************************************"
    96  echo ""
GenPi64 /usr/local/src/gstreamer #

I then identified the values the script is extracting so I could recreate the command in a console:

GenPi64 /usr/local/src/gstreamer # cat -n /proc/mounts
     1  /dev/mmcblk0p2 / btrfs rw,noatime,compress=zstd:3,ssd,discard,space_cache=v2,subvolid=5,subvol=/ 0 0
     2  devtmpfs /dev devtmpfs rw,nosuid,relatime,size=10240k,nr_inodes=958456,mode=755 0 0
     3  proc /proc proc rw,nosuid,nodev,noexec,relatime 0 0
     ...
     
GenPi64 /tmp # export ROOTPART="$(awk '/\s\/\s/{print $1}' /proc/mounts)"
GenPi64 /tmp # echo $ROOTPART
/dev/mmcblk0p2
GenPi64 /tmp # export ROOTFS="$(awk '/\s\/\s/{print $3}' /proc/mounts)"
GenPi64 /tmp # echo $ROOTFS
btrfs
GenPi64 /tmp # export ROOTDRIVE="$(sed 's/[0-9]*$//;s/\([0-9]\)p$/\1/' <<<"${ROOTPART}")"
GenPi64 /tmp # echo $ROOTDRIVE
/dev/mmcblk0
GenPi64 /tmp # export PNUM="$(egrep -o "[0-9]+$" <<<"${ROOTPART}")"
GenPi64 /tmp # echo $PNUM
2
GenPi64 /tmp #

I inserted SDHC into Hermes (reg. Gentoo system), exposed as sdc, sdc2 is the 59GB partition

hermes /etc/xen # sfdisk -l /dev/sdc2
Disk /dev/sdc2: 59.23 GiB, 63596134400 bytes, 124211200 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
hermes /etc/xen #

Key line No. 59 from script above:

if ! sfdisk --no-reread --no-tell-kernel -N ${PNUM} <<<", +" "${ROOTDRIVE}" &>/dev/null; then

consulted: http://karelzak.blogspot.com/2015/05/resize-by-sfdisk.html where it is stated "it's possible to enalarge as much as possible:" giving this example:

echo ", +" | ./sfdisk -N 1 /dev/sdc

karelzak's example matches the above line No. 59, so we can

Try:

echo ", +" | sfdisk -N 2 /dev/sdc

Result:

hermes /etc/xen # echo ", +" | sfdisk -N 2 /dev/sdc
Checking that no-one is using this disk right now ... OK

Disk /dev/sdc: 59.48 GiB, 63864569856 bytes, 124735488 sectors
Disk model: Flash Reader
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: dos
Disk identifier: 0xdd08d348

Old situation:

Device     Boot  Start       End   Sectors  Size Id Type
/dev/sdc1  *      2048    524287    522240  255M  c W95 FAT32 (LBA)
/dev/sdc2       524288 124735487 124211200 59.2G 83 Linux

/dev/sdc2:
New situation:
Disklabel type: dos
Disk identifier: 0xdd08d348

Device     Boot  Start       End   Sectors  Size Id Type
/dev/sdc1  *      2048    524287    522240  255M  c W95 FAT32 (LBA)
/dev/sdc2       524288 124735487 124211200 59.2G 83 Linux

The partition table has been altered.
Calling ioctl() to re-read partition table.
Syncing disks.
hermes /etc/xen #

Tested modified disk on Rpi4; result: expansion failed

jlpoole@GenPi64 ~ $ date; df -lh
Sat Mar 12 05:03:56 PM UTC 2022
Filesystem      Size  Used Avail Use% Mounted on
/dev/mmcblk0p2  3.9G  3.5G  480K 100% /
devtmpfs         10M     0   10M   0% /dev
tmpfs           1.6G  756K  1.6G   1% /run
cgroup_root      10M     0   10M   0% /sys/fs/cgroup
shm             3.9G     0  3.9G   0% /dev/shm
/dev/mmcblk0p1  255M   70M  186M  28% /boot
tmpfs           782M     0  782M   0% /run/user/1001
jlpoole@GenPi64 ~ $

