Workaround for loop device bug

Creating a loop device sometimes fails due to a mismatch between kernel and user
code: (https://lore.kernel.org/lkml/8bed44f2-273c-856e-0018-69f127ea4258@linux.ibm.com/)
Work around this by ignoring losetup errors if the loop device was created, and retrying
failed mounts.

run/archiveloop

@@ -90,7 +90,7 @@ function timestamp () {
 function fix_errors_in_image () {
   local image="$1"
   log "Running fsck on $image..."
-  loopback=$(losetup --show -f -P "$image")
+  loopback=$(losetup_find_show -P "$image")
   # Use -p repair arg. It works with vfat and exfat.
   /sbin/fsck "${loopback}p1" -- -p |& timestamp '| ' >> "$LOG_FILE" || echo ""
   losetup -d "$loopback"

run/make_snapshot.sh

@@ -208,7 +208,7 @@ function snapshot {
   # /dev/loop0p1
 
   # Use -p repair arg. It works with vfat and exfat.
-  LOOP=$(losetup --show -P -f "$newsnapname")
+  LOOP=$(losetup_find_show -P "$newsnapname")
   PARTLOOP=${LOOP}p1
 
   if [ "$1" = "fsck" ]

setup/pi/create-backingfiles.sh

@@ -96,7 +96,7 @@ function add_drive () {
   local partition_offset
   partition_offset=$(first_partition_offset "$filename")
 
-  loopdev=$(losetup -o "$partition_offset" -f --show "$filename")
+  loopdev=$(losetup_find_show -o "$partition_offset" "$filename")
   log_progress "Creating filesystem with label '$label'"
   if [ "$useexfat" = true  ]
   then

setup/pi/envsetup.sh

@@ -202,3 +202,31 @@ else
   export PICONFIG_PATH=/dev/null
 fi
 
+# losetup sometimes fails because of a mismatch between kernel and user land
+# (https://lore.kernel.org/lkml/8bed44f2-273c-856e-0018-69f127ea4258@linux.ibm.com/)
+# but even when it fails like that, testing shows the loop device gets created anyway
+function losetup_find_show {
+  local lastarg="${@:$#}"
+  local loop=$(losetup -n -O NAME -j "$lastarg")
+  if losetup -f --show "$@"
+  then
+    return
+  fi
+  if [ -n "$loop" ]
+  then
+    # losetup failed, and there was already a previous loop device for the
+    # given file.
+    # Rather than trying to determine if a new loop device was created, just return
+    # an error.
+    return 1
+  fi
+  local newloop=$(losetup -n -O NAME -j "$lastarg")
+  if [ -z "$newloop" ]
+  then
+    # losetup truly failed and didn't even create a loop device
+    return 1
+  fi
+  echo "$newloop"
+}
+
+export -f losetup_find_show

setup/pi/setup-teslausb

@@ -279,6 +279,19 @@ function update_backingfiles_fstab_entry () {
   fi
 }
 
+function mount_with_retry () {
+  for i in {1..5}
+  do
+    if mount "$@"
+    then
+      return
+    fi
+    setup_progress "'mount $*' failed, retrying"
+    sleep 1
+  done
+  return 1
+}
+
 function create_usb_drive_backing_files () {
   if [ ! -e "$BACKINGFILES_MOUNTPOINT" ]
   then
@@ -323,7 +336,7 @@ function create_usb_drive_backing_files () {
   umount /mnt/cam > /dev/null || true
   local ret=0
   /root/bin/disable_gadget.sh || ret=$?
-  if (( (ret == 0) || (ret == 2) )) && mount /mnt/cam
+  if (( (ret == 0) || (ret == 2) )) && mount_with_retry /mnt/cam
   then
     mkdir -p /mnt/cam/TeslaCam
     mkdir -p /mnt/cam/TeslaTrackMode
@@ -345,20 +358,20 @@ function create_usb_drive_backing_files () {
 
     umount /mnt/cam
 
-    if mount /mnt/music
+    if [ -e /backingfiles/music_disk.bin ] && mount_with_retry /mnt/music
     then
       touch /mnt/music/.metadata_never_index
       umount /mnt/music
     fi
 
-    if mount /mnt/lightshow
+    if [ -e /backingfiles/lightshow_disk.bin ] && mount_with_retry /mnt/lightshow
     then
       mkdir -p /mnt/lightshow/LightShow
       touch /mnt/lightshow/.metadata_never_index
       umount /mnt/lightshow
     fi
 
-    if mount /mnt/boombox
+    if [ -e /backingfiles/boombox_disk.bin ] && mount_with_retry /mnt/boombox
     then
       mkdir -p /mnt/boombox/Boombox
       touch /mnt/boombox/.metadata_never_index

tests/losetuptest.sh

@@ -0,0 +1,31 @@
+#!/bin/bash
+
+# Repro case for the issue described here: https://lore.kernel.org/lkml/8bed44f2-273c-856e-0018-69f127ea4258@linux.ibm.com/
+# When running this script on an affected kernel/userspace combo,
+# such as Raspberry Pi OS with kernel 5.10.103 and losetup 2.33.1,
+# you should see an occasional message about losetup having failed.
+
+TESTIMG=/backingfiles/losetuptest.bin
+while true
+do
+  echo -n "$(date) looping: "
+  dd if=/dev/zero of=$TESTIMG bs=1M count=128 status=none
+  if losetup --sector-size=4096 -fP $TESTIMG
+  then
+    LOOP=$(losetup -n -O NAME -j /backingfiles/losetuptest.bin)
+    echo "$LOOP"
+    losetup -d $LOOP
+  else
+    echo losetup failed:
+    LOOP=$(losetup -n -O NAME -j /backingfiles/losetuptest.bin)
+    if [ -z "$LOOP" ]
+    then
+      echo "No loop device exists"
+    else
+      echo "loop device exists: $LOOP"
+      losetup -d $LOOP
+    fi
+    losetup -l
+  fi
+  rm $TESTIMG
+done