zfsbootmenu: log errors for unimportable pools

Before landing at the emergency_shell in the main import loop, log the
output of a failed zpool import for every pool that's marked UNAVAIL .

In the emergency_shell function, write any log messages at the 'err'
level directly to the active console.

The 'print_kmsg_logs' helper has been added that accepts a
comma-separated list of log levels (as understood by util-linux dmesg).
If the feature flags supplied to dmesg fail, fall back to 'dmesg -r'
which is understood by both Busybox and util-linux dmesg. Manually
convert err,info,... log levels to a numeric code and grep for that at
the start of each line.

Author: zdykstra

zfsbootmenu/bin/zlogtail

@@ -2,6 +2,7 @@
 
 # shellcheck disable=SC1091
 source /lib/zfsbootmenu-lib.sh >/dev/null 2>&1 || exit 1
+source /lib/kmsg-log-lib.sh >/dev/null 2>&1 || exit 1
 
 [ -f "${BASE}/have_errors" ] && rm "${BASE}/have_errors"
 [ -f "${BASE}/have_warnings" ] && rm "${BASE}/have_warnings"
@@ -29,10 +30,4 @@ fi
 
 export FZF_DEFAULT_OPTS="${fuzzy_default_options[*]}"
 
-# Try to use feature flags found on dmesg from util-linux
-if output="$( dmesg --notime -f user -l err,warn 2>/dev/null )" ; then
-  echo "${output}" | ${FUZZYSEL}
-else
-  # fall back to manually parsing dmesg output; will show all ZBM generated logs up to zinfo level
-  dmesg | awk '/ZFSBootMenu:/{ for (i=3; i<=NF; i++){ printf("%s ", $i)}; printf "\n" }' | ${FUZZYSEL}
-fi
+print_kmsg_logs "err,warn" | ${FUZZYSEL}

zfsbootmenu/lib/kmsg-log-lib.sh

@@ -84,3 +84,58 @@ zerror() {
   : > "${BASE}/have_errors"
   echo "<3>ZFSBootMenu: $1" > /dev/kmsg
 }
+
+# arg1: comma-separated log levels to print
+# prints: all logs at that level
+# returns: nothing
+
+print_kmsg_logs() {
+  local levels levels_array grep_args
+
+  levels="${1}"
+  if [ -z "${levels}" ]; then
+    zerror "levels is undefined"
+    return
+  fi
+
+  # Try to use feature flags found in dmesg from util-linux
+  if output="$( dmesg --notime -f user --color=always -l "${levels}" 2>/dev/null )" ; then
+    echo -e "${output}"
+  else
+    # Both util-linux and Busybox dmesg support the -r flag. However, the log level that is
+    # reported by Busybox dmesg is larger than that reported by util-linux dmesg. Busybox dmesg
+    # is too bare-bones to do much of anything, so we just need to grep for both integers at 
+    # a given log level, then refly on matching ZFSBootMenu for info and lower, and ZBM for debug.
+
+    IFS=',' read -r -a levels_array <<<"${levels}"
+    for level in "${levels_array[@]}"; do
+      case "${level}" in
+        err)
+          grep_args+=( "-e" "^<11>" "-e" "^<3>" )
+          ;;
+        warn)
+          grep_args+=( "-e" "^<12>" "-e" "^<4>" )
+          ;;
+        notice)
+          grep_args+=( "-e" "^<13>" "-e" "^<5>" )
+          ;;
+        info)
+          grep_args+=( "-e" "^<14>" "-e" "^<6>" )
+          ;;
+        debug)
+          grep_args+=( "-e" "^<15>" "-e" "^<7>" )
+          ;;
+        *)
+          grep_args+=( "-e" "." )
+          ;;
+      esac
+    done
+
+    dmesg -r | grep "${grep_args[@]}" \
+      | awk '
+          /ZFSBootMenu:/{ for (i=3; i<=NF; i++){ printf("%s ", $i)}; printf "\n" }
+          /ZBM:/{ for (i=3; i<=NF; i++){ printf("%s ", $i)}; printf "\n" }
+        '
+  fi
+}
+

zfsbootmenu/lib/zfsbootmenu-core.sh

@@ -196,6 +196,38 @@ match_hostid() {
   return 1
 }
 
+# args: no arguments
+# prints: nothing
+# returns: nothing
+
+log_unimportable() {
+  local pool id state line error_line
+
+  while read -r line; do
+    case "${line}" in
+      pool*)
+        pool="${line#pool: }"
+        ;;
+      id*)
+        id="${line#id: }"
+        ;;
+      state*)
+        state="${line#state: }"
+        if [ "${state}" == "UNAVAIL" ]; then
+          while read -r error_line; do
+            zerror "${error_line}"
+          done <<<"$( zpool import -N "${id}" 2>&1 )"
+        fi
+        state=
+        pool=
+        id=
+        ;;
+      *)
+        ;;
+    esac
+  done <<<"$( zpool import )"
+}
+
 # args: none
 # prints: nothing
 # returns: 0 if at least one pool is available
@@ -1815,7 +1847,12 @@ emergency_shell() {
 
 	EOF
 
-  command -v efibootmgr >/dev/null 2>&1 && mount_efivarfs "rw"
+  if [ -f "${BASE}/have_errors" ]; then
+    print_kmsg_logs "err"
+    echo ""
+  fi
+
+  command -v efibootmgr >/dev/null 2>&1 && mount_efivarfs "rw" 
 
   # -i (interactive) mode will source $HOME/.bashrc
   /bin/bash -i

zfsbootmenu/libexec/zfsbootmenu-init

@@ -116,6 +116,7 @@ fi
 # If a boot pool is specified, that will be tried first
 try_pool="${boot_pool}"
 zbm_import_attempt=0
+
 while true; do
   if [ -n "${try_pool}" ]; then
     zdebug "attempting to import preferred pool ${try_pool}"
@@ -173,6 +174,7 @@ while true; do
       continue
   fi
 
+  log_unimportable
   # Allow the user to attempt recovery
   emergency_shell "unable to successfully import a pool"
 done