So I inserted the SDHC into Hermes and performed a check:

hermes /etc/xen #  e2fsck /dev/sdc2
e2fsck 1.46.4 (18-Aug-2021)
ext2fs_open2: Bad magic number in super-block
e2fsck: Superblock invalid, trying backup blocks...
e2fsck: Bad magic number in super-block while trying to open /dev/sdc2

The superblock could not be read or does not describe a valid ext2/ext3/ext4
filesystem.  If the device is valid and it really contains an ext2/ext3/ext4
filesystem (and not swap or ufs or something else), then the superblock
is corrupt, and you might try running e2fsck with an alternate superblock:
    e2fsck -b 8193 <device>
 or
    e2fsck -b 32768 <device>

/dev/sdc2 contains a btrfs file system
hermes /etc/xen #

I then compared the results above ("corrupr"?) with another SDHC from an image I obtained from Q Engineering with Bullseye 64 just to see what e2fsck tells me:

hermes /etc/xen #  e2fsck /dev/sdc2
e2fsck 1.46.4 (18-Aug-2021)
rootfs: clean, 246881/941616 files, 2332422/3822976 blocks
hermes /etc/xen # fdisk -l /dev/sdc2
Disk /dev/sdc2: 14.58 GiB, 15658909696 bytes, 30583808 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
hermes /etc/xen #

And the whole Q Engineering disk shows "83 Linux":

hermes /etc/xen # fdisk -l /dev/sdc
Disk /dev/sdc: 14.84 GiB, 15931539456 bytes, 31116288 sectors
Disk model: Flash Reader
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: dos
Disk identifier: 0x88f8ebc8

Device     Boot  Start      End  Sectors  Size Id Type
/dev/sdc1         8192   532479   524288  256M  c W95 FAT32 (LBA)
/dev/sdc2       532480 31116287 30583808 14.6G 83 Linux
hermes /etc/xen #

Here are some further comparisons:

Q Engineering disk:

hermes /etc/xen # blkid  /dev/sdc
/dev/sdc: PTUUID="88f8ebc8" PTTYPE="dos"
hermes /etc/xen # blkid  /dev/sdc1
/dev/sdc1: LABEL_FATBOOT="boot" LABEL="boot" UUID="E5B6-FD41" BLOCK_SIZE="512" TYPE="vfat" PARTUUID="88f8ebc8-01"
hermes /etc/xen # blkid  /dev/sdc2
/dev/sdc2: LABEL="rootfs" UUID="a4b3ef34-532d-430b-a401-589e4a6a5d2c" BLOCK_SIZE="4096" TYPE="ext4" PARTUUID="88f8ebc8-02"
hermes /etc/xen #


hermes /etc/xen # file -sL /dev/sdc
/dev/sdc: DOS/MBR boot sector; partition 1 : ID=0xc, start-CHS (0x40,0,1), end-CHS (0x3ff,3,32), startsector 8192, 524288 sectors; partition 2 : ID=0x83, start-CHS (0x3ff,3,32), end-CHS (0x3ff,63,32), startsector 532480, 30583808 sectors
hermes /etc/xen # file -sL /dev/sdc1
/dev/sdc1: DOS/MBR boot sector, code offset 0x58+2, OEM-ID "mkfs.fat", Media descriptor 0xf8, sectors/track 32, heads 64, sectors 524288 (volumes > 32 MB), FAT (32 bit), sectors/FAT 4033, serial number 0xe5b6fd41, label: "boot       "
hermes /etc/xen # file -sL /dev/sdc2
/dev/sdc2: Linux rev 1.0 ext4 filesystem data, UUID=a4b3ef34-532d-430b-a401-589e4a6a5d2c, volume name "rootfs" (extents) (large files)
hermes /etc/xen # date; echo above is Q Engineering Disk
Sat Mar 12 10:14:49 PST 2022
above is Q Engineering Disk
hermes /etc/xen #

GenPi64 disk:

hermes /etc/xen # date; echo Below is GenPi64 disk
Sat Mar 12 10:16:02 PST 2022
Below is GenPi64 disk
hermes /etc/xen # file -sL /dev/sdc
/dev/sdc: DOS/MBR boot sector; partition 1 : ID=0xc, active, start-CHS (0x10,0,1), end-CHS (0x3ff,3,32), startsector 2048, 522240 sectors; partition 2 : ID=0x83, start-CHS (0x0,0,1), end-CHS (0x29f,3,32), startsector 524288, 124211200 sectors
hermes /etc/xen # file -sL /dev/sdc1
/dev/sdc1: DOS/MBR boot sector, code offset 0x3c+2, OEM-ID "mkfs.fat", sectors/cluster 8, reserved sectors 8, root entries 512, Media descriptor 0xf8, sectors/FAT 256, sectors/track 63, heads 255, hidden sectors 2048, sectors 522207 (volumes > 32 MB), serial number 0x3d025fe8, unlabeled, FAT (16 bit)
hermes /etc/xen # file -sL /dev/sdc2
/dev/sdc2: BTRFS Filesystem sectorsize 4096, nodesize 16384, leafsize 16384, UUID=b58ab1e7-3c13-41c1-8b74-0a25ff372bfc, 3365949440/4152360960 bytes used, 1 devices
hermes /etc/xen #

hermes /etc/xen # blkid /dev/sdc
/dev/sdc: PTUUID="dd08d348" PTTYPE="dos"
hermes /etc/xen # blkid /dev/sdc1
/dev/sdc1: SEC_TYPE="msdos" UUID="3D02-5FE8" BLOCK_SIZE="512" TYPE="vfat" PARTUUID="dd08d348-01"
hermes /etc/xen # blkid /dev/sdc2
/dev/sdc2: UUID="b58ab1e7-3c13-41c1-8b74-0a25ff372bfc" UUID_SUB="fad65da3-92d5-46fe-a4d1-2b798171f494" BLOCK_SIZE="4096" TYPE="btrfs" PARTUUID="dd08d348-02"
hermes /etc/xen #

[jlpoolen]

For the GenPi64 disk:

hermes /etc/xen # lsblk -o NAME,FSTYPE,FSSIZE,FSUSE%,PTTYPE,PARTLABEL,PARTFLAGS,SIZE /dev/sdc
NAME   FSTYPE FSSIZE FSUSE% PTTYPE PARTLABEL PARTFLAGS  SIZE
sdc                         dos                        59.5G
+-sdc1 vfat                 dos              0x80       255M
+-sdc2 btrfs                dos                        59.2G
hermes /etc/xen #

[jlpoolen]

Definitely a difference (aside from size): ext4 vs. btrfs

hermes /etc/xen # date; echo Q Engineer;  lsblk -o NAME,FSTYPE,FSSIZE,FSUSE%,PTTYPE,PARTLABEL,PARTFLAGS,SIZE /dev/sdc
Sat Mar 12 10:30:08 PST 2022
Q Engineer
NAME   FSTYPE FSSIZE FSUSE% PTTYPE PARTLABEL PARTFLAGS  SIZE
sdc                         dos                        14.8G
+-sdc1 vfat                 dos                         256M
+-sdc2 ext4                 dos                        14.6G
hermes /etc/xen #

So, is the file type of btrfs affecting the tool, sfdisk's, ability to alter the partition? I'm thinking I'll have to go deeper and examine cylinders.

[jlpoolen]

On a RaspberryPi4, these two report are contradicting each other: df says there is a 3.9 GB limit, parted says we have to the end of the disk, e.g. 59GB+.

