Add a umount wrapper to suppress gvfsd/trashd breaking umount calls.

Specifically, detect if the umount failed due to current access, if so,
give it up to 9 more runs (w/ 1s pauses) continuing only if it's still
failing due to currently open files.

Via this, it should suppress the race of gvfs/trashd looking at
quick mounted/umounted pathways.

This CL is a two parter; this adds the script, and converts common.sh
consumers over to using the override.

The next CL will modify the chroot itself to ensure our script gets
picked up/used.

BUG=chromium-os:23443
TEST=trybot, manul validation.

Change-Id: I92dedd91d6133c2063b1e5dbbc1a68366844801d
Reviewed-on: https://gerrit.chromium.org/gerrit/32087
Commit-Ready: Brian Harring <[email protected]>
Reviewed-by: Brian Harring <[email protected]>
Tested-by: Brian Harring <[email protected]>

Author: Brian Harring

clean_loopback_devices

@@ -39,6 +39,6 @@ if [ "${SURE}" != "y" ]; then
   exit 1
 fi
 
-sudo umount "$OUTPUT_DIR"/*/* 2> /dev/null
-sudo umount /tmp/esp.* 2> /dev/null
+safe_umount "$OUTPUT_DIR"/*/* 2> /dev/null
+safe_umount /tmp/esp.* 2> /dev/null
 sudo losetup -d /dev/loop* 2> /dev/null

common.sh

@@ -562,20 +562,30 @@ safe_umount_tree() {
   fi
 
   # First try to unmount in one shot to speed things up.
-  if sudo umount -d ${mounts}; then
+  if safe_umount -d ${mounts}; then
     return 0
   fi
 
   # Well that didn't work, so lazy unmount remaining ones.
   mounts=$(sub_mounts "$1")
   warn "Failed to unmount ${mounts}"
   warn "Doing a lazy unmount"
-  if ! sudo umount -d -l ${mounts}; then
+  if ! safe_umount -d -l ${mounts}; then
     mounts=$(sub_mounts "$1")
     die "Failed to lazily unmount ${mounts}"
   fi
 }
 
+
+# Helper; all scripts should use this since it ensures our
+# override of umount is used (inside the chroot, it's enforced
+# via configuration; outside is the concern).
+# Args are passed directly to umount; no sudo args are allowed.
+safe_umount() {
+  sudo "${SCRIPT_ROOT}/path-overrides/umount" "$@"
+}
+
+
 get_git_id() {
   git var GIT_COMMITTER_IDENT | sed -e 's/^.*<\(\S\+\)>.*$/\1/'
 }

image_to_usb.sh

@@ -348,7 +348,7 @@ if [ -b "${FLAGS_to}" ]; then
   if [ -n "${mount_list}" ]; then
     echo "Attempting to unmount any mounts on the target device..."
     for i in ${mount_list}; do
-      if sudo umount "$i" 2>&1 >/dev/null | grep "not found"; then
+      if safe_umount "$i" 2>&1 >/dev/null | grep "not found"; then
         die_notrace "$i needs to be unmounted outside the chroot"
       fi
     done

image_to_vm.sh

@@ -146,8 +146,8 @@ dd if="${SRC_IMAGE}" of="${TEMP_PMBR}" bs=512 count=1
 TEMP_MNT=$(mktemp -d)
 TEMP_ESP_MNT=$(mktemp -d)
 cleanup() {
-  sudo umount "${TEMP_MNT}"
-  sudo umount "${TEMP_ESP_MNT}"
+  safe_umount "${TEMP_MNT}"
+  safe_umount "${TEMP_ESP_MNT}"
   rmdir "${TEMP_MNT}" "${TEMP_ESP_MNT}"
 }
 trap cleanup INT TERM EXIT

make_netboot.sh

@@ -68,8 +68,8 @@ fi
 
 # Prepare to mount rootfs.
 umount_loop() {
-  sudo umount r || true
-  sudo umount s || true
+  safe_umount r || true
+  safe_umount s || true
   false
 }
 
@@ -141,7 +141,7 @@ sudo cp "s/dev_image/etc/lsb-factory" "r/${LSB_FACTORY_DIR}"
 
 # Clean up mounts.
 trap - EXIT
-sudo umount r s
+safe_umount r s
 sudo rm -rf r s
 
 # Generate an initrd fo u-boot to load.

mod_image_for_recovery.sh

@@ -77,7 +77,7 @@ get_install_vblock() {
   sudo cp "$stateful_mnt/vmlinuz_hd.vblock"  "$out"
   sudo chown $USER "$out"
 