GenPi64 /home/jlpoole # parted /dev/mmcblk0 'unit chs print'
Model: SD SD64G (sd/mmc)
Disk /dev/mmcblk0: 7764,108,23
Sector size (logical/physical): 512B/512B
BIOS cylinder,head,sector geometry: 7764,255,63.  Each cylinder is 8225kB.
Partition Table: msdos
Disk Flags:

Number  Start     End          Type     File system  Flags
 1      0,32,32   32,162,1     primary  fat16        boot, lba
 2      32,162,2  7764,108,23  primary  btrfs

GenPi64 /home/jlpoole # df -lh
Filesystem      Size  Used Avail Use% Mounted on
/dev/mmcblk0p2  3.9G  3.5G  428K 100% /
devtmpfs         10M     0   10M   0% /dev
tmpfs           1.6G  752K  1.6G   1% /run
cgroup_root      10M     0   10M   0% /sys/fs/cgroup
shm             3.9G     0  3.9G   0% /dev/shm
/dev/mmcblk0p1  255M   70M  186M  28% /boot
tmpfs           782M     0  782M   0% /run/user/1001
GenPi64 /home/jlpoole #

[jlpoolen]

Logged an inquiry on the Gentoo Forum: https://forums.gentoo.org/viewtopic-p-8694549.html#8694549

[jlpoolen]

In Discord, @jonesmz pointed out that line 72 of the script does the btrfs change. I have overlooked that and focused on the sfdisk point. The execution of the command line below fixed the problem.

GenPi64 /home/jlpoole # date; time btrfs filesystem resize max "/";date
Sat Mar 12 07:51:36 PM UTC 2022
Resize device id 1 (/dev/mmcblk0p2) from 3.87GiB to max

real    0m0.101s
user    0m0.003s
sys     0m0.013s
Sat Mar 12 07:51:36 PM UTC 2022
GenPi64 /home/jlpoole # df -lh
Filesystem      Size  Used Avail Use% Mounted on
/dev/mmcblk0p2   60G  3.5G   56G   6% /
devtmpfs         10M     0   10M   0% /dev
tmpfs           1.6G  752K  1.6G   1% /run
cgroup_root      10M     0   10M   0% /sys/fs/cgroup
shm             3.9G     0  3.9G   0% /dev/shm
/dev/mmcblk0p1  255M   70M  186M  28% /boot
tmpfs           782M     0  782M   0% /run/user/1001
GenPi64 /home/jlpoole #

[samip5]

systemd or openrc image? Oh, saw that it's the openrc one. 🤔

[samip5]

Is this still a problem?

[jonesmz]

sadly it is.
Now that we should have the osuosl image working, i'm hoping to revisit switching to dracut, which supports auto-resizing prior to root being mounted. That should circumvent the problem of the kernel refusing to update the partition table.

[samip5]

i'm hoping to revisit switching to dracut, which supports auto-resizing prior to root being mounted. That should circumvent the problem of the kernel refusing to update the partition table.

Should that work even with btrfs?

[jonesmz]

can you elaborate on the question? i'm not sure what you're asking.

[samip5]

can you elaborate on the question? i'm not sure what you're asking.

Should the resizing work with an btrfs filesystem?

[jonesmz]

It's not an issue of the filesystem, it's an issue of the partition that the root filesystem belongs to not being modifable at runtime.

if you use parted / fdisk / gdisk whatever to change the partition table, even if you manually run partprobe, the kernel claims that the partition is the same size. You can read the changed size with the partition editor tool you used no problem, but the kernel itself refuses to acknowledge the change.

when you reboot, suddenly it sees the new size, and the resize can proceed accordingly.

btrfs can do a live resize, but only if the partition has room to grow.

Doing the resize inside the initramfs completely avoids this problem because the initramfs can change the partition table before it pivots root to the real rootfs.