-  sudo umount "$stateful_mnt"
+  safe_umount "$stateful_mnt"
   rmdir "$stateful_mnt"
   switch_to_strict_mode
   echo "$out"
@@ -186,7 +186,7 @@ create_recovery_kernel_image() {
   # safe.
   sudo sed  -i -e "s/cros_efi/cros_efi kern_b_hash=$kern_hash/g" \
     "$efi_dir/efi/boot/grub.cfg" || true
-  sudo umount "$efi_dir"
+  safe_umount "$efi_dir"
   sudo losetup -a | sed 's/^/16651 /'
   sudo losetup -d "$efi_dev"
   rmdir "$efi_dir"
@@ -243,7 +243,7 @@ install_recovery_kernel() {
     local esp_mnt=$(mktemp -d)
     sudo mount -o loop,offset=$((esp_offset * 512)) "$RECOVERY_IMAGE" "$esp_mnt"
     sudo cp "$vmlinuz" "$esp_mnt/syslinux/vmlinuz.A" || failed=1
-    sudo umount "$esp_mnt"
+    safe_umount "$esp_mnt"
     rmdir "$esp_mnt"
     switch_to_strict_mode
   fi
@@ -322,7 +322,7 @@ maybe_resize_stateful() {
   set +e
   sudo mount -o loop $small_stateful $new_stateful_mnt
   sudo cp "$INSTALL_VBLOCK" "$new_stateful_mnt/vmlinuz_hd.vblock"
-  sudo umount "$new_stateful_mnt"
+  safe_umount "$new_stateful_mnt"
   rmdir "$new_stateful_mnt"
   switch_to_strict_mode
 

mount_gpt_image.sh

@@ -85,13 +85,13 @@ unmount_image() {
       "${FLAGS_stateful_mountpt}"
     fix_broken_symlinks "${FLAGS_rootfs_mountpt}"
   fi
-  sudo umount "${FLAGS_rootfs_mountpt}/usr/local"
-  sudo umount "${FLAGS_rootfs_mountpt}/var"
+  safe_umount "${FLAGS_rootfs_mountpt}/usr/local"
+  safe_umount "${FLAGS_rootfs_mountpt}/var"
   if [[ -n "${FLAGS_esp_mountpt}" ]]; then
-    sudo umount "${FLAGS_esp_mountpt}"
+    safe_umount "${FLAGS_esp_mountpt}"
   fi
-  sudo umount "${FLAGS_stateful_mountpt}"
-  sudo umount "${FLAGS_rootfs_mountpt}"
+  safe_umount "${FLAGS_stateful_mountpt}"
+  safe_umount "${FLAGS_rootfs_mountpt}"
   switch_to_strict_mode
 }
 

path-overrides/umount

@@ -0,0 +1,29 @@
+#!/bin/bash
+
+# Work around a bug on precise where gvfs trash goes looking in mounts
+# it shouldn't, resulting in the umount failing when it shouldn't.
+# See crosbug.com/23443 for the sordid details.
+
+suppressed_dir=$(dirname "$(readlink -f "$0")")
+cleaned_path="$(echo "$PATH" | sed -e 's+\(^\|:\)/usr/local/sbin\(:\|$\)++g')"
+binary="$(PATH="${cleaned_path}" type -P umount)"
+if [ $? -ne 0 ]; then
+  echo "umount: command not found" >&2
+  exit 127
+fi
+
+for x in {1..10}; do
+  # umount doesn't give use a distinct exit code for device is busy; thus grep
+  # the output.
+  output=$(LC_ALL=C "${binary}" "$@" 2>&1)
+  ret=$?
+  if [ ${ret} -eq 0 ] || [[ "${output}" == *"device is busy"* ]]; then
+    # Nothing to do in these scenarios; either ran fine, or it failed in a non
+    # busy fashion.
+    break
+  fi
+  sleep 1
+done
+
+echo -n "${output}" >&2
+exit ${ret}

update_bootloaders.sh

@@ -159,13 +159,13 @@ fi
 ESP_FS_DIR=$(mktemp -d /tmp/esp.XXXXXX)
 cleanup() {
   set +e
-  if ! sudo umount "${ESP_FS_DIR}"; then
+  if ! safe_umount "${ESP_FS_DIR}"; then
       # There is a race condition possible on some ubuntu setups
       # with mounting and unmounting a device very quickly
       # Doing a quick sleep/retry as a temporary workaround
       warn "Initial unmount failed. Possibly crosbug.com/23443. Retrying"
       sleep 5
-      sudo umount "${ESP_FS_DIR}"
+      safe_umount "${ESP_FS_DIR}"
   fi
   if [[ -n "${ESP_DEV}" && -z "${ESP_DEV//\/dev\/loop*}" ]]; then
     sudo losetup -d  "${ESP_DEV}"
@@ -211,7 +211,7 @@ if [[ "${FLAGS_arch}" = "x86" || "${FLAGS_arch}" = "amd64" ]]; then
   # Install the syslinux loader on the ESP image (part 12) so it is ready when
   # we cut over from rootfs booting (extlinux).
   if [[ ${FLAGS_install_syslinux} -eq ${FLAGS_TRUE} ]]; then
-    sudo umount "${ESP_FS_DIR}"
+    safe_umount "${ESP_FS_DIR}"
     sudo syslinux -d /syslinux "${ESP_DEV}"
     # mount again for cleanup to free resource gracefully
     sudo mount -o ro "${ESP_DEV}" "${ESP_FS_DIR